Introducing The Jupiter Release of Galileo: An Inside Story
What new data sources and benchmarks were added to Galileo to enhance its relevance for enterprise decision-making?
The Jupiter release integrated executive compensation data from Equilar, job titles and skills for every major position across all functional areas, and salary benchmarks covering approximately 120 countries , alongside metrics on turnover, business practices, and hundreds of datasets on vendor market technologies and HR solutions. This unified intelligence layer enables organizations to conduct comprehensive salary analysis, skills assessment, and vendor evaluation in a single platform.
Building a comprehensive human capital data foundation
Galileo's Jupiter release represents a deliberate expansion of the platform's data architecture. Rather than adding isolated datasets, the release systematically integrated sources that directly address the most common enterprise questions: compensation competitiveness, talent skill requirements, and vendor technology fit.
The inclusion of Equilar's executive compensation data gives organizations real-time insight into pay practices at peer companies, addressing one of the most frequently audited HR functions. Combined with Wagescape salary benchmarks and complementary sources, this coverage spans functional roles from entry-level individual contributors through senior leadership.
As discussed in the episode , the geographic scope—approximately 120 countries—ensures that global enterprises can make localized decisions rather than relying on a single regional baseline. This is critical for multinational organizations operating in markets with vastly different labor conditions and regulatory requirements.
From salary analysis to vendor intelligence
Beyond compensation, Jupiter extended Galileo's reach into vendor market technologies and HR solutions , incorporating hundreds of datasets on tools, platforms, and services that organizations use to manage their workforce. This transforms Galileo from a research repository into an integrated decision platform.
The addition of turnover benchmarks and business practice data creates a feedback loop: organizations can now see not only what peers pay, but also what technologies those peers use, how their turnover compares, and what practices correlate with better outcomes. Josh Bersin explains in the podcast that this approach ensures data remains actionable and relevant to specific industries, company sizes, and geographies—not just generic best practices.
"We essentially built a new architecture that takes the essence of Galileo and carefully indexed it and tagged it in eight different dimensions so that the LLM can read it extremely well and does not have to hallucinate."
Josh Bersin — Founder and CEO of The Josh Bersin Company. Bersin has spent nearly 30 years studying human capital practices, management, and HR. Over this period, he has identified 95 functional capability areas of human capital and developed maturity models, benchmarks, assessments, and nearly 1,800 case studies to provide actionable solutions tailored to specific industries, locations, and company sizes.
The structural foundation behind these data additions matters as much as the data itself. The episode goes deeper into how Galileo's research content was restructured to be consumed by AI systems—a critical technical decision that allows the platform to scale these new data sources reliably.
Jupiter added executive compensation data from Equilar, enabling peer-to-peer pay analysis for organizational benchmark assessments.
Job title and skills data now map every major position across all functional areas, supporting talent planning and internal equity reviews.
Salary benchmarks from Wagescape and related sources cover approximately 120 countries, enabling localized decision-making for global enterprises.
Hundreds of datasets on vendor market technologies and HR solutions allow organizations to evaluate tools against peer adoption and business outcomes.
Turnover and business practice benchmarks create a comprehensive intelligence layer connecting compensation, skills, technology, and organizational performance.
How does the Jupiter release integrate human capital intelligence with enterprise systems and business problem-solving?
Jupiter embeds Galileo's intelligence directly into enterprise data systems like SAP, Oracle, and Workday, allowing users to ask business questions that automatically surface relevant human capital insights. The system chains business problems to HR solutions using the HR 2030 model, answering questions like "What is the span of control of our sales department?" and comparing compensation to market benchmarks.
Connecting Intelligence to Enterprise Data
Jupiter works by building a semantic layer that understands the data structures and terminology of major enterprise systems. When a user asks a business question—whether about organizational hierarchy, workforce planning, or pay equity—the system translates that inquiry into the underlying data models of Workday, SAP, or Oracle, then applies Galileo's knowledge base to deliver actionable answers.
The integration goes beyond simple data lookup. Jupiter maps business problems directly to functional HR capabilities , using the nearly 1,800 case studies and 95 functional capability areas that Bersin's company has documented over 30 years. This means the answer includes not just what your data shows, but context from similar companies facing the same challenge.
A Chain of Thought from Problem to Solution
The architecture creates what Bersin describes as a chain of thought connecting business problems to human capital issues , embedded in the HR 2030 model. This framework ensures that when someone asks about sales department efficiency, the system can identify whether the issue stems from span of control, leadership capability, compensation structure, or other structural factors—and suggest solutions based on what has worked in comparable organizations.
HiBob, Workday, ServiceNow, and other platforms are now integrating this layer. Users don't need to switch tools or navigate multiple systems —they simply ask their question in the platform they already use, and Jupiter surfaces the connected insights. As Bersin explains in this podcast episode , this shift was driven by customer demand: teams wanted Galileo integrated where they already work, not as a separate tool.
"We essentially built a new architecture that takes the essence of Galileo and carefully indexed it and tagged it in eight different dimensions so that the LLM can read it extremely well and does not have to hallucinate."
Josh Bersin — Founder and CEO, The Josh Bersin Company. For nearly 30 years, Bersin has led research into human capital practices across 95 functional capability areas, compiling nearly 1,800 case studies from organizations worldwide. His firm helps companies understand how to structure, develop, and manage their people in relation to measurable business outcomes.
The real advantage of Jupiter lies in its precision. The episode details how this multi-dimensional tagging prevents hallucination , ensuring that HR answers are grounded in both your company's actual data and validated research benchmarks from comparable organizations.
What performance advantage does the optimized Agent Ready Corpus deliver compared to general-purpose language models?
The Agent Ready Corpus is 10 to 100 times more efficient at token usage for answering complex HR questions than general-purpose models like ChatGPT or Claude. While those models hallucinate and misattribute information when handling nuanced human capital scenarios, the optimized ARC delivers accurate, predictable, and contextually relevant answers grounded in real case studies and benchmarks.
Precision Over Generalization: The Core Difference
General-purpose language models lack the deep, structured knowledge required to answer complex HR and management questions accurately. When posed with a question about performance management practices in a mid-sized tech company in the UK, ChatGPT or Claude will often either invent plausible-sounding answers or conflate unrelated best practices, because they have no reliable way to distinguish between what applies to Google, a small retailer, or a fast-growing startup.
The Agent Ready Corpus solves this by indexing and tagging research content across eight distinct dimensions —industry, company size, geography, functional capability, and others—so the language model can retrieve and reference the exact case studies and benchmarks that apply to the user's specific scenario. This structured approach eliminates hallucination and ensures every answer is backed by real evidence.
Benchmarking Against Real-World Complexity
The efficiency gain became measurable through rigorous testing. Bersin's team benchmarked the ARC against general LLMs using 30 golden prompts—complex, multifaceted HR questions drawn from real advisory work. The results were decisive: token usage dropped dramatically while accuracy, reliability, and contextual relevance all improved. As explained in detail in the podcast episode , the ARC architecture allows the model to "read extremely well and does not have to hallucinate."
"We essentially built a new architecture that takes the essence of Galileo and carefully indexed it and tagged it in eight different dimensions so that the LLM can read it extremely well and does not have to hallucinate."
Josh Bersin — Founder and CEO of The Josh Bersin Company. For nearly 30 years, Bersin has studied human capital practices across management, leadership, and HR, identifying 95 functional capability areas and developing nearly 1,800 case studies. His research spans practices and benchmarks from approximately 120 countries, making him one of the most comprehensive sources of actionable HR intelligence grounded in comparative organizational data.
This specificity matters enormously for practical use. A CFO in Japan asking about pay equity practices for a distribution company of 5,000 employees receives an answer that references similar organizations in similar markets, not a generic summary that conflates disparate contexts. The ARC's efficiency gain isn't just about fewer tokens—it's about eliminating wasted computation on uncertainty and hallucination.
The practical outcome is also speed and cost reduction: the Jupiter Release, as discussed in this episode , makes it viable to run Galileo as a portable intelligence layer across multiple platforms—integrated into ChatGPT, Claude, Microsoft Copilot, or Google Gemini—without the infrastructure burden or latency penalties that would come from relying on those models' raw capability alone.
What structural changes were made to optimize research content for AI consumption, and what is the Agent Ready Corpus?
The Josh Bersin Company realized their traditionally structured research—written like books with introductions, tables of contents, and conclusions—was fundamentally inefficient for AI systems. They built ARC (Agent Ready Corpus), a new architecture that extracts research content into machine-readable chunks and indexes it with eight dimensions of metadata tagging , allowing large language models to read and cite sources precisely without hallucinating or guessing.
Agent Ready Corpus (ARC) is a proprietary architecture developed by The Josh Bersin Company that restructures research data into a format optimized for AI consumption. Rather than delivering content in traditional document format, ARC breaks down nearly 1,800 case studies, maturity models, benchmarks, and frameworks into machine-readable segments, each tagged across eight distinct dimensions to enable large language models to understand context, retrieve accurate citations, and deliver precise answers without reliance on generative hallucination.
From Document Format to Machine-First Structure
The core problem was structural. Bersin's company had spent 28 years accumulating research across 95 functional capability areas in human capital management—everything from performance management to pay equity to employee wellbeing. Yet when large language models tried to extract value from this corpus, the traditional book-like format with narrative introductions and conclusions actually made it harder for AI to isolate and cite factual claims without inventing answers.
The solution was not to write more reports. As Bersin explains in the Jupiter Release episode , the company stripped away boilerplate entirely and rebuilt their entire knowledge base from scratch into what they call the Agent Ready Corpus.
Eight Dimensions of Tagging for Precision
The ARC architecture doesn't simply chunk text randomly. Each piece of research is tagged across eight different dimensions of metadata , creating a multi-layered index that lets LLMs understand not just what a fact is, but what context it applies to—industry, company size, geography, business challenge, and more. This granular tagging is what prevents hallucination.
The result is concrete: when a user queries Galileo (the Josh Bersin Company's AI-powered research platform) about performance management in retail companies in Germany, the system doesn't guess. It finds the exact relevant case studies and benchmarks from their corpus of nearly 1,800 real-world examples, pulled from approximately 120 countries, and returns them with proper citations. The system is reported to be 10 to 100 times more efficient at token usage compared to general-purpose LLMs querying the same research in unstructured format.
"We essentially built a new architecture that takes the essence of Galileo and carefully indexed it and tagged it in eight different dimensions so that the LLM can read it extremely well and does not have to hallucinate."
Josh Bersin — Founder and CEO, The Josh Bersin Company. Bersin has studied human capital practices and organizational management for nearly 30 years, developing maturity models, benchmarks, and nearly 1,800 case studies across 95 functional capability areas. His research serves over 2,000 companies seeking evidence-based solutions tailored to their specific industry, location, and company size.
The deeper insight Bersin explores in this episode reveals how 18 months of close collaboration with Microsoft shaped the optimization of this architecture specifically for Copilot integration—a detail that underscores the industrial maturity of the approach.
This shift represents a fundamental rethinking of how research institutions can serve AI-driven workflows. Rather than asking AI to understand human-authored content, The Josh Bersin Company engineered their content to be natively understandable by AI, without sacrificing the depth or specificity that made their research valuable in the first place.
Why did the strategy shift from building a specialized AI agent to creating a portable intelligence layer across multiple platforms?
Customers already owned AI platforms like ChatGPT, Gemini, Claude, and Microsoft Copilot—they wanted Galileo integrated there rather than as a separate tool. This insight sparked partnerships with ServiceNow, SAP, and Workday, revealing that Galileo's real power came from embedding intelligence directly into existing workflows and combining it with company policies and employee data , not from operating in isolation.
The initial vision was to build Galileo as a standalone AI agent. However, customer feedback fundamentally reshaped the strategy. Organizations did not want to add yet another platform to their stack; they wanted to leverage the intelligence Galileo offered within the tools they already used daily.
This shift revealed a critical insight about where AI value actually lives. As discussed in the Jupiter Release episode , the constraint was not Galileo's capabilities—it was that embedding intelligence into isolated platforms limited its potential . The breakthrough came from understanding that interleaving Galileo's knowledge with enterprise data sources created exponentially more value.
The architectural redesign for this portability was substantial. As Bersin explains in the podcast , the team "essentially built a new architecture that takes the essence of Galileo and carefully indexed it and tagged it in eight different dimensions so that the LLM can read it extremely well and does not have to hallucinate." This technical foundation enabled Galileo to operate effectively across ServiceNow, SAP, Workday, and other enterprise systems without losing accuracy or context.
"We essentially built a new architecture that takes the essence of Galileo and carefully indexed it and tagged it in eight different dimensions so that the LLM can read it extremely well and does not have to hallucinate."
Josh Bersin — Founder and CEO of The Josh Bersin Company. For nearly 30 years, Bersin has studied human capital management practices and built a corpus of nearly 1,800 case studies across 95 functional capability areas in HR and organizational management. His research foundation—combined with AI—now powers Galileo, which serves over 2,000 companies seeking intelligence that bridges employee data, company policies, and strategic talent decisions.
The deals that followed with major enterprise vendors were not cosmetic integrations. Each partnership required Galileo to function as a genuine intelligence layer—combining the depth of human capital research with live organizational data. Listen to the full episode to understand how 18 months of collaboration with Microsoft on Copilot optimization shaped this new architecture and what the Jupiter Release means for the future of embedded AI in enterprise software.
Portability as the Path to Real Enterprise Value
What the strategy shift exposed is that platform-agnostic intelligence scales faster than proprietary agents . When Galileo lived only in its own interface, it competed for attention and required new user behavior. Integrated into ServiceNow, SAP, and Workday—systems where HR teams and managers already spend their day—Galileo became invisible infrastructure, not another tool to learn.
This shift also opened a path to richer data. In a standalone agent, Galileo could only reference published research and case studies. Embedded in enterprise systems with permission to access organizational hierarchies, compensation data, and talent pipelines, it became capable of answering questions like "How does our succession planning compare to similar companies in our geography?" rather than generic advice. The episode details how this combination of curated research intelligence with live organizational context transformed Galileo from a reference tool into a decision support system.
Customers preferred integrating AI intelligence into existing platforms rather than adopting new standalone tools, forcing a fundamental architecture redesign.
The new strategy positioned Galileo as a portable intelligence layer capable of operating across ServiceNow, SAP, Workday, and Microsoft Copilot.
Platform portability required indexing and tagging Galileo's knowledge across eight dimensions to enable accurate, hallucination-free operation across different systems.
Embedding intelligence directly into workflows where users already work proved more valuable than operating as an isolated agent with its own interface.
What opportunity did the emergence of large language models like ChatGPT create for HR research organizations?
When ChatGPT launched, Josh Bersin recognized an immediate opportunity: build a search engine where users could query an entire corpus of human capital research and receive instant answers, then drill down to supporting reports. This led to Galileo, a generative AI tool that not only retrieves information but creates new content—implementation plans, RFP checklists, hiring guides, and behavioral interview guides—using the organization's intellectual property .
The opportunity was clear: ChatGPT had proven that large language models could answer complex questions in natural language. Bersin saw that his company's 28 years of accumulated research—95 functional capability areas, nearly 1,800 case studies, maturity models, and benchmarks—could be transformed into an interactive search experience powered by AI. Rather than selling static reports, the model shifted to dynamic knowledge retrieval and generation.
This shift fundamentally changed how organizations access and apply HR research. As Bersin explains in the episode , Galileo could take a user's specific question—say, how to design a performance management system for a tech startup in the UK—and generate a tailored implementation plan instantly, drawing from case studies and best practices relevant to that exact context. The AI didn't just retrieve; it synthesized and created.
"We essentially built a new architecture that takes the essence of Galileo and carefully indexed it and tagged it in eight different dimensions so that the LLM can read it extremely well and does not have to hallucinate."
Josh Bersin — Founder and CEO of The Josh Bersin Company. For nearly 30 years, Bersin has studied human capital practices across organizations worldwide, identifying 95 distinct functional capability areas in HR and management. His firm has developed maturity models, benchmarks, and nearly 1,800 case studies covering practices from approximately 120 countries, positioning it as a leading source of actionable HR intelligence.
The technical architecture mattered deeply. Bersin's team didn't simply feed raw research into a generic LLM; they built the Agent Ready Corpus (ARC) , a carefully structured knowledge base indexed and tagged across eight dimensions so that large language models could retrieve and reason about information with precision. This prevented hallucination—a critical flaw in generic LLMs when applied to specialized domains—while improving efficiency by 10 to 100 times compared to general-purpose models.
One compelling detail from the episode: the team developed 30 golden prompts used to benchmark AI performance , ensuring that Galileo delivered consistent, high-quality answers across different use cases and LLM providers—including ChatGPT, Claude, and Google Gemini.
The business opportunity was equally significant: more than 2,000 companies now use Galileo , shifting from a research membership model to a subscription-based AI co-pilot that generates real-time answers and artifacts. This transformation from selling finished research to selling instant, personalized synthesis of knowledge represented a fundamental evolution in how HR research monetizes and delivers value.
How has the business model for monetizing human capital research evolved over the past decade?
The Josh Bersin Company shifted from a traditional research membership model —selling access to reports, frameworks, and graphics similar to Gartner's approach—to role-based subscription models that deliver tailored videos, podcasts, and educational content. This evolution reflects a fundamental insight: consumers need research and resources tailored to their specific job function and industry, not a one-size-fits-all library.
From Library Access to Personalized Content Delivery
The original business model, launched alongside the company's founding in 1998, followed the industry standard of research membership: package all research into reports and frameworks, then sell access to the entire collection. This approach worked well for a broad audience, but it had a critical flaw.
The key insight was that relevance matters far more than volume. A C-level executive, a recruiter, an L&D specialist, and a technology vendor each need fundamentally different resources—yet a traditional membership forced them all to sift through the same massive library. As Josh Bersin explains in the episode , telling a small retailer how Walmart or Google manages performance is interesting academically, but it may not be actionable for their specific context.
Over the past decade, this realization drove a complete redesign of the business model. Rather than pushing all research through the same channel, the company now segments by role and delivers curated content formats —videos, podcasts, frameworks, and educational materials—that match how each audience actually works and learns.
Scaling Relevance Through Role-Based Subscriptions
The new subscription architecture reflects decades of research across nearly 1,800 case studies and 95 functional capability areas in human capital management. Instead of a single product, there are now differentiated offerings for executives, HR practitioners, recruiters, L&D leaders, and technology vendors.
Each segment receives content shaped for their job and their industry context. A vendor looking to understand HR benchmarks gets analysis tailored to their product category; an HR leader gets actionable insights from companies similar in size and geography. This model aligns revenue directly with value delivered, and it reflects a shift in how knowledge workers prefer to consume research—through curated, relevant narrative rather than exhaustive archives.
The underlying IP remains the same: the research foundation described in the podcast includes maturity models, benchmarks, assessments, and hundreds of case studies. What changed is the distribution layer—how that IP is packaged, personalized, and delivered to increase adoption and impact.
"We take the research, we produce it into reports and models and frameworks and graphics… we talk to vendor customers. We talk to the vendors themselves. We have pretty much the same examples and same process."
Josh Bersin — Founder and CEO, The Josh Bersin Company. Over nearly 30 years, Bersin has built a research practice studying human capital management across 95 functional capability areas, from talent management and pay equity to diversity and employee experience. His company now combines research, advisory services, education, and consulting to help organizations align people practices with business strategy.
One detail worth exploring further: the episode reveals how Galileo, the company's AI-powered research platform, now enables even more precise content personalization by indexing research across eight different dimensions—a capability the original membership model could never offer.
The traditional research membership model (reports + frameworks for all) proved less effective than role-based subscriptions that deliver tailored content to specific job functions.
Relevance to industry, company size, and location drives adoption far more than comprehensive but generic best-practice libraries.
Modern subscription models package the same research IP into multiple formats—videos, podcasts, educational materials—optimized for how different audiences prefer to learn and work.
The shift reflects a deeper insight: HR professionals, vendors, and executives need curated guidance, not unlimited access to everything.
What is the foundational approach to conducting human capital research across different company sizes and industries?
Effective human capital research must deliver actionable solutions specific to your industry, location, and company size rather than applying generic best practices from large enterprises. What works for Google or Walmart may not be relevant to a fast-growing tech company or small retailer—which is why comparable case studies from organizations like yours become the foundation of truly applicable insights.
The risk of one-size-fits-all research is real. When you study how a global multinational manages performance, you might learn something interesting, but the operational reality of your organization often demands a completely different approach. Location, industry dynamics, and scale create distinct constraints that make generic frameworks insufficient.
As Bersin explains in the episode , the solution is to build research on a foundation of nearly 1,800 real-world case studies spanning different industries, geographies, and company sizes. These cases serve as mirrors—when you find organizations with similar characteristics to your own, their experiences become immediately relevant to your situation.
Over close to 30 years of studying how companies actually manage their people, this research has identified 95 distinct functional capability areas of human capital. For each area, the research includes maturity models, benchmarks where appropriate, assessments, and most importantly, case studies from real implementations. That empirical foundation transforms research from theoretical to practical.
Why generic benchmarks miss the mark
A retail distribution company in the UK faces supply chain realities, hiring cycles, and operational pressures that differ fundamentally from a SaaS startup burning through growth capital. A fast-growing tech company prioritizes scaling rapidly while managing turnover; a legacy manufacturer focuses on stability and retention. The same performance management system, succession planning approach, or organizational design simply will not work equally well across these contexts.
The research model described in The Josh Bersin Company podcast reverses this logic. Instead of asking "what do best-in-class companies do?", it asks "what do companies similar to mine do?" That shift—from aspiration to applicability—is what makes research actionable.
"The purpose of this research is to give you actionable solutions to problems in your industry, in your location, in your company size. If I tell you how Google or Walmart does performance management and you're a fast-growing tech company or a small retailer or a retail distribution company in the UK, it may not apply at all."
Josh Bersin — Founder and CEO, The Josh Bersin Company. Bersin has spent nearly 30 years researching human capital management across thousands of organizations. He has identified 95 functional capability areas of HR and management, developed maturity models and benchmarks, and compiled nearly 1,800 case studies to make research relevant to organizations of all sizes and types across multiple continents.
The depth of this research ecosystem is itself the key differentiator. Bersin's research team has indexed and categorized these insights across industry, geography, and company stage, making it possible to surface examples directly comparable to your own situation. When you're evaluating how to restructure your HR function or implement a new performance system, you can examine what actually happened in organizations shaped like yours—not in unicorns or century-old conglomerates.
The research membership model: making context-specific insights accessible
To monetize and distribute this knowledge systematically, the research operates through a membership model. Rather than selling one-off reports, members get access to an evolving library of frameworks, maturity models, vendor analysis, and case studies—all filtered and searchable by industry, location, and organizational size. This allows practitioners to pull exactly the relevant slice of the research at the moment they need it.
The case study component is particularly vital. The research framework includes continuous interviews with companies and vendors , capturing not just what was implemented, but why it worked—or didn't—in their specific context. That narrative context is what transforms data into wisdom.
How does AI enable HR to shift from transactional to strategic value delivery?
By automating monotonous operational work, Seneca's HR team freed capacity to focus on higher-value activities that excite and develop their capability.…
What foundational elements must precede successful AI implementation in HR transformation?
Before pursuing AI initiatives, organizations must first establish data governance, process redesign, and knowledge management —the three foundational blocks that enable AI to deliver real impact. These fundamentals create the clean data, efficient processes, and accessible knowledge that AI systems depend on to function effectively.
Why the order matters: foundations before innovation
Many organizations rush to implement AI without preparing the ground beneath it. Diana Mouhan, Executive Director of Transformation and Change at Seneca Polytechnic, observed this tension during her institution's HR transformation journey. When budget pressures emerged, the team had a choice: pause the entire transformation indefinitely, or redesign it using the tools and fundamentals already at hand.
They chose the second path. Rather than asking which new technologies to invest in next, they refocused on working with what they had and ensuring the basics were right . This shift proved decisive. By strengthening data governance first—ensuring consistent, clean data across systems—and redesigning processes to eliminate unnecessary steps, they created the conditions for both current operations and future AI adoption to succeed.
As Diana explains in the episode , this discipline meant refreshing over 100 knowledge articles to improve user experience and self-service, which in turn fed into faster response times—eventually dropping from nearly two weeks to less than 24 hours.
The three pillars: data, processes, knowledge
Data governance is the first pillar. Without clear ownership, validation rules, and maintenance protocols, AI systems inherit garbage and amplify it. Seneca made this explicit: who owns each data field? When is it updated? What counts as valid? These questions, while unglamorous, determined everything downstream.
Process redesign is the second. Seneca's team didn't simply automate broken workflows—they examined every HR process and asked whether it was necessary at all. Some steps vanished. Others were consolidated. The result: employees interacting with HR through a single, modern platform rather than scattering requests across email, phone, and fax. More than 90% of interactions consolidated onto ServiceNow.
Knowledge management is the third. Seneca invested in making their internal knowledge base searchable, current, and useful. This doesn't sound like "AI transformation," but it enables self-service, reduces support volume, and gives AI agents better content to reference when they do engage. A point detailed in this podcast .
"Trust isn't just telling people don't be afraid. Trust is actually built on acknowledging the concerns that they have and being super vulnerable in that journey with them."
Diana Mouhan — Executive Director of Transformation and Change at Seneca Polytechnic. A transformation specialist with over a decade of expertise in organizational strategy, AI adoption, and the future of work, Diana has led transformations across Deloitte and Kearney before moving into executive roles, where she now guides institutions through the intersection of technology and human change.
Trust, in Diana's view, isn't manufactured through reassurance alone. It emerges from honest acknowledgment of concerns and transparent partnership in the change process. When she describes building foundations first, she's also describing a change management posture: admit what is uncertain, show up alongside people in the discomfort, and build capability together rather than imposing solutions.
Seneca's team began with only 25% to 30% of the HR staff trusting AI for workspace utilization . Rather than dismiss that skepticism, they used it as feedback. They strengthened the fundamentals, ran pilots transparently, and let results speak. The cultural shift followed naturally once people saw faster service, fewer errors, and their own workload becoming more strategic.
The broader lesson surfaced in the full conversation : when you establish these three pillars before deploying AI, the technology becomes an accelerator of what's already working, not a band-aid for what's broken.
What operational improvements result from digitizing HR service delivery and automating routine work?
Seneca Polytechnic moved more than 90% of employee HR interactions onto the ServiceNow platform , consolidating them from scattered email, phone, fax, and other channels. Average response times plummeted from nearly two weeks to less than 24 hours—a transformational shift in service speed.
This dramatic acceleration wasn't achieved through technology alone. The team refreshed over 100 knowledge articles to enable self-service capabilities, allowing employees to find answers independently without waiting for HR staff responses.
As Diana Mouhan explains in the episode , the shift required rethinking how HR operates fundamentally. Rather than asking what new technology to invest in, Seneca redesigned its service delivery model using existing tools—a pragmatic approach that freed resources and focused effort on user experience and data quality.
The consolidation onto a single platform also enabled stronger data governance and quality controls , allowing the HR team to analyze trends and make decisions based on complete, reliable information rather than fragmented records scattered across multiple systems.
"Trust isn't just telling people don't be afraid. Trust is actually built on acknowledging the concerns that they have and being super vulnerable in that journey with them."
Diana Mouhan — Executive Director of Transformation and Change at Seneca Polytechnic. With over a decade of organizational strategy and AI adoption expertise, Mouhan brings experience from leadership roles at Deloitte and Kearney, where she specialized in guiding organizations through complex technology transformations and change management initiatives.
The implementation succeeded not by forcing adoption, but by understanding where each team member stood in their technology journey. The transparency built trust, turning skepticism into buy-in—critical when asking people to abandon familiar channels and embrace a new way of working. Discover how Seneca's team navigated HR team readiness and adoption barriers in the full conversation .
Consolidating HR interactions onto a single platform (ServiceNow) reduced response times from nearly two weeks to under 24 hours.
Refreshing over 100 knowledge articles enhanced self-service capabilities, empowering employees to resolve issues independently.
Improved data governance enabled better trend analysis and decision-making across HR operations.
Change management and stakeholder trust were as critical as the technology itself to successful adoption.
Why should AI agents be formally integrated into organizational charts?
Adding AI agents to your organizational chart creates formal ownership and accountability for each agent's performance. This governance layer lets you define clear workflows, assign defined responsibilities, establish performance metrics, and continuously improve—treating AI as managed team roles, not simple tools.
When Diana Mouhan embedded AI agents directly into Seneca Polytechnic's organizational structure, she moved beyond thinking of them as chatbots or software features. Each agent received its own place in the hierarchy, a designated owner, and explicit responsibility for outcomes. This structural change forced clarity: who decides what the agent does, who monitors its performance, and who fixes it when something goes wrong?
The governance benefit runs deeper than accountability. As discussed in this episode of The Josh Bersin Company , formal integration enables you to design and refine workflows specific to each agent's role. Rather than letting an AI tool drift into whatever function seems useful at the moment, you intentionally architect what it should do, how it should do it, and what success looks like.
Performance management becomes possible once agents have defined roles. You can measure response quality, track efficiency gains, and adjust the agent's parameters based on real data —just as you would for a human team member. This continuous improvement cycle treats AI not as a static installation but as an evolving, managed resource that improves over time.
At Seneca, this approach transformed response times across their HR operations. When agents are properly integrated and managed, they don't languish in limbo; they deliver measurable results because someone owns them and cares about their performance. The organizational chart becomes a living document that reflects how your organization actually works—human and AI resources working together under clear accountability structures.
"Trust isn't just telling people don't be afraid. Trust is actually built on acknowledging the concerns that they have and being super vulnerable in that journey with them."
Diana Mouhan — Executive Director of Transformation and Change at Seneca Polytechnic. Diana is a transformation specialist with over a decade of experience in organizational strategy, AI adoption, and the future of work. She previously held senior roles at Deloitte and Kearney before leading Seneca's enterprise transformation initiatives, where she architected the integration of AI agents into HR operations at scale.
One insight that extends beyond this fiche: Diana's team didn't just drop AI into workflows overnight—they first assessed each team member's readiness and understanding. Some staff didn't yet know what a prompt was, while others were already exploring advanced use cases. This granular readiness assessment shaped how and when agents were deployed. Discover more about how Seneca measured team readiness in the full podcast .
Org chart integration clarifies ownership—each agent has a designated person responsible for its performance and continuous improvement.
Formal roles enable defined workflows and measurable success criteria for every AI agent in your HR operations.
Governance structures prevent AI agents from drifting into ad-hoc uses; they stay aligned with strategic HR objectives.
Performance management and tracking become possible when agents are treated as managed team members, not auxiliary tools.
What practical AI use cases can HR teams implement across workforce operations?
Seneca Polytechnic deployed AI across five critical workforce areas: workforce planning, organizational design, job architecture, change management, and compensation analysis . Their standout approach creates personas for change management by analyzing organizational data to build archetypes, which they use as a testing ground to improve deliverables and navigate complexity.
From broad strategy to concrete change management personas
The scope of Seneca's AI deployment reveals how HR teams can think systematically about where AI adds value. Rather than piloting AI in isolation, Diana Mouhan's team embedded it across the full talent lifecycle—from planning headcount needs to designing how compensation aligns with organizational strategy.
The persona-building approach stands out because it translates abstract organizational complexity into human profiles. By analyzing data at the thematic level, the team creates archetypes that represent distinct employee segments or change readiness profiles . These personas become a testing mechanism: HR leaders can stress-test new policies, communication strategies, or process designs against each persona before rolling out to the full organization.
As discussed in this episode of the What Works podcast , this method helps teams anticipate friction and refine solutions early. It's particularly powerful during organizational change, where one-size-fits-all communication often fails. Personas reveal where different groups sit on adoption curves, what concerns matter most to each segment, and how to craft messaging that resonates.
"Trust isn't just telling people don't be afraid. Trust is actually built on acknowledging the concerns that they have and being super vulnerable in that journey with them."
Diana Mouhan — Executive Director of Transformation and Change at Seneca Polytechnic. With over a decade of expertise in organizational strategy and AI adoption, Diana has built transformation practices across Deloitte and Kearney before leading Seneca's innovation hub, where her team drives HR modernization through data-driven personas and adaptive change strategies.
The persona approach also echoes a broader insight from Seneca's journey: understanding where your team sits on the AI readiness spectrum is the foundation for any sustainable deployment. What Diana's team discovered was that readiness ranges dramatically—some staff members still needed to learn what a prompt was, while others were already experimenting with advanced use cases.
The compensation analysis and job architecture applications show that AI doesn't have to be flashy to deliver value. Market data, internal pay equity assessments, and role-based competency mapping are all areas where AI accelerates accuracy and reduces manual analysis time. These use cases often deliver faster returns than more experimental applications.
If you're mapping where AI fits in your own HR operation, Diana's full conversation covers how she prioritized which areas to tackle first and how budget constraints actually forced her team to be more creative in using existing tools.
AI delivers fastest impact when deployed across the full talent lifecycle (planning, design, compensation, change management) rather than isolated pilots.
Persona-building using organizational data creates testable archetypes to validate HR strategies before rolling out organization-wide.
Change management personas reveal readiness profiles and allow HR leaders to craft targeted communications for different employee segments.
Foundational use cases like job architecture and compensation analysis often deliver steadier ROI than more experimental AI applications.
What assessment approach reveals HR team readiness levels for AI adoption?
Start by listening to where each person actually stands on their AI journey—without assumptions. Only 25 to 30 percent of Seneca's HR team initially trusted AI enough to use it in the workspace , a baseline that revealed the true readiness gap and enabled the team to build meaningful shifts in confidence.
The assessment Diana Mouhan's team at Seneca Polytechnic employed was fundamentally human-centered. Rather than deploying surveys or formal readiness audits, they spent time in genuine conversation with each team member to understand their starting point.
This listening approach uncovered a fragmented landscape: some employees didn't yet know what a prompt was, while others were already experimenting independently with AI tools. Mapping this spectrum proved essential , because it showed the team that readiness wasn't binary—it was a continuum that demanded different interventions for different groups.
Knowing that trust was the foundation, Diana's team recognized that the 25–30% figure became their starting point for designing messaging, training, and peer-led adoption pathways. The baseline revealed which skeptics needed reassurance and which early adopters could become internal champions.
"Trust isn't just telling people don't be afraid. Trust is actually built on acknowledging the concerns that they have and being super vulnerable in that journey with them."
Diana Mouhan — Executive Director of Transformation and Change at Seneca Polytechnic. Diana brings over a decade of organizational strategy and AI adoption expertise, with prior leadership roles at Deloitte and Kearney. Her work focuses on redesigning HR operations to embrace technology while maintaining human-centered change management.
The specificity of this finding—not just "we measured trust," but "we discovered exactly what percentage trusted us and built from there"—reveals why listening as an assessment tool outperforms formal readiness models. Diana's approach at Seneca shows how vulnerability and acknowledgment anchor adoption faster than mandates or cheerleading alone.
As the team rolled out AI-driven tools like ServiceNow and knowledge-management platforms, this readiness baseline informed every design choice —from how they framed AI's role to which features they prioritized. The result wasn't instant universal adoption, but rather a sustainable rhythm in which skeptics gradually moved toward experimentation once their real concerns were named and addressed.
From baseline trust to measurable behavioral shift
The 25–30% baseline wasn't a ceiling; it was a launchpad. Once identified, Seneca's team redirected their entire change strategy to close that gap systematically. Rather than assuming everyone was equally ready, they could now segment communication, training, and peer support according to where each cohort actually stood.
This granular readiness picture also revealed allies: the 25–30% who already trusted AI became mentors and co-creators of the transformation journey. Their confidence became visible to hesitant colleagues, accelerating social proof far more than any leader's mandate could.
What is the foundation for building trust in AI adoption within an HR team?
Trust in AI adoption isn't built through reassurance alone—it requires acknowledging team members' concerns vulnerably and openly, preserving human judgment and employee agency, and anchoring accountability with leadership . Diana Mouhan emphasizes that bringing leadership together to co-create use cases and develop shared AI principles transforms fear into partnership.
When Seneca Polytechnic began its AI adoption journey in HR, the team faced a common barrier: fear. Yet Diana Mouhan recognized that telling people not to worry about AI replacing their jobs would ring hollow. Instead, she took a fundamentally different approach centered on honesty and inclusion.
The foundation for trust starts with acknowledging real concerns rather than dismissing them. Diana brought her transformation team and HR leadership together not as a top-down mandate, but as partners in shaping how AI would actually work within the organization. This vulnerability—admitting uncertainty and inviting the team to solve the problem together—signaled that leadership wasn't operating from a hidden agenda. They were genuinely exploring the technology alongside their teams.
A critical element was making explicit what would and would not change. As Diana explains in the episode , the team reinforced that human judgment remains with employees and HR specialists —AI would be a tool for enhancement, not replacement. This distinction matters deeply: workers retain final decision-making authority in their domains, and their expertise isn't being automated away.
"Trust isn't just telling people don't be afraid. Trust is actually built on acknowledging the concerns that they have and being super vulnerable in that journey with them."
Diana Mouhan — Executive Director of Transformation and Change at Seneca Polytechnic, a transformation specialist with over a decade of organizational strategy and AI adoption experience, previously at Deloitte and Kearney.
Equally important is clarity about where accountability lives. Leadership owns the responsibility for how AI is deployed, what it optimizes for, and the outcomes it produces. This boundary protects team members from feeling that AI failures or misaligned decisions are somehow their burden to carry. When people see leadership standing behind both the technology and its ethical application, trust deepens significantly.
The practical mechanism for building this trust at Seneca was experimentation and co-creation. Rather than announcing AI solutions, Diana's team ran pilots with HR staff, gathered their input on what worked and what didn't, and revised principles collaboratively. This approach generated buy-in before full rollout —people weren't surprised by technology arriving; they'd helped shape it. You can explore more about how Seneca redesigned its entire transformation strategy when funding constraints emerged , revealing how trust and flexibility work hand-in-hand.
Initial adoption metrics at Seneca reflected the challenge: only 25% to 30% of the HR team initially trusted AI for workspace utilization. Yet by grounding trust-building in vulnerability, transparency, and shared ownership, the team moved past fear toward genuine partnership with the technology.
Trust requires acknowledging and validating team concerns openly, not dismissing or minimizing them.
Human judgment and decision-making authority must remain clearly with employees and specialists, not be absorbed by AI.
Leadership must own accountability for AI outcomes and ethical deployment, shielding teams from carrying that burden.
Co-creation and pilot-based experimentation build buy-in and trust before enterprise-wide rollout.
How can an HR organization continue transformation when technology funding disappears unexpectedly?
Rather than pausing the transformation until funding returns on an indefinite timeline, redesign the transformation itself to work with available tools and reduced dependencies. This means shifting from "what technology must we invest in?" to "how do we accomplish our goals with what we already have?"
From perfect vision to practical reality
When Seneca Polytechnic faced an unexpected budget cut in the higher education sector in Canada—stemming from structural changes in regulatory funding—Diana Mouhan's transformation team faced a binary choice: pause everything until money returned, or rebuild their roadmap entirely.
They chose the latter. The team stopped asking which new technologies they needed to acquire and started asking fundamentally different questions: how to redesign with existing tools , how to reduce funding requirements without abandoning vision, and how to adapt to a leaner operating model. As Diana Mouhan explains in the episode , this wasn't a retreat—it was a recalibration.
The shift required vulnerability and transparency with the team about constraints. Rather than presenting a false sense of unlimited possibility, Mouhan's leadership acknowledged the real barriers and invited the organization to problem-solve within them. This honest approach actually strengthened trust and ownership across the transformation office.
Practical outcomes of constraint-driven design
Working within these tighter parameters, Seneca's HR team automated processes using ServiceNow—a platform they already had—and redesigned their knowledge management. Over 100 knowledge articles were refreshed to enable more self-service, driving employee response times down from nearly two weeks to less than 24 hours .
More than 90 percent of employee interactions moved to the ServiceNow platform, a result achieved not through expensive new technology but through thoughtful redesign of existing systems and clearer user pathways. The transformation continued forward, proving that funding constraints don't have to mean transformation pause—they can mean transformation refocus, as detailed in this podcast conversation .
"Trust isn't just telling people don't be afraid. Trust is actually built on acknowledging the concerns that they have and being super vulnerable in that journey with them."
Diana Mouhan — Executive Director of Transformation and Change at Seneca Polytechnic. Diana is a transformation specialist with over a decade of experience in organizational strategy, AI adoption, and the future of work. Her career spans roles at Deloitte and Kearney, bringing deep expertise in navigating organizational change and technology transformation during uncertainty.
What made Seneca's pivot especially instructive is how Diana Mouhan's team maintained their strategic vision while fundamentally changing their operational approach. Rather than scaling back ambition, they scaled back assumptions about how the work had to be done. Learn more about how her team maintained agility and managed stakeholder expectations in the full episode on Listenly .
Transformation doesn't require new technology—redesign your roadmap to maximize existing tools and capabilities.
Shift from "what do we need to buy?" to "what can we achieve with what we have?" as your guiding question.
Acknowledge constraints openly and invite your team to problem-solve within them—this builds trust and ownership.
Practical redesign can yield measurable results: Seneca reduced response times from two weeks to under 24 hours using existing systems.
Inside Databricks: The Dynamic Learning Operation That Fuels Hypergrowth
What is the adoption curve when introducing AI-driven enablement tools to traditional L&D and business teams?
Greater than 60 percent of people initially fear that AI will replace their jobs , but this resistance dissolves when organizations shift messaging to emphasize human-AI collaboration and the human's critical role in guiding the technology. Once teams understand they remain the intelligent actor and that AI makes them better at their work, adoption accelerates naturally.
The fear barrier isn't a technical problem — it's a confidence one
When AI-driven enablement tools first land in a traditional L&D or business team, the reaction is predictably defensive. The fear that technology will make their role redundant is immediate and widespread. This isn't an edge case—it's the baseline reaction for the majority of people encountering AI for the first time in their workflow.
Yet Roshana Golani explains in the podcast , the fear itself isn't the barrier to adoption. The barrier is what the organization communicates about the role of humans in the system . When messaging centers on partnership rather than replacement, and when people see concrete proof that they're the decision-maker and the technology is their amplifier, the adoption curve shifts dramatically.
Moving from fear to competence through visibility
The turning point comes when team members actually use the tool and recognize their own authority within the system. They see that they still own the judgment —choosing what to prioritize, what to refine, what to reject. The AI surfaces patterns and accelerates research; the human chooses direction. This division of labor, once visible, reverses the fear narrative entirely.
As discussed at length in this episode , organizations that frame AI enablement this way—not as replacement, but as a tool that makes skilled professionals more capable—see team members shift from resistance to enthusiasm. They begin discovering new possibilities within their own roles because they're freed from routine work and empowered to apply their expertise at a higher level.
Roshana Golani — VP of Enablement at Databricks, with over two decades of experience leading enablement and skills transformation across major technology transformations at VMware, Amazon, and Google. Her career has focused on enabling people to thrive during periods of disruptive technology adoption, from virtualization and cloud migration to data and AI ecosystems. At Databricks, she reports directly to a co-founder and leads enablement across go-to-market functions including sales, marketing, and product engineering, helping field organizations scale capability with rapid business growth.
A practical insight from the conversation: the episode also covers how solution architects apply AI tools daily , revealing where adoption friction actually lives and how to meet teams where they work—a concrete pattern that translates across L&D and business enablement functions.
The majority of teams start from a place of job displacement anxiety, not technical skepticism—messaging and visibility are the levers that move adoption.
Adoption accelerates when team members see themselves as the active intelligence guiding the system, not as passive users of it.
AI makes skilled professionals more capable by handling routine work, allowing them to apply expertise at a higher level and discover new opportunities in their own roles.
Real-world visibility into how peers use the tool—seeing concrete examples and trusted colleagues benefiting—is often more persuasive than any top-down communication.
How can data integration from multiple organizational systems enable skills inference and personalized enablement at scale?
A unified data lakehouse consolidating Salesforce, product infrastructure, and customer interaction data allows organizations to infer actual employee skills and performance patterns in real time . This enables systems to understand what employees are doing, where they're doing it, and deliver enablement in their immediate context—a capability that works across industries from software to retail and manufacturing.
Organizations sitting on fragmented data across multiple systems—CRM platforms, product usage logs, customer communication records—often miss a critical insight: the patterns already exist in that data . By bringing these sources together into a single analytical environment, an organization gains visibility not just into what happened, but what skills and capabilities its people actually demonstrate.
The practical value emerges when this consolidated view becomes actionable. As Roshana Golani explains in The Josh Bersin Company podcast , the question becomes: "We need to understand what is it that the solution architects were doing? Where were they doing it? And can I bring it to them in a way that makes sense?" That shift—from retrospective skill assessment to real-time, context-aware enablement—defines the scale at which modern organizations can operate.
This approach transcends industry boundaries. Software companies benefit from understanding which product features their teams are mastering. Retail organizations use the same logic to optimize staff training against actual customer interaction patterns. Manufacturing firms apply it to floor operations and supply chain skill development. The underlying principle remains constant: data reveals capability, and capability drives enablement decisions .
The implementation complexity—integrating Salesforce with internal product telemetry, customer systems, and learning platforms—is real. But organizations like Databricks have proven this model works at scale , serving over 20,000 customers globally. The data lakehouse architecture itself becomes the enabler: a single source of truth capable of processing both historical skill patterns and real-time performance signals.
"We need to understand what is it that the solution architects were doing? Where were they doing it? And can I bring it to them in a way that makes sense?"
Roshana Golani — VP of Enablement at Databricks. Golani brings deep expertise in skills transformation across disruptive technology waves, having led enablement initiatives at VMware, Amazon, and Google during virtualization, cloud migration, and data/AI adoption cycles. At Databricks, she sits within the field engineering organization, reporting to a co-founder, and oversees enablement across sales, marketing, and product engineering functions.
For those curious about how Databricks operationalizes this strategy internally, the full episode explores how tools like Genie and Skills Navigator automate skill discovery and training content generation based on the unified data model.
A unified data lakehouse consolidates Salesforce, product infrastructure, and customer interaction data to reveal real employee skills and performance patterns.
Real-time, context-aware enablement becomes possible when organizations understand not just what employees need to learn, but where and how they work.
This capability-driven approach scales across industries—software, retail, manufacturing—because the principle of data-driven skill inference is industry-agnostic.
Successful implementation requires integrating multiple data sources into a single analytical environment, enabling both pattern recognition and personalized learning delivery.
What role do subject matter experts and non-technical business users play in AI-powered learning systems?
Non-technical business users tend to adopt and innovate with AI technology more effectively than technologists because they focus on solving specific business problems rather than engineering complexity . They approach tools pragmatically, which accelerates real-world impact, while subject matter experts remain critical for providing domain knowledge and ensuring accuracy.
This pattern reveals a fundamental difference in how each group approaches technology. When technologists encounter AI tools like Databricks, they often spend time optimizing infrastructure, debating architectural decisions, or exploring edge cases. Business users, by contrast, ask a simpler question: "How does this solve my immediate problem?"
Program managers without technical backgrounds have successfully built their own applications and solutions using Databricks, demonstrating that the right platform design can lower barriers to entry. As Roshana Golani discusses in the episode , the technology should be architected specifically to make it easy for both subject matter experts and non-technical users to contribute, rather than requiring deep engineering knowledge before someone can create value.
The key insight is that business friction, not technical sophistication, should drive adoption . When organizations remove the friction points between a business user's need and the ability to address it with AI, adoption accelerates naturally. This doesn't diminish the role of subject matter experts—their knowledge of domain rules, edge cases, and quality standards remains irreplaceable—but it does reshape how enablement and product design should prioritize accessibility.
"We need to understand what is it that the solution architects were doing? Where were they doing it? And can I bring it to them in a way that makes sense?"
Roshana Golani — VP of Enablement at Databricks. With a career spanning VMware, Amazon, and Google, Golani has pioneered skills transformation and enablement during major technology shifts, including virtualization, cloud adoption, and now AI. At Databricks, she leads enablement across sales, marketing, and product engineering within the field organization, reporting to a company co-founder.
This principle challenges the traditional assumption that technical depth is the prerequisite for AI innovation. Instead, it suggests that organizations seeking to scale AI adoption should prioritize designing systems that meet users where they are, not where engineers wish they were. The episode explores how Databricks structures its own internal learning and enablement to operationalize this insight at scale across its customer base and go-to-market functions.
How can AI systems like Genie generate personalized training content dynamically rather than relying on static course libraries?
AI systems consume multiple input sources—PRDs, product documentation, webinars, and knowledge transfer sessions —then dynamically generate customized training content tailored to individual learner needs and context. The system maintains instructional quality through human-in-the-loop oversight and uses ontology plus product telemetry to ensure content remains relevant to specific use cases, such as teaching a solutions architect how to demo Databricks to a manufacturing customer.
From Static Libraries to Real-Time Customization
Traditional training platforms lock content into predetermined courses. Dynamic AI systems like Genie fundamentally invert that model by treating training as a real-time response to actual job context , not a one-size-fits-all repository.
Rather than asking "What course should this person take?", the system asks "What is this person trying to do right now, and what knowledge do they need to succeed at it?" This shift is explained in detail in this episode of The Josh Bersin Company , where enablement leaders discuss how AI is reshaping learning operations.
The Human-in-the-Loop Safeguard
Raw AI generation without oversight risks hallucination and instructional errors. Databricks maintains human review at critical checkpoints , ensuring that dynamically generated content meets pedagogical standards before it reaches learners.
This hybrid approach—AI for speed and scale, humans for judgment and accuracy—is what separates responsible enablement from reckless automation. As Roshana Golani emphasizes in the full podcast episode , the goal is not to replace instructional expertise but to amplify it.
"We need to understand what is it that the solution architects were doing? Where were they doing it? And can I bring it to them in a way that makes sense?"
Roshana Golani — VP of Enablement at Databricks. Golani brings over a decade of experience transforming how organizations skill their workforces across disruptive technology cycles—from virtualization and cloud migration to AI adoption. She sits within Databricks' field engineering organization, directly accountable to a co-founder, and leads enablement across sales, marketing, and product engineering functions.
That question captures the entire philosophy: context matters. Personalization means understanding not just who is learning, but where and when they need to learn it. A deeper dive into how Databricks operationalizes this principle is available in the full conversation .
Ontology and Telemetry: The Intelligence Layer
Behind every personalized recommendation sits a knowledge map. Ontology —a structured model of how concepts relate—allows the system to understand which skills connect to which outcomes. Telemetry from actual product usage shows where learners struggle and what they're trying to accomplish.
Together, these inputs let Genie infer not just what content to generate, but how to frame it. Teaching someone to demo Databricks to a manufacturing customer requires different emphasis than teaching a data engineer to optimize a pipeline. The system adapts both substance and tone based on audience and use case.
For more on how Databricks integrates its internal tool stack—Skills Navigator, Genie, and others—to create a cohesive learning operation, listen to Roshana Golani's full interview .
What is Skills Navigator and how does it infer employee skills without traditional taxonomies?
Skills Navigator is an internally-built tool at Databricks that uses a skills inference engine running on the Databricks data lakehouse to automatically detect employee capabilities from their actual work activities. Rather than requiring a pre-built skills taxonomy, the system starts with an initial taxonomy and continuously refines it through data, learning what actual skills look like as employees perform their jobs.
The power of Skills Navigator lies in its ability to observe real work patterns rather than rely on employees to self-assess or managers to manually categorize competencies. By analyzing how employees actually spend their time—the projects they work on, the tools they use, the problems they solve—the system builds a living, evolving picture of what skills genuinely exist within the organization.
As Roshana Golani explains in the episode , this data-driven approach enables managers and employees to have more informed conversations about proficiency development. Instead of guessing what skills matter or waiting for an annual performance review, they can reference concrete evidence of capability—what someone has actually demonstrated through their work.
From static taxonomy to continuous learning
The traditional approach to skills management relies on a fixed taxonomy—a predefined list of competencies that often becomes outdated the moment a new technology or role emerges. Skills Navigator sidesteps this problem by treating the taxonomy as a starting point, not a destination. The system learns continuously as it observes employee behavior , updating its understanding of which skills are most relevant and how they combine in real work contexts.
This dynamic refinement matters especially in fast-moving industries like data and AI, where Databricks operates. New techniques, tools, and job requirements emerge constantly, and Skills Navigator adapts without waiting for a manual taxonomy refresh. The inference engine does the heavy lifting: as discussed in this podcast , the system understands what solution architects and other technical roles actually do, and it brings relevant skill insights to those individuals in real time.
"We need to understand what is it that the solution architects were doing? Where were they doing it? And can I bring it to them in a way that makes sense?"
Roshana Golani — VP of Enablement, Databricks. Roshana has spent her career at the intersection of disruptive technology and organizational transformation, working previously at VMware, Amazon, and Google where she led enablement initiatives during major shifts including virtualization, cloud migration, and data/AI adoption. At Databricks, she reports directly to a co-founder and drives enablement strategy across sales, marketing, and product engineering functions.
The inference engine's ability to work in context—delivering insights and guidance exactly where employees need them, in the flow of their actual work—represents a fundamental shift. Rather than treating skills as abstract competencies tracked in an HR database, Skills Navigator embeds skill awareness into the daily work experience itself, making proficiency development actionable and immediate.
What makes this approach particularly powerful is that it removes friction from skill recognition. Employees don't need to remember to update a skills profile; managers don't need to conduct manual assessments; and the organization doesn't need to maintain an expensive taxonomy infrastructure. The data lakehouse becomes the source of truth , capturing real evidence of what people can do, and the inference engine translates that evidence into meaningful development conversations.
Skills Navigator infers capabilities directly from observed work activities, removing reliance on self-reported or manually-curated skill taxonomies.
The system continuously refines its understanding as employees perform their jobs, ensuring the skills model stays relevant in fast-moving fields like data and AI.
Rather than treating skills as static HR data, Skills Navigator embeds proficiency insights into the flow of work itself, enabling real-time development conversations.
Built on the Databricks data lakehouse, the inference engine uses actual work patterns as the authoritative source of employee capability, removing guesswork from skills management.
How can organizations move from traditional push-based training certification models to AI-driven performance enablement in the flow of work?
Stop designing courses first. Instead, identify the actual job being performed and deliver just-in-time training exactly when the person needs it , using telemetry from the tools they already work in. By measuring real performance and productivity—not test scores—enablement becomes personalized and context-aware, shifting from "in case you need this someday" to "here's what you need right now."
Moving from certification mindset to performance measurement
Traditional L&D asks: "What course should we build?" The new model asks: "What is the person actually trying to accomplish right now?" This is not a small shift—it fundamentally changes where and when learning happens. As Roshana Golani explains in The Josh Bersin Company , the real power comes from embedding support directly into the workflow , not pulling people out of their work to sit in training.
The measurement changes too. Instead of tracking course completion rates or certification badges, organizations measure what actually matters: did productivity go up? Did the sales rep close the deal? Did the solution architect solve the customer's problem? Telemetry from CRM systems, product infrastructure, and collaboration tools reveals exactly where gaps exist and who needs help.
Building context-aware, just-in-time enablement at scale
At Databricks, which serves over 20,000 customers globally and grows 80 percent year over year, the scale of traditional training is impossible. You cannot build enough courses. Instead, Golani's team uses data from where work actually happens—tools like Salesforce, Slack, and the product itself—to understand what solution architects and sales engineers are doing and where they struggle.
This approach means personalization becomes automatic, not manual . Rather than offering the same curriculum to everyone, the system identifies that Person A needs help with a specific integration in this customer conversation, while Person B needs coaching on discovery questions. The training is delivered in context, at the moment of need, often via conversational AI or micro-learning tools integrated into existing workflows. As Golani notes, "We need to understand what is it that the solution architects were doing? Where were they doing it? And can I bring it to them in a way that makes sense?"
The shift also addresses a deeper truth about transformation: over 60 percent of people initially fear that AI will replace their jobs . When enablement feels personalized and supportive rather than mandatory compliance, it becomes a signal that the organization is investing in people, not just implementing technology.
"We need to understand what is it that the solution architects were doing? Where were they doing it? And can I bring it to them in a way that makes sense?"
Roshana Golani — VP of Enablement at Databricks. Golani has led skills transformation and enablement at major technology companies including VMware, Amazon, and Google, guiding organizational change during periods of virtualization, cloud migration, and now data and AI adoption. At Databricks, she sits within the field engineering organization and reports directly to a co-founder, giving her unique authority to reshape how the entire go-to-market team—spanning sales, marketing, and product engineering—learns and performs.
For a deeper look at how this plays out in practice at a hypergrowth company, listen to Roshana discuss Databricks' learning infrastructure and the specific tools like Skills Navigator that power this model .
Start with the job and the workflow, not with course design—understand where people are struggling before you build enablement.
Use telemetry and data from real work tools (CRM, product, messaging) to identify gaps and deliver micro-learning in context, not classroom training.
Measure performance outcomes—productivity, deal closure, customer satisfaction—rather than certification completion rates.
Personalized, just-in-time enablement signals organizational investment in people and reduces fear around technology adoption.
The Messy World of AI-Powered Recruiting Where Nobody Is Happy
What are the key limitations of AI video interviewing tools currently?
Even the most sophisticated AI video interviewing tools from companies like Eightfold and HireVue are designed primarily for screening, not final assessment . While they offer candidates the convenience of applying without scheduling constraints, they fail to capture the subtle environmental and interpersonal signals that emerge in face-to-face interactions, and rejected candidates often have no one to contact for meaningful feedback.
Screening versus assessment: the real limitation
AI video interviewing tools excel at volume processing—they can handle thousands of applications in parallel without the bottleneck of scheduling human time. Yet as Josh Bersin explains in The Josh Bersin Company podcast , this efficiency comes at the cost of depth. These platforms remain effective primarily for initial screening phases , where the goal is simply to filter candidates by basic qualifications or communication readiness.
The moment a hiring decision requires nuance—assessing how a candidate collaborates, handles pressure, or fits a team's culture—the AI tool's limitations become apparent. A video interview cannot replicate the organic flow of conversation, the unscripted moments where real competencies and personality emerge, or the mutual evaluation that happens when recruiter and candidate truly engage.
The feedback void: rejection without dialogue
Perhaps the most damaging gap in current AI video interviewing systems is their lack of feedback mechanisms for rejected candidates . When a human recruiter declines a candidate, that person can at least request a conversation, ask what went wrong, or understand where to improve. With an AI tool, rejection is often a silent door.
Candidates who invest time in recording responses to algorithmic questions receive no explanation—no insight into which skills were flagged as insufficient, whether it was their communication style, tone recognition, or something else entirely. This absence of dialogue creates frustration and erodes trust in the hiring process itself, especially when candidates have no recourse or clarity.
The broader challenge, as discussed in this episode of The Josh Bersin Company , is that hiring costs money—sometimes $5,000 to $10,000 or more for a skilled role , with executive placements reaching up to half of the first-year salary in recruiter fees. When AI tools remove human judgment and dialogue from the equation, they may accelerate screening but ultimately create a less robust and less humane hiring experience.
Josh Bersin — Global Industry Analyst & CEO of The Josh Bersin Company, a research and advisory firm focused on talent, learning, and organizational effectiveness in the enterprise. Bersin has spent decades analyzing the talent acquisition industry, benchmarking hiring costs, and tracking how technology reshapes recruitment workflows. His analysis spans the full recruiting lifecycle, from job board evolution to modern AI-driven screening and assessment tools.
One concrete detail worth exploring further: the episode digs into how AI-generated resumes are now flooding the system , creating a cascade effect where AI consumes AI-written content, making the screening phase even noisier and less reliable.
AI video interviewing tools are optimized for high-volume screening, not nuanced final assessments or leadership evaluation.
These systems fail to capture the interpersonal and environmental cues that human interviewers naturally pick up in real conversations.
Rejected candidates typically receive no feedback or dialogue from AI-driven tools, creating a frustrating and opaque experience.
The cost of hiring a skilled worker ($5,000–$10,000+) demands human judgment and clear communication—elements AI screening alone cannot replace.
What role does industrial-organizational psychology play in the future of AI recruiting?
Industrial-organizational psychology integrated into AI recruiting creates job-specific, culture-aligned assessments that filter candidates more effectively than generic algorithms alone. Before deploying AI to source candidates, companies work with a consulting firm or the AI provider to build realistic assessments tailored to the role and organizational culture — a process that takes more upfront time but produces significantly better candidate quality.
The traditional recruiting AI stack often automates at scale without understanding what actually predicts success in a specific role or company. This creates a mismatch: the AI floods recruiters with resumes, but many lack genuine fit. Smarter IO assessments flip this approach by embedding proven hiring science into the AI model itself.
Companies like Maki People and Paradox exemplify this model. Rather than letting AI cast the widest possible net, they work upfront with employers to define what job success actually looks like — technical skills, cultural values, communication style, and role-specific competencies — and then build assessments that filter for those attributes from the start.
The upfront investment that saves time downstream
This approach requires more initial effort than deploying an off-the-shelf resume parser. Teams must collaborate to answer hard questions: What does a high performer in this role actually do? Which soft skills matter most? How does this role interact with team dynamics and company culture? The payoff emerges in higher-quality candidate pipelines and fewer mismatches that lead to costly bad hires.
When IO psychology is absent, the result is what recruiters describe as a flood — high volume, low signal. As Bersin explains in the episode , the recruiting industry is caught in an arms race where AI creates resumes and AI consumes them, making it harder for hiring teams to surface genuine talent.
"We have AI creating resumes and producing and sending them out and AI consuming them — slop talking to slop — and the recruiters constantly tell us they're getting so flooded with fraudulent resumes that they really can't sort through it all."
Josh Bersin — Global Industry Analyst & CEO, The Josh Bersin Company. Bersin specializes in talent acquisition and organizational performance, drawing on decades of research and benchmarking data on hiring costs, time-to-fill, and candidate quality across industries.
In this context, IO psychology becomes a shield — a systematic way to define signal and filter out noise. The full episode explores how this plays out across different job types and skill levels , and why some companies are already seeing measurably better outcomes by investing in smarter assessments upfront.
How should companies differentiate their AI recruiting approach by job type?
For entry-level or low-skilled roles, companies can automate heavily with AI, moving through a quick screening process focused on basic skills, availability, and physical requirements. For highly technical, licensed roles — like nurses, electricians, and software engineers — and especially for executives and professionals, companies should invest significantly more in human recruiting and validated assessments, because the cost of a bad hire and the difficulty of replacement are much higher.
The recruiting process carries vastly different stakes depending on the role. Hiring costs range from hundreds of dollars for high-volume, low-skilled positions to $5,000–$10,000 or more for highly-skilled professionals. For senior engineers or executives, companies may pay almost one-third to one-half of the first-year salary in recruiter fees alone — a massive investment that demands careful vetting.
Automation Works for High-Volume, Low-Skill Roles
Entry-level and low-skilled positions are ideal candidates for heavy AI automation. These roles typically prioritize speed, basic competency matching, and rapid screening over deep evaluation of culture fit or advanced capability. Resume parsing, automated phone screening, and AI-driven initial qualification work well here because the cost of a hiring mistake is contained and replacement is straightforward.
Companies in retail, customer service, logistics, or manufacturing benefit most from this approach. The volume is high, turnover is expected, and the hiring cycle can be compressed without significant risk. An AI system that flags candidates with the right availability, location, or baseline skills accomplishes the goal efficiently.
Human Expertise and Validated Assessment for High-Stakes Roles
Skilled and professional positions demand a different strategy. Human recruiters and validated assessment tools become cost-effective investments when a hiring error carries serious consequences — a misplaced nurse puts patient safety at risk, a bad software engineer compounds technical debt, and a wrong executive pick damages strategy and culture.
As discussed in this episode of The Josh Bersin Company , roles requiring technical certifications, specialized knowledge, or leadership capacity need human judgment to evaluate nuance, potential growth, and cultural contribution. AI can support this process — skills assessment tools like Galileo help, for example — but the final evaluation must involve seasoned recruiters and hiring managers.
The alternative — over-automating these roles — risks flooding your team with technically marginalized candidates or, worse, introducing fraud. Even advanced screening faces challenges when AI systems are exposed to fraudulent resume data in the first place.
"We have AI creating resumes and producing and sending them out and AI consuming them — slop talking to slop — and the recruiters constantly tell us they're getting so flooded with fraudulent resumes that they really can't sort through it all."
Josh Bersin — Global Industry Analyst & CEO, The Josh Bersin Company. Bersin leads research and advisory services on talent, HR technology, workforce analytics, and organizational culture, advising Fortune 500 companies and HR leaders on the practical application of AI and digital transformation in recruiting and talent management.
This friction reveals why human judgment remains irreplaceable for critical roles. A recruiter can smell inconsistencies, sense cultural alignment, and dig deeper when something feels off. An algorithm cannot — not yet.
For more detail on how AI tools themselves vary in reliability, and how the choice of AI model can bias hiring outcomes, listen to the full episode on Listenly , which also explores the geopolitical dimension of recruiting fraud and the organizational safeguards companies need in place today.
How does AI perform on skills assessment compared to holistic candidate evaluation?
AI tools perform strongly at skills-based assessment—identifying required skills for specific roles, spotting gaps in resumes, and generating behavioral interview questions. But they fall short at capturing the human dimension: company culture fit, interpersonal dynamics, and subtle aptitudes that only face-to-face conversation reveals.
What AI does well in recruitment
AI recruiting platforms loaded with comprehensive skills libraries, such as Galileo, which taps Lightcast's massive skills database , excel at the technical side of candidate screening. They can quickly parse a resume, extract competencies, and benchmark them against the official skill requirements for any role—whether that's a flight attendant or baggage handler for an airline.
The practical output is clear: AI can identify skill gaps automatically , flag missing certifications, and even generate targeted behavioral interview questions tailored to the role's actual needs. This standardization and speed represent a genuine productivity gain, especially for high-volume hiring where human screeners would take weeks to reach the same assessment.
The human signals AI cannot detect
The hard limit appears when hiring moves beyond skills matching. AI struggles with the intangible human signals that determine whether a candidate will thrive in a team or align with company values. Company culture, interpersonal chemistry, communication style under pressure, and non-writing aptitudes—empathy, leadership presence, creative problem-solving in real time—remain invisible to automated systems.
A resume or even a video interview transcript cannot fully capture whether a candidate will click with their future manager, contribute to team dynamics, or adapt to your specific organizational way of working. These judgments still require human presence and intuition, which is why the strongest hiring processes pair AI assessment with in-person or live-video interview stages. The conversation itself, not the skills checklist, often makes the difference.
Josh Bersin — Global Industry Analyst & CEO, The Josh Bersin Company. Bersin is a leading voice on HR technology and talent acquisition trends, known for rigorous benchmarking of recruiting processes, cost-per-hire analysis, and critical evaluation of AI's real impact on talent practices.
What is the North Korean fraudulent engineer threat to AI-powered hiring?
North Korea orchestrated a large-scale operation in which fraudulent software engineers applied to legitimate jobs at banks, insurance companies, and tech firms using AI-mediated hiring systems, conducted fake video interviews, accepted positions, performed some coding work while collecting salaries, and funneled all earnings back to the North Korean government. This scheme exploited the speed and automation of AI recruiting tools, exposing a critical security and identity verification gap in modern hiring processes.
According to reporting covered in The Josh Bersin Company podcast , the Wall Street Journal revealed this operation after investigating hiring anomalies at several major organizations. The fraudsters were not American citizens and did not match the identities they claimed, yet they successfully passed initial AI screenings and automated resume filters that typically reject candidates at scale.
The vulnerability lies in how AI hiring systems prioritize volume and speed over identity authentication. As Josh Bersin explains in the episode , employers have built massive recruiting infrastructure to handle 20 to 30% of Americans changing jobs annually—a high-volume churn that forces companies to automate screening and initial selection. Video interviews, which many organizations use as a first-pass human touchpoint, proved insufficient to catch the deception.
This case also intersects with a broader problem discussed in this podcast analysis of AI recruiting : recruiters report being flooded with AI-generated and fraudulent resumes. When legitimate candidates use ChatGPT to write resumes and fraudsters use the same tools to craft fake profiles, AI systems struggle to distinguish real from fake at scale.
The cost of hiring has risen dramatically—$5,000 to $10,000 or more for skilled roles, and up to one-third of first-year salary for senior engineers—yet companies rushed to automate the process without reinforcing identity verification. The North Korean operation succeeded precisely because AI hiring tools were designed to move candidates quickly through funnels, not to authenticate who they really are.
Why AI systems failed to catch the deception
Traditional resume screening—parsing job titles, skills, experience dates—works on the assumption that the resume belongs to the person submitting it. AI tools trained on legitimate job applications have no pattern to flag a coordinated state-sponsored fraud ring impersonating engineers.
Video interviews added a layer of apparent human verification, but without background checks, reference verification, or rigorous identity confirmation before onboarding, the system remained porous. Companies prioritized speed and candidate experience over friction, creating an opening fraudsters exploited.
"We have AI creating resumes and producing and sending them out and AI consuming them — slop talking to slop — and the recruiters constantly tell us they're getting so flooded with fraudulent resumes that they really can't sort through it all."
Josh Bersin — Global Industry Analyst & CEO, The Josh Bersin Company. A leading expert in talent acquisition, HR technology, and enterprise workforce strategy, Bersin has advised Fortune 500 companies and conducted extensive benchmarking on recruiting costs, AI adoption, and hiring process design across industries.
Is there a proven bias in AI recruiting tools toward resumes written with specific AI models?
Research by computer scientists examining hundreds of resumes found that if a job search tool uses ChatGPT and the candidate also used ChatGPT to prepare…
How has the rise of ChatGPT and off-the-shelf LLMs disrupted AI recruiting tools?
The arrival of ChatGPT triggered a flood of cheap, off-the-shelf AI recruiting tools — while simultaneously giving job seekers the means to auto-generate and mass-submit polished resumes. The result is a broken loop: AI systems screening candidates are now drowning in AI-fabricated applications, a dynamic Josh Bersin bluntly calls "slop talking to slop." Recruiters report being so overwhelmed by fraudulent, AI-generated resumes that they can no longer sort through them effectively.
Hundreds of new tools, and a market that suddenly looked easy to disrupt
Before ChatGPT, building a credible AI-powered recruiting product required significant investment in proprietary models and proprietary data. Companies like Eightfold, HireVue, HiredScore, and SmartRecruiters had spent years developing systems trained on millions of candidate profiles and hiring outcomes. That barrier was largely what kept the market from being overrun.
After ChatGPT launched, that barrier collapsed almost overnight. Hundreds of startups began wrapping large language models into recruiting workflows — resume screening, candidate matching, interview scoring — and selling the result as enterprise-grade AI. As Josh Bersin covers in depth in The Josh Bersin Company podcast , what had been an expensive capability suddenly appeared to be available off the shelf to anyone willing to build a thin product layer on top of a general-purpose model.
Established players like Workday, SAP, LinkedIn, and Paradox had to contend with dozens of new entrants making similar-sounding promises. The market became genuinely hard to navigate — for buyers and candidates alike.
"Slop talking to slop" — when both sides of the hiring funnel automate at once
The disruption didn't only hit the employer side. Job seekers responded to the AI screening wave with their own automation. Tools proliferated to help candidates generate tailored, AI-polished resumes and submit them at scale — sometimes to hundreds of openings simultaneously with minimal human effort.
This created a deeply problematic symmetry. The same technology powering the screening tools was now powering the applications those tools were supposed to evaluate. One particularly striking finding: AI recruiting tools built on ChatGPT are 50 to 75% more likely to accept resumes also written with ChatGPT, compared to resumes written by hand or generated with Claude or Gemini. The system, in effect, rewards matching the model's own stylistic fingerprint — a finding explored further in this episode of The Josh Bersin Company .
The research behind this comes from computer scientists who examined hundreds of resumes across different AI-assisted and human-written conditions. It's not a marginal effect. It points to a fundamental validity problem: tools that are supposed to identify qualified candidates may instead be selecting for candidates who know which AI to use.
"We have AI creating resumes and producing and sending them out and AI consuming them — slop talking to slop — and the recruiters constantly tell us they're getting so flooded with fraudulent resumes that they really can't sort through it all."
Josh Bersin — Global Industry Analyst & CEO, The Josh Bersin Company. Bersin is one of the most widely cited analysts in the HR technology industry, tracking talent acquisition, HR systems, and workforce trends across large enterprises globally.
Recruiters at companies using tools from vendors like SeekOut, Maki People, SHL, or Paradox report the same on-the-ground reality: the volume of inbound applications has surged, but the signal-to-noise ratio has collapsed. Screening at scale was supposed to make hiring faster. Instead, it has made it harder for human recruiters to identify genuinely qualified candidates — a central tension unpacked in The Josh Bersin Company .
"Slop talking to slop" — Josh Bersin's term for the closed loop in which AI tools generate candidate resumes at scale and AI-powered ATS systems consume and score them, with neither side representing genuine human signal. The phrase captures a quality-collapse dynamic specific to the post-ChatGPT recruiting market. The economics of recruiting make this problem particularly high-stakes. Hiring a single highly-skilled employee already costs between $5,000 and $10,000 — and filling a senior executive or software engineering role can require paying an executive recruiter nearly a third to a half of that person's first-year salary. When the screening layer breaks down, those costs don't shrink. They compound. Every bad hire, every wasted screening cycle, every missed qualified candidate adds to a bill that was already significant — a point Bersin has long emphasized on his podcast .
What legal and bias challenges have emerged around AI scoring tools in recruiting?
Multiple U.S. jurisdictions — including New York State and Illinois — have passed laws requiring companies to prove that their AI scoring and inference tools used in hiring are not biased. Two active lawsuits, one against Stateful and one against Workday, are directly challenging these systems. The question of whether AI recruiting tools systematically discriminate remains largely unresolved across the industry.
States Moved First — and Put the Burden of Proof on Employers
Rather than waiting for federal regulation, individual states acted. New York State and Illinois both enacted legislation requiring organizations deploying AI scoring or inference technology in hiring to demonstrate that those systems are not biased. The burden is on the employer — and, by extension, the vendor — to prove compliance.
This is a significant shift. Before these laws, companies could deploy AI screening tools with little obligation to audit for disparate impact. As Josh Bersin details in this episode of The Josh Bersin Company podcast , the legislative pressure is real and already in force — not a future risk.
Stateful and Workday: Two Lawsuits That Could Define the Entire Sector
Two active lawsuits are now testing whether AI-based candidate scoring constitutes illegal discrimination. One targets Stateful, a company whose systems evaluate candidates based on inferred signals. The other targets Workday, one of the most widely deployed HR platforms in the world — meaning a ruling against Workday could have industry-wide consequences.
Neither case has been resolved. As Bersin discusses in The Josh Bersin Company , the outcome of these suits will likely shape how AI scoring is regulated for years. Vendors like HireVue, Eightfold, and HiredScore — all of which use AI to evaluate or rank candidates — are watching closely.
The core legal question is consistent across both cases: does an algorithmic system that produces ranked scores or pass/fail decisions introduce bias along protected characteristics like race, gender, or age? That question, as of now, has no settled legal answer.
"We have AI creating resumes and producing and sending them out and AI consuming them — slop talking to slop — and the recruiters constantly tell us they're getting so flooded with fraudulent resumes that they really can't sort through it all."
Josh Bersin — Global Industry Analyst & CEO, The Josh Bersin Company. Bersin has spent decades benchmarking HR technology and talent acquisition practices. This observation, drawn from direct conversations with recruiting leaders, illustrates how AI-generated input compounds the bias problem: when the data fed into scoring systems is itself distorted, the outputs become even harder to audit or trust.
This dynamic matters for the bias debate. A study by computer scientists examining hundreds of resumes found that AI recruiting tools using ChatGPT are 50 to 75% more likely to accept resumes also written with ChatGPT, compared to those written by hand or with Claude or Gemini. If AI scoring systems are already skewed by the format of the input, auditing for demographic bias becomes significantly more complex — a point Bersin raises in this podcast episode .
The bias challenge is not purely theoretical. With 20 to 30% of Americans changing jobs every year, AI recruiting tools operate at enormous scale. Even a small systematic bias baked into a scoring algorithm can affect millions of candidates annually — which is precisely why regulators and plaintiffs' attorneys are paying attention. The full scope of that context is unpacked in The Josh Bersin Company podcast .
What was Eightfold's pioneering contribution to AI-powered recruiting?
Eightfold was a pioneering company that scraped profiles of individuals from across the internet, anonymized them, and used AI to study skills and technical…
How did AI first enter the recruiting space before large language models existed?
AI in recruiting didn't start with ChatGPT. It began in the era of job boards like Monster.com and CareerBuilder, which shifted job advertising from print to online. Applicant tracking software companies then layered on matching tools that scored resumes against job descriptions — primarily through word matching and pre-LLM BERT technology — before more sophisticated inference engines emerged.
From newspaper classifieds to job boards: the first digital shift
Before the internet, finding an open position required calling contacts, sending letters, or scanning newspaper and magazine ads. Companies like CareerBuilder and Monster.com — now part of Radency — changed everything by moving job advertising online. For the first time, candidates could search and apply at scale, and employers could reach a far wider pool of applicants.
This transition created a new problem: volume. As Josh Bersin explains in this episode of The Josh Bersin Company , the sheer number of incoming applications made manual screening unsustainable. Recruiters needed tools to filter candidates automatically — and that's where early AI entered the picture.
Applicant Tracking System (ATS): Software used by employers to collect, sort, and manage job applications. ATS platforms were among the first to embed automated matching logic — scoring resumes against job descriptions using keyword and phrase detection — well before modern large language models existed. Word matching and BERT: the pre-LLM toolkit that screened millions of resumes
Applicant tracking software companies built matching tools on top of their platforms. These tools compared the words on a resume to the words in a job description, assigning scores based on overlap. It was blunt but scalable — and it defined how most companies screened candidates for years.
The next step was BERT (Bidirectional Encoder Representations from Transformers), a pre-LLM language model developed by Google that could understand context and semantic similarity better than raw keyword matching. Recruiting tools that adopted BERT could, for example, recognize that "managed a team" and "led a group" meant roughly the same thing — a real improvement over pure word matching. This is a technical inflection point discussed directly in The Josh Bersin Company podcast .
After BERT, more sophisticated inference engines emerged — systems capable of modeling candidate fit across broader dimensions than keyword overlap. Companies like Eightfold built talent intelligence platforms on this foundation, moving recruiting AI from simple scoring toward genuine predictive matching. This evolution is covered in depth across Josh Bersin's podcast series .
"We have AI creating resumes and producing and sending them out and AI consuming them — slop talking to slop — and the recruiters constantly tell us they're getting so flooded with fraudulent resumes that they really can't sort through it all."
Josh Bersin — Global Industry Analyst & CEO, The Josh Bersin Company. Bersin has spent decades benchmarking HR technology and talent acquisition practices. This observation captures the unintended consequence of the very automation pipeline that began with early job boards and resume-matching tools.
The quote above underscores how far the original matching logic has traveled. What started as simple word scoring on Monster.com applications has become a fully automated loop — one where AI-generated resumes are now being evaluated by AI-powered screeners. The recruiting context in which these early tools were built, covered thoroughly in this episode , makes that trajectory all the more striking.
How large and costly is the recruiting industry in the United States?
20 to 30% of Americans change jobs every year, even during the worst recessions, making talent acquisition a massive and ongoing industry. The cost of…
