The build vs. buy question has followed people analytics leaders for the better part of a decade. It was never a simple question, but it was at least a familiar one. Should we invest in a custom data platform our team controls, or buy something that gets us to insights faster? But the question just changed in a fundamental way.
The rise of agentic AI, systems that don't just answer questions but plan, reason, and take action across workflows, means that the choice you make about your people data infrastructure isn't just about analytics anymore. It's about whether your organization can safely and effectively participate in the AI-driven enterprise of the next five years.
If you're a CHRO still treating build vs. buy as a reporting and dashboarding decision, you're solving the wrong problem.
For those newer to the conversation: the traditional build vs. buy debate in people analytics comes down to three common paths, each with its own pitfall.
Data is scattered across systems, manual exports, and a people analytics team buried in ad hoc reporting requests. The pitfalls are stagnation and frustration, but it's worse than that. When PA can't move fast enough, the business doesn't wait. Managers pull their own data from Workday, finance builds its own headcount model, and suddenly five people in five departments are reporting five different turnover numbers. Nobody is wrong on purpose. They just don't have a shared source of truth, so they each build their own. The result is an organization making decisions on numbers that don't agree with each other, and no one with the authority or infrastructure to fix it.
This means using internal engineering or a consulting firm to create a custom data warehouse and analytics platform. The appeal is control. The reality, more often than not, is a costly multi-year initiative that is fragile, hard to maintain, and functionally outdated the day it launches. And even when the build succeeds, HR rarely controls the roadmap. Every new data source, every new metric, every change to a calculation goes into an IT queue behind a dozen other priorities. The team that was promised self-sufficiency ends up waiting weeks for changes that should take hours. It's almost never built to support the kind of governed, AI-ready data architecture that the next decade of HR technology demands.
This entails buying a packaged analytics solution that promises fast answers. And it does deliver speed, at first. But the pitfall here is rigidity. You can't adapt the data model to your business. You can't see how metrics are calculated. When a leader asks how you got a number, you can't answer. And when AI predictions surface inside the product, you have no way to audit them.
All three paths lead to the same dead end: slow answers, low trust, and an inability to be the strategic partner the business needs.
Historically, people analytics teams simply had to choose which pitfalls they could endure. However, a new variable has transformed the buy vs. build decision, turning classic pitfalls into existential defects. Here's why.
Agentic AI is qualitatively different from the generative AI tools that captured everyone's attention over the past few years. A generative AI tool helps you write or summarize. An agent reasons through a problem, calls tools, executes steps, and completes workflows, often without a human in the loop for each action.
Organizations across industries are beginning to deploy agents for work that previously required human coordination: scheduling, data gathering, analysis, reporting, anomaly detection. In HR contexts specifically, the use cases are compelling. Agents can connect to workforce data, identify trends, model scenarios, and surface recommendations without waiting for a team to run the analysis manually. A task that might take a team of data engineers weeks can be completed in under an hour.
That's the promise. But the promise only holds if the agents have something reliable to work with.
This is where the build vs. buy decision becomes existential for CHROs.
Agents don't just read data. They act on it. If your people data is fragmented, poorly governed, or locked inside a platform you can't interrogate, the AI will offer misinformed analysis that doesn’t reflect the full reality of your workforce. It’s poised to make incorrect judgments and assumptions about how to govern and model your data. At worst, it will make recommendations that categorically shouldn’t be followed, and someone will act on them.
In this case, you face three bad options:
That last option is already happening. Shadow AI, meaning employees using general-purpose AI tools like ChatGPT to answer questions about headcount, attrition, and workforce costs, is already touching your people data. It's ungoverned, unlogged, and invisible to you. The risk isn't theoretical.
It's a question of when, not if, someone makes a sensitive decision using an AI that had no business accessing the data it used.
This is the part that catches a lot of organizations off guard.
Many CHROs look at their IT team's data warehouse and think the foundation is already in place. The data is centralized. It's secure. It seems like it should be ready for AI. And the obvious next step looks simple: point Copilot or some other AI assistant at the warehouse and let people ask questions.
But a data warehouse, on its own, doesn't do what most people assume it does.
A warehouse stores data. It doesn't calculate your metrics. It doesn't know that your organization defines "voluntary turnover" differently from the HRIS default, or that your headcount logic excludes contingent workers in some business units but not others. Those calculations live in someone's head, or in a spreadsheet, or in a BI tool's configuration. They don't live in the warehouse itself.
A warehouse also doesn't have a semantic layer, meaning the AI doesn’t know how your company defines its metrics, or what leaders actually care about. When an AI agent queries a warehouse directly, it's guessing at what "attrition" means, how to join the tables, which filters to apply. It might return a number. It might even return a plausible number. But it won't return your number, calculated the way your organization has agreed it should be.
And then there's security. A warehouse has database-level access controls, but it typically doesn't enforce the row-level, role-based security that people data demands. Your VP of Sales shouldn't see individual compensation data for the engineering team. A warehouse doesn't inherently prevent that. A governed people data platform does.
The net result is that a CHRO who thinks the warehouse is "close enough" is often in a worse position than they realize.
They have the illusion of readiness without the substance of it.
They'll deploy an AI tool, it will return confident-sounding answers built on incomplete logic and loose security, and the organization will trust those answers because they came from "the system." That's not a foundation. It's a trap.
It's worth being direct about something that often gets glossed over in these conversations: people data is not like other enterprise data.
It's not just commercially sensitive. It's legally sensitive. It's ethically sensitive. Getting it wrong doesn't just mean a bad forecast; it means discriminatory outcomes, compliance violations, and a fundamental erosion of employee trust. The stakes of ungoverned AI acting on people data are categorically higher than those of ungoverned AI acting on inventory or revenue data.
There are several properties that any people data infrastructure needs to have before it can serve as a safe foundation for agentic AI.
These requirements aren't naturally met by an in-house build, a raw data warehouse, or a packaged black-box tool. Each of those paths gets you part of the way there. None of them gets you all the way.
If you're evaluating your people analytics infrastructure in light of the agentic AI moment, here are the questions that matter more than any feature comparison:
Can you explain how your numbers are calculated? Explainability in AI starts with explainability in the data layer beneath it. If you can't trace a metric back to its source and its logic today, an agent built on top of that data won't be able to either.
Can agents connect to your people data safely? The question isn't whether to allow it. Shadow AI means it's already happening. The question is whether it's happening with governance, role-based security, and an audit trail, or without any of those things.
Is your data model flexible enough to keep pace with the business? Agentic AI is most powerful when it can be directed to answer questions your business hasn't thought to ask yet. That requires a data model that isn't locked to last year's org structure or the defaults of your HRIS vendor, and that your team can modify without filing a ticket and waiting three weeks.
Are you building a data foundation that gets smarter over time? The best agentic outcomes don't come from a single data source. They come when AI can connect the dots across your HRIS, engagement surveys, performance reviews, compensation benchmarks, recruiting pipelines, and external labor market data. Today you might ask "what's our attrition rate?" Tomorrow an agent should be able to answer "which teams are most likely to lose high performers in the next quarter, and what's driving it?" That second question requires layers of data working together. If your architecture can only answer the questions you thought to ask at implementation, it's already falling behind.
Most organizations that invested in people analytics infrastructure made reasonable choices with the information they had at the time. If you chose to build in-house, you were optimizing for control. If you chose a packaged tool, you were optimizing for speed. Those weren't bad instincts. But what most buyers didn't fully appreciate was what they were giving up: the ability to see inside the data model, to modify it without dependence on IT or the vendor, and to trust that the numbers being reported were actually calculated the way the business agreed they should be. Those gaps were manageable when analytics was primarily about dashboards and reports. They become serious liabilities when AI agents start acting on that same data.
We built One Model around a belief we've held since 2014: that every company struggles with its HR data, and that the correct management of that data is the real key to becoming successful with people analytics. For years, that meant helping organizations escape the stagnation of the status quo, the fragility of the in-house build, and the rigidity of the black-box tool, by providing an open, governed, flexible platform that gives teams both speed and control.
The foundation we built is what makes agentic AI safe and useful in an HR context. When agents need to reason over workforce data, they need a source of truth that is connected, interpretable, governed, and trusted. They need a semantic layer that translates raw data into the metrics and definitions the business actually uses. They need security that is granular enough for people data. And they need all of this to be maintained and adaptable without a permanent IT dependency. That's what a well-built people data platform provides. It's not an add-on to an AI strategy; it's the center of one.
Organizations that have invested in getting their people data infrastructure right aren't starting from scratch as agentic capabilities mature. They're positioned to move fast because their data is already in a form that agents can use safely and intelligently.
The build vs. buy conversation has always been about enabling your people analytics function to do more. In the age of agentic AI, it's become a conversation about whether you're in control of how AI engages with your most sensitive data at all.
That's a conversation worth having now.
Chris Butler is CEO and Co-Founder of One Model, a people analytics platform that gives HR teams and their AI tools a governed, trusted foundation for workforce data. Learn more at onemodel.co.