AI will expose your People Analytics. Your strengths and your weaknesses.

RAND Corporation, a leading U.S. public policy think tank, recently examined why AI initiatives fail. The takeaway is unambiguous: over 80% do not deliver. Based on 65 interviews with experienced data scientists and AI engineers, two root causes stand out: organizations struggle to scale beyond pilots, and the data they rely on is not fit for purpose.

This deserves a more structural reading. Companies are not failing at AI because they lack technology maturity. They are failing because they invest in the wrong sequence.

For HR leaders, the implication is sharper. HR data has historically been the least structured in the enterprise. The question is no longer whether AI will transform the function. It is what it will reveal.

AI is an amplifier. In a function where data is fragmented, partial and weakly governed, that amplifier can work against you.

 

The AI paradox: widespread experimentation, limited value

This pattern is now visible at scale. MIT NANDA’s study (The GenAI Divide: State of AI in Business, 2025), shows that only 5% of GenAI pilots drive measurable revenue acceleration. The remaining 95% stall, with no impact on the P&L.

Researchers refer to a “learning gap”: a failure to integrate AI into the operational fabric of the company its data, its processes, its teams. A generic model does not compensate for missing foundations. It builds on them and amplifies whatever it finds.

In HR, this is critical. HR Data is historically fragmented across payroll, HRIS, LMS, ATS, performance tools, and countless local spreadsheets. It is often accurate for payroll, but incomplete on skills. Precise on headcount, silent on attrition drivers. Rich in some business units, poor in others.

This is the foundation AI will operate on.

This is also where many organizations fall into a trap: using AI to compensate for a data gap, when AI presupposes data it does not create it.

The outcome is predictable. They end up funding the same problem twice:

1. First, through AI initiatives that fail to create value
2. Then, through data remediation efforts once failure is visible

The MIT finding is not a technology issue. It is a sequencing issue.

And that sequencing problem has a direct consequence: technology is placed upstream, where data should be the foundation.

 

HR data as the foundation

By design, a LLM is generic. It reflects average human language. That is a strength when rewriting an email. It is a limitation when asked to understand the specific realities of a company.

Without structured information about those realities, AI does not generate wrong answers. It generates average ones. And in people-related decisions, “average” is rarely acceptable.

AI enhances the data it is given. If that data is scattered, incomplete or poorly defined, it produces noise rather than signal. It blends accurate and inaccurate inputs without distinction.

AI treats noise as signal. All available data is assumed to be valid.

The priority, therefore, is clear: ensure that AI is fed with data that is both reliable and sufficiently rich to move from generic responses to context-specific insight.

Once the data is usable, the question becomes how AI should be applied.

 

The real blind spot: HR data governance

Before AI, poorly governed HR data created compliance risks and isolated errors.

With AI, it creates systemic decision risk and exposure. Three questions become critical:

• Who has access to what?

This may seem basic. All HR systems have native permission layers. The risk does not come from AI bypassing them. It comes from AI applying them as they are.

The first blind spot is the complexity of access rights. In mature HR systems, permissions are layered over time, often opaque and not systematically updated after role changes, reorganizations or scope shifts.

An AI querying these systems will not flag that an access right has not been reviewed for 18 months. It will simply surface information to the user who queried it even if that access is no longer appropriate.

The second risk lies in HR data outside the system of record. Excel files on SharePoint, interview trackers on shared drives, talent review decks circulating without controls.

When access was limited, obscurity acted as protection. With AI connectivity, that protection disappears. Data can be exposed without any explicit rule being broken.

• What is up to date?

AI does not assess data freshness. It processes what is available: outdated repositories, partially populated fields, abandoned processes that are still documented.

The outputs carry the apparent authority of the model despite flawed inputs.

• What data informed which decision?

AI-driven recommendations on mobility, promotion or compensation must be explainable, challengeable and auditable. This requires clarity upfront on which data sources are used and the ability to demonstrate it afterwards.

Data governance is not bureaucracy. It is the condition for decisions to be defensible, explainable and correctable. Without it, HR faces a double risk:

1. Making decisions distorted by misinterpreted data
2. Being unable to explain or defend those decisions to employees, governing bodies or regulators

AI will not compensate for weak or superficial governance. It will make it visible and amplify its consequences.

 

People readiness: what AI does not change

Clean data and solid governance are necessary but not sufficient. Value only emerges through the teams who use them.

AI dramatically lowers the technical barrier. Building dashboards, structuring reports, cross-referencing datasets, formatting outputs these tasks are now accessible to many. What AI does not replace is analytical discipline. Apply AI to a poorly framed question, and you will get an answer that is confident and useless. Sometimes worse: confident and misleading.

The operational implication is clear. Upskilling HR teams must be methodological before it is technological.

Anticipating the value of quantitative analysis, framing the right questions, identifying the relevant data, choosing the right analytical angle, interpreting results with critical judgment these are not delegable to AI.

 

Conclusion

Data, governance, capabilities these are not prerequisites to AI. They are what AI will expose. AI will not mechanically transform People Analytics. It will reveal its true state at scale and with direct impact on decisions.

The real question is no longer: “Are you ready for AI?”. It is: “Are you ready to face what AI will make visible?”

The data you capture, the governance you enforce and the practices you embed today will determine the value or the risk of your AI use cases tomorrow.

In that context, the first step is often simple: establish an objective view of your true data maturity, and build your AI acceleration roadmap from there.

 

 

Sources :

• RAND Corporation, The Root Causes of Failure for Artificial Intelligence Projects and How They Can Succeed: Avoiding the Anti-Patterns of AI (RRA2680-1), 2024.
• MIT NANDA, The GenAI Divide: State of AI in Business 2025.