Why People Analytics and AI Fail Without Data Foundations

Ian O'Keefe has spent 25 years building people analytics functions inside some of the most data-sophisticated companies in the world: Amazon, JPMorgan Chase, Google, and American Express. At Amazon, he led a multidisciplinary team whose data products supported more than 400,000 corporate employees. In 2024 he founded Ikona Analytics to help HR organizations get ready for the AI era.

That vantage point gives him an unusually clear answer to a question most HR leaders are asking right now: why does workforce analytics succeed in some companies and quietly fail in others? His answer has almost nothing to do with tools.

The biggest foundational mistake: starting with tools, not decisions

Asked for the single biggest mistake organizations make before they even start measuring, Ian does not hesitate: starting with data, tools, and outputs instead of the decisions those outputs are supposed to serve.

Rushing to stand up reports and dashboards is the classic version of the trap. When a leader says the report is wrong, fixing the report is not the issue. Everything upstream of it is: production systems, workflows, the architecture, the point at which transactions are generated by HR teammates and managers. Working only at the point where data is consumed, without acknowledging that ecosystem, is where credibility starts to leak.

His grounding question for any new study, model, or product is disarmingly simple: what would you do differently if you knew the answer? If the leader cannot name a process, policy, or approach they would change, the request needs reframing before it needs building.

Rebuilding a broken data foundation

At JPMorgan Chase, Ian inherited fragmented HR data across lines of business and functions, which he notes is the common starting point for most organizations. The first move is restraint: resist the urge to fix everything in all directions at once, and do not confuse reorganization with a solve for fragmentation.

The second move is to treat data as an asset with owners, all the way back to where it originates. Recruiting, compensation, talent, learning, operations: everyone who hits submit or controls a platform is a data generator. If that data does not land somewhere common, the gaps express themselves downstream as a foundation that is cracked and broken.

From there, the sequence is the one he repeats throughout the conversation: identify the key decisions leadership needs to make, define the data and metrics those decisions require, then work backwards into the ecosystem to shore up the most critical workflows with unglamorous fundamentals like data definitions and quality scanning routines.

Reporting is the what. Analytics is the why

Most HR leaders conflate reporting with analytics, and Ian argues the confusion quietly kills the credibility of the entire function. Reporting looks back at what happened; it is an extension of your tools, processes, and transactions. Analytics asks why something happened and what will probably happen next: hypotheses, models, experiments, research.

Even basic reporting is hard: counting humans consistently across an ATS, payroll, talent systems, contingent workforces, and a finance function that counts people differently than HR does. That difficulty lives in governance, definitions, and core platforms.

The trap comes when the ecosystem is less understood than leadership believes. Talented, high-priced researchers and data scientists get pulled into reporting reconciliation quagmires, which is not a good use of their skill set, and the analytics agenda stalls before it starts.

Scaling without losing integrity: lessons from Amazon

Supporting 400,000 corporate employees taught Ian that technology platforms and the governance mechanisms on top of them are part and parcel. At that scale, a central team cannot inspect everything, so the answer is accountability loops: data engineering routines that sweep for problems, plus practitioners embedded in the business lines with publish rights and access paths to inspect, root-cause, and remediate data close to where it is generated.

The balance to strike is matching governance, security permissions, and architecture access patterns to the humans actually doing analytical work across the organization, not just the central team.

HR's internet moment

On AI, Ian is direct about the scale of the shift: this is an internet moment, probably bigger than the one from 1999. The question companies asked then, how does this new technology redefine the way things get done, is exactly the question HR should be asking now.

His starting point is unexpected: most of how work actually gets done in HR is not encoded anywhere. It is tribal knowledge, tacit and sitting in the minds of people in the seat. The clues are the manual, out-of-product workflows every team runs: someone downloads from system A, emails it to a colleague, who sends it on. Those loops are signals of human effort doubling for what technology could do.

That is where AI genuinely helps: augmenting and accelerating the drudgery, consolidating information, spotting patterns, sending signals. He is measured on the jobs question. There is evidence of roles being displaced, and plenty of evidence that AI replaces tasks rather than whole humans. Understanding the tasks and the context around them is what determines whether an AI investment returns anything at all.

The missing glue: turning tacit knowledge into a data asset

Ian's diagnostic method at Ikona is deliberately human. Instead of leading with questions about data definitions and integrity, he asks people to talk about where they spend their time, which tools they use, and where the data they need shows up badly. Ten minutes of storytelling will surface a dozen data capabilities, history, and context that no system captures and no AI can access.

The next level for people analytics teams, he argues, is turning what might be dismissed as water-cooler talk into a data asset: capturing the spoken experience, digitizing it, and making it readable by machines so it can improve a workflow or guide an AI contextually. Change management is underestimated for exactly this reason. People analytics is less about the analytics than about the people who generate the data in the first place.

The 90-day non-negotiable

Before investing another dollar in analytics or AI workforce tools, Ian's non-negotiable for every CHRO is governance, ownership, and understanding of the data ecosystem. Do not expect AI to fix it. Every technology wave has carried the same temptation, from won't the internet fix this, to won't big data fix this, to let's build a machine learning model for that.

AI raises the stakes because it fails persuasively. Take a hard look at data foundations, governance, and ownership across the whole HR data ecosystem first, then pick small, high-value AI use cases that accelerate decisions from a solid foundation. That loop, decisions to data to foundation and back, is the whole playbook.

Where employee listening fits

Asked how real-time listening platforms strengthen the people data foundation rather than becoming another disconnected source, Ian points to the part of listening that analytics teams used to dread: open-ended comments. The unstructured spoken word, he argues, is where the truth lies more than in a quantified value.

What was out of reach until recently is now table stakes: capturing the shared experience of people and teams intentionally, digitizing and synthesizing it, and putting new kinds of metrics and pattern-spotting on top. Survey providers not moving in that direction ought to be. Platforms that connect regular pulse listening with employee sentiment analysis give organizations a way to transform, not just understand, and feed the people science foundation everything else builds on.