Why AI on top of flawed self-report data produces faster bad answers, not better ones
By Jonathan Hawkins
Open any HCM vendor’s website right now and you’ll see the same message: “We’ve added AI.” Workday has AI. UKG has AI. SAP SuccessFactors has AI. Oracle HCM has AI. Every major platform has spent the last eighteen months racing to add machine learning, natural language processing, and generative capabilities to their people analytics stack.
The marketing is consistent. The fundamental problem is unchanged.
AI on Bad Data Produces Confident Bad Answers
The data foundation underneath most HCM analytics hasn’t changed. It’s still engagement surveys, pulse checks, self-reported sentiment, and periodic performance reviews. Adding AI to this layer doesn’t fix the structural limitations of self-report. It accelerates them.
A machine learning model trained on survey data inherits every bias in that data: social desirability, recency, response fatigue, and the fundamental gap between what people say and what they do. The AI just processes those biases faster and presents them in a more polished dashboard.
The Missing Interpretive Layer
What’s missing from the HCM stack isn’t intelligence. It’s interpretation. The operational data that actually predicts behaviour — scheduling patterns, attendance records, productivity shifts, workforce management signals — sits in these platforms already. But nobody is reading it for what it reveals about human sentiment and intent.
That’s the gap Anthrolytics fills. We’re not competing with HCM platforms. We’re the interpretive layer they’re missing. The same data they use for scheduling and compliance, we use for prediction.
Why This Matters for Buyers
If you’re a CHRO evaluating AI-powered people analytics from your existing HCM vendor, ask one question: what data is the AI trained on? If the answer is survey data, sentiment scores, or self-report, the AI hasn’t solved the prediction problem. It’s just dressed up the measurement problem.
Prediction requires objective, behavioural, continuous data. It requires signals that employees generate by doing their jobs, not by completing a questionnaire. And it requires a model purpose-built to connect those signals to future outcomes — not a general-purpose AI bolted onto an existing analytics module.
The HCM platforms have the data. They’re building the AI. What they haven’t built is the layer that connects the two in a way that actually predicts what people will do next. That’s what we built.