AI and people analytics: questions before predictions

Build useful people analysis with clear definitions, reliable data and careful interpretation. This practical guide to AI people analytics brings together a step-by-step approach, illustrative examples, a reusable worksheet and answers to common questions. Start with the section closest to your current challenge, then use the working session to turn the guidance into a clear next action.

Start with a decision and a question

People analytics should help someone make a better decision about work. Begin with the question, the possible action and the evidence needed. “Where are onboarding access delays occurring?” is more actionable than asking a model to find something interesting in all employee data.

Distinguish descriptive analysis from prediction and causal claims. A pattern in historical records does not prove why it occurred or what will happen to an individual. Avoid using a model’s fluent explanation as evidence of a causal relationship.

Establish data meaning and quality

Define each field, population and time period. Check missing values, duplicate records and changes in how data was collected. An apparent change in turnover may reflect a new classification rule rather than a change in the workforce.

Review whether the data is appropriate for the purpose and whether the proposed analysis is lawful and proportionate. Limit access and reporting to authorised users. Sensitive information and small groups need particular care; aggregation alone does not remove every identification risk.

Use AI where results can be verified

An assistant may help draft a query, explain a chart or suggest questions to investigate. Verify calculations and query logic independently. Check joins, filters and denominators before interpreting the result.

If considering a predictive model, involve qualified statistical, legal and privacy reviewers. Examine validity, uncertainty, potential unequal effects and the proposed action. Do not turn an unvalidated “flight risk” score into a judgement about an employee’s loyalty or an automatic employment decision.

Worked example: onboarding delays

Illustrative scenario: HR wants to understand why new starters wait for system access. The analyst defines the start date, request date and confirmed access date, then checks missing records.

An AI assistant drafts a summary of the verified aggregate results. The analyst checks that it distinguishes request delays from provisioning delays and does not blame individuals without evidence. The action is to improve the handoff that causes the delay, then measure the same process again.

Analysis-design worksheet

• Decision: Who will act on the result, and what options do they have?
• Question: What specific uncertainty should the analysis reduce?
• Definitions: Which population, fields and period are included?
• Quality: What missing or inconsistent information affects interpretation?
• Verification: How will calculations and generated explanations be checked?
• Privacy: What access and reporting restrictions apply?
• Limitations: What can the result not establish?

Write the limitations beside the conclusion. If the data does not answer the question, report that directly rather than presenting a confident substitute.

Evaluate the action and keep an audit trail

Record the approved data source, transformation steps, analysis version and reviewer. Retain enough information to reproduce the result under the organisation’s retention rules. Avoid keeping unnecessary raw personal data just because it might be useful later.

Check whether the resulting action improved the original problem. Revisit the analysis if definitions or workforce context change. The NIST AI Risk Management Framework provides a general risk reference; the employee data guide supports the information-handling plan.

Put the guide into practice

Set aside a working session with the people who own this process and one or two people who experience it. Use a fictional or appropriately authorised case, so the discussion can be specific without sharing unnecessary personal information. The purpose is to leave with a usable decision or document, not just agreement that the topic matters.

Prepare the case

Choose a business question with a clear decision owner, such as where onboarding delays occur. Define the population, period and fields needed. Ask what action could follow the result. Avoid beginning with every available employee record and hoping a model finds an interesting pattern worth presenting.

Write the starting assumptions down before discussing solutions. If the group disagrees on what happened, identify the information needed to resolve that difference rather than building a plan on an untested story.

Work through the decision

Check a small sample of records and independently calculate a simple result. Review missing values, denominators and changes in definitions. If AI drafts a query or explanation, compare it with the agreed question. A technically valid query can still answer the wrong question or join records incorrectly.

Ask each participant to explain the proposed decision in their own words. Differences in interpretation often reveal an unclear criterion, a missing responsibility or an instruction that will be difficult to follow.

Test an exception

Introduce a small group, a change in data collection or an apparent correlation. Decide what can be reported and what remains uncertain. Check whether a proposed explanation claims causation without supporting evidence. Do not turn an aggregate pattern into a prediction about a named employee without an appropriately reviewed basis.

Record what changes in this situation and what remains the same. An exception should lead to a clear next step, with an owner, rather than an informal workaround that nobody can explain later.

Agree the handoff

Give the decision maker a conclusion with definitions, limitations and a practical next step. Keep the analysis reproducible through controlled records of sources and transformations. The handoff should make uncertainty understandable rather than hide it behind a polished narrative or a precise-looking number.

Finish by confirming the owner, the next action and the date when the result will be reviewed. Give the person receiving the work enough context to continue without repeating the whole discussion.

Frequently asked questions

What is the difference between reporting and predictive analytics?

Reporting describes information about a defined population and period. Predictive analysis estimates an outcome using a model and assumptions. Neither automatically explains causation. Predictive uses affecting employees need careful validation and specialist review of the proposed action, fairness and privacy. Be explicit about what the analysis can establish and avoid presenting a model estimate as a known fact about a person.

Can AI write SQL or spreadsheet formulas for HR analysis?

It may draft them, but a competent reviewer should check the logic, filters, joins and calculations before use. Test on approved sample data and compare results with an independent calculation. Ensure the tool and information are approved. A formula that runs without an error can still produce a misleading answer if it uses the wrong denominator or omits relevant records.

Which people analytics project is a sensible starting point?

Choose a well-defined operational problem with reliable data and a proportionate action, such as locating delays in an administrative process. Establish the baseline and decision owner before analysing. Avoid starting with opaque individual predictions simply because they sound advanced. A clear descriptive analysis that leads to a useful process change may provide more value than a complex model nobody can explain.

Review the first cycle

Review whether the action improved the original process and whether the analysis could be reproduced. Check that the reported population and definitions stayed consistent. If the conclusion depends on missing or unreliable data, improve collection before making a stronger claim. Keep sensitive detail out of reports whose audience does not need it.

Keep a brief record of what was tried, what participants found useful and what needs to change. Compare the result with the original problem rather than judging success only by completion. If the process created extra work without improving clarity, quality or support, simplify it and test again. Share the agreed change with the people who will use it, and name the person responsible for keeping the guidance current.

About New Dynamics

New Dynamics connects goals, feedback, recognition and reviews around the way organisations work. This guide is published by the New Dynamics Editorial Team as part of our practical library for HR leaders, managers and People teams.

Use the examples and worksheets to structure your own discussions and adapt them to your organisation. Illustrative scenarios are not customer case studies. Policy and employment guidance needs appropriate local review before adoption.

For questions about this guide, corrections or a conversation about your performance management process, email contact@new-dynamics.com. Explore the complete guide library for related resources.