People analytics dashboards: choose metrics that answer a decision

What should a people analytics dashboard show?

Show the measures that help a named audience make a specific decision. An HR service manager reviewing unresolved requests needs different information from a leader deciding whether to invest in onboarding. Put the decision, metric definition and source beside the chart before adding more visualisation.

This guide uses a fictional HR service example. It proposes a planning method rather than a universal set of benchmarks. Your measures should reflect the service you operate and the information you can use appropriately.

Start with one decision

Suppose the team needs to decide whether to change how onboarding requests are routed. A total count of tickets tells you how busy the queue is, but not whether routing is the problem. You might also examine reassignments, time waiting for missing information and cases reopened after being marked complete.

Discuss what each measure can and cannot establish. A long case duration could reflect a difficult request, a delayed response or a deliberately early submission. It does not automatically identify poor performance by the person handling it.

Write a metric definition before building the chart

Use the following brief for every measure:

• Decision: what action might change after reviewing the result?
• Definition: exactly what is counted or calculated?
• Population: which people or cases are included and excluded?
• Time period: does the measure use opened, closed or active cases?
• Source and owner: where does the information come from, and who checks it?
• Known limitations: what is incomplete, inconsistent or open to interpretation?
• Review action: who investigates a change and how?

For “onboarding readiness”, define the required tasks and the agreed readiness date. Include the total number of eligible starters alongside the percentage. A small population can produce a large percentage change from only one case.

Check the records behind the measure

Inspect a sample before drawing conclusions. Look for duplicate cases, missing completion times, inconsistent categories and records that changed system halfway through the period. Keep a note of corrections rather than silently revising historical numbers.

If the data cannot support the question, change the question or improve collection. AI-generated explanations cannot repair an undefined metric. The HR automation checklist helps identify where reliable events should be recorded in the workflow.

Use AI as an assistant to interpretation

An assistant can draft a plain-language description of a supplied table or suggest questions to investigate. Check every numerical statement against the table and distinguish observed patterns from possible explanations. Ask it to state what the data does not show.

For example, a rise in reopened cases after a workflow change is a reason to investigate. It does not prove that the change caused the increase. Compare case types and inspect examples before deciding what to do next.

Limit access to the information the audience needs. Consider whether a filtered chart could identify individuals in a small group. Agree appropriate aggregation and access controls with your information owners before publishing the dashboard.

Make the review part of the service

Give the dashboard a review cadence and an action log. Record the question raised, evidence checked, decision taken and the next review date. Retire measures that nobody uses to make a decision; adding charts indefinitely makes the important signals harder to find.

Develop the approach in People Analytics for Better HR Decisions, or address the foundations with HR Data Quality & Connected Records. Talk to Experrt about a practical programme using synthetic examples from your type of service.