Short answer
People analytics moves beyond dashboards when a metric starts a disciplined decision process. Write the question the chart cannot answer, collect the smallest safe set of relevant context, assign a human reviewer, choose one observable action, and set a return date. Keep measurement, explanation, and judgement distinct.
A dashboard is evidence, not the decision
A people dashboard can show headcount, absence, movement, representation, hiring, learning, or engagement patterns across time and groups. That visibility matters. The mistake is asking the chart to do a different job: explain a cause, represent every employee's experience, or select an intervention.
The CIPD people analytics factsheet defines the field around analysing people data to solve business problems and distinguishes concepts including correlation and causation. That distinction should shape the operating workflow. If absence and a manager change occur together, you have an observed association. You do not yet know that the manager change caused the absence pattern.
Moving beyond the dashboard therefore does not mean replacing it. It means connecting the metric to a question, additional evidence, an accountable decision, and a later review.
The metric-to-decision worksheet
Use one row for one decision. If the row contains several business problems, split it.
| Field | What to write | Guardrail |
|---|---|---|
| Metric | The measure, population, period, and comparison | Do not hide the denominator or response base |
| Observed pattern | What changed in neutral language | Describe association, not cause |
| Unanswered question | The smallest question needed for action | Avoid “Why is everything worse?” |
| Qualitative collection | Who can supply relevant examples and how | Collect only what the decision requires |
| Human reviewer | Named role responsible for interpretation | AI output cannot own the judgement |
| Possible action | One bounded change within an owner's control | Do not treat a theme as an instruction |
| Return date | When evidence and the metric will be reviewed | Define what would change the decision |
The worksheet prevents two common failures. First, the team debates a red number without deciding what evidence would resolve the debate. Second, it collects a large volume of comments with no owner, purpose, or return date.
A fictional absence-pattern example
Imagine a US customer operations group. Its monthly dashboard shows a higher absence rate for one regional team than for comparable teams. The increase appears during the same period as a scheduling change and a new manager appointment.
The dashboard establishes the pattern and timing. It does not establish that the schedule or manager caused it. Seasonal illness, workload, recording practice, local staffing, or other conditions may contribute.
The People Analytics lead completes the worksheet:
- Metric: recorded absence for the regional team, compared with its prior period and similar teams;
- Observed pattern: absence increased after two workplace changes occurred;
- Unanswered question: which work situations, if any, are making attendance harder?
- Qualitative collection: short voluntary conversations with a suitable cross-section of the team, plus the manager's account of scheduling and recording changes;
- Human reviewer: the HR business partner, supported by the analyst and the relevant Operations leader;
- Possible action: test a clearer shift-change process if the evidence supports that explanation;
- Return date: the next agreed operating review, when participation, examples, and the absence pattern are examined together.
Suppose several employees describe late schedule changes, while others describe an equipment problem that extends shifts. Those are competing explanations, not proof. Operations can verify the rota records and equipment incidents before deciding whether either action fits.
Separate group patterns from individual judgement
A group metric describes a group under a defined method. It does not describe every person in the group. A theme found in several conversations also does not tell you what a silent participant thinks.
Use these rules when moving from analysis to action:
- Keep the population and response base visible.
- Do not attach a group trend to a named employee.
- Distinguish reported experience from verified operational records.
- Record contradictions and missing voices, not only the dominant theme.
- Assign individual employment decisions to authorised people using appropriate evidence and process.
- Let employees correct relevant context where the workflow allows it.
These rules matter even when no AI is involved. AI can make weak inference faster, so buyers should be especially clear about scope and review.
Where AI can help safely
AI can assist with bounded tasks such as organising responses into provisional themes, finding repeated examples, drafting questions for a reviewer, or tracking whether an agreed action has a follow-up. The output should keep source, date, uncertainty, and access conditions visible.
The NIST AI Risk Management Framework core calls for context-specific measurement, documented human roles, and ongoing evaluation. Applied to people analytics, that means testing the system in the decision where it will be used and naming who reviews the output before action.
AI should not convert a group association into a claim about cause, infer an individual's state from a team score, or make an employment decision. A fluent summary still needs evidence and accountable review.
Design the smallest useful qualitative collection
Do not open a broad listening campaign every time a metric moves. Start with the decision and ask what additional evidence would actually change it.
A proportionate collection plan answers:
- Which population experienced the pattern?
- Whose examples are relevant, including less-heard shifts or locations?
- Which question is close enough to the work to produce usable detail?
- Who can view individual material and who receives only an aggregate?
- What route handles an urgent or sensitive issue?
- When will collection stop?
- How will participants hear what happened?
For one recurring team problem, explore the fictional team diagnostic. It shows how employee descriptions, competing explanations, and one action to test can fit together. Lontra can support that conversation preparation and human review. It does not replace the dashboard, establish causality, or decide what the organisation should do.
The Lontra trial covers one campaign with up to 30 invitations for 60 days, with no card required. Use it for one bounded question and a defined return date. Studio is a subscription module and is not included as free standalone access.
Review the action, not just the metric
At the return date, ask whether the evidence supported the original explanation, whether the action happened as intended, and what else changed. The metric may improve, worsen, or stay flat for reasons beyond the intervention. Record that uncertainty.
A strong people analytics practice creates a chain that leaders can inspect: metric, question, evidence, interpretation, decision, action, and review. Dashboards remain essential because they keep the measure consistent. The work beyond them makes the measure useful without pretending it says more than it does.