The core idea
There is no single universally accepted people analytics maturity model. Our capability map distinguishes operational reporting, descriptive, diagnostic, predictive and prescriptive work, with impact evaluation throughout. Reporting establishes dependable totals; description identifies patterns; diagnosis investigates explanations; prediction estimates future outcomes; prescription compares possible actions. Evaluation asks what changed because of an intervention. These capabilities overlap. A well-designed comparison of two onboarding approaches can be more useful than an elaborate individual flight-risk score.
Source and attribution [1]Using it in practice
Start with a recurring decision and assess whether the current data can support it. Check definitions, completeness, ownership and how the result reaches a decision-maker. Improve one weak link before buying more complex technology. Build evaluation into the plan from the beginning: a team should not wait until it reaches a supposed final maturity stage to ask whether its work helped. Include qualitative evidence and employees’ explanations of their experience.
An example, not a reported case
Worked example · illustrative
A fictional employer initially reports monthly leavers. It then separates voluntary departures from contract endings and compares eligible new-starter cohorts after 90 days. Interviews suggest that unpredictable shift availability deserves investigation. The team pilots clearer scheduling communication and compares subsequent retention with a suitable reference group. It has moved towards a better decision without needing to predict which named employee will leave.
What to watch for
Do not present this six-part teaching route as an industry-certified scale or attribute it to a single author. More data, dashboards or machine learning do not automatically mean greater capability. An organisation may be strong in one analytical area and weak in another. Governance, trust, data quality and the ability to act matter at every stage.
Reporting and description
Create a metric dictionary before debating a trend. Define who counts as an employee, how casual-worker inactivity is treated and which date records a departure. A dashboard should show the relevant population and period. Reconcile totals with source systems and explain corrections rather than quietly changing historical values.
Diagnosis and prediction
A diagnostic question seeks plausible reasons and evidence that distinguishes them. A predictive question asks how well an outcome can be anticipated for new cases. The most predictive variable may not be a useful intervention target. Keep those purposes separate when discussing what an analysis can support.
Prescription and impact
A recommendation depends on costs, feasibility, fairness and evidence about what an action changes. Prediction alone does not establish that offering a particular intervention will help the predicted group. Plan a proportionate evaluation and define an outcome that matters beyond whether staff opened a dashboard.
A practical development plan
Choose one business question, one data-quality improvement, one skill to develop and one decision owner. Set a date to review what changed. Record unresolved limitations openly. This creates an achievable capability plan without making advanced modelling a status symbol.
Assess capability through a completed decision
Choose one recent analysis and trace it from question to action. Were definitions agreed, data checked, uncertainty explained and an owner able to respond? Did anyone later assess the result? This reveals capability more clearly than counting dashboards. A team with simple tools may perform this chain well, while a sophisticated platform may produce reports that nobody trusts or uses. Identify the weakest practical link and make it the next improvement priority.
Establish a minimum dependable reporting service
Create a small metric dictionary, named data owners, a reconciliation routine and a correction log. Explain what each report is for and when it is refreshed. If two systems disagree on headcount, resolve the boundary and timing before developing a predictive model. Keep the service small enough to maintain. Reliability includes telling users when a figure is incomplete or provisional, rather than allowing visual polish to imply a level of certainty the data does not support.
Build skills around an actual question
For an original learning exercise, investigate whether early exits are concentrated in a particular joining cohort. An analyst can learn cohort construction, a manager can contribute operational context and HR can organise appropriate employee feedback. Agree the decision the work might inform. Review the analysis together, including alternative explanations. This develops analytical capability and organisational understanding at the same time, rather than treating a technical training course as evidence that the whole organisation has become mature.
Decide what not to build
Before commissioning a model, ask whether a simpler process change or descriptive analysis would answer the question. Consider maintenance, governance and the consequences of error. A model that cannot lead to a fair, feasible action may add cost without value. Record why an approach was chosen and what would justify greater complexity later. Maturity includes restraint: being able to explain why an attractive analytical technique is unnecessary for the decision at hand.
Assess the capability needed for the next decision
Do not assume predictive modelling is the next step simply because it appears above reporting in a diagram. A reliable definition of starters and leavers may create more value than an untrusted turnover score. Examine data quality, analytical skill, stakeholder understanding, governance and the capacity to act. An organisation can be strong in one area and weak in another, so a single maturity label can conceal the development work that matters.
Choose a bounded capability improvement
Khan and Millner connect analytical development with real organisational questions. Select one recurring decision, agree its data definitions and produce a repeatable analysis with a named owner. For example, build a reliable early-tenure retention view before attempting individual prediction. Evaluate whether the information changes a decision and whether users understand its limits. Advanced technique is not the outcome: useful, proportionate and responsible action is. Maintain simple reporting where it serves a continuing purpose rather than treating it as a stage to abandon.
Take it into your next conversation
Three useful questions.
- Which decision will this analysis change?
- Are the population, denominator, time period and assumptions explicit?
- What alternative explanation or unintended effect must we check?
Related terms
Go to the evidence
Sources & attribution
Further reading: Nadeem Khan and Dave Millner (2023), Introduction to People Analytics: A practical guide to data-driven HR, 2nd edition, PDF pp. 227, 244 and 294–295. Edition & reading notes ↗
The core idea is an original summary of the cited work. Application notes, examples and sketches are our interpretations, not quotations or reproductions of the authors’ figures. Publisher records may require access to read the full original work.
Application guidance expanded on 26 September 2026. These practical suggestions and invented examples are editorial explanations; further-reading sources are not the original authors of the historical models unless expressly identified. Original overview review dates are retained.
Published 2026-09-20 · Reviewed 2026-09-22. Editorial approach