flowchart LR A[Descriptive<br/>what happened] --> B[Predictive<br/>what is likely] B --> C[Prescriptive<br/>what to do] style A fill:#E8F0FE,stroke:#1A73E8 style B fill:#FEF7E0,stroke:#F9AB00 style C fill:#E6F4EA,stroke:#137333
9 Overview of HR Analytics
9.1 Why HR Analytics Has Earned a Chapter of Its Own
A metric is a number that describes the past. An analytic is a question put to that number that changes what the organisation does next.
The previous module showed how to build a credible HR-metrics programme: choose the right metrics, organise them by efficiency, effectiveness, and impact, render them well on a page. That programme is necessary but not sufficient. A scorecard, no matter how well designed, describes the workforce as it is. It does not, on its own, tell you what to do about it. The leap from describing the workforce to deciding what to do about it is the leap that HR analytics makes.
The term itself is younger than its practice. Firms have been doing rough HR analytics — paired comparisons, regression on attrition, scenario forecasts — for decades. What changed, as Ben Waber (2013) documents in his work on people analytics, is the arrival of cheap storage, fast compute, and rich behavioural data that lets the analyst ask questions earlier generations could not have computed in time to act on. The discipline matured into something the executive committee could rely on, and the language settled on “HR analytics” or “people analytics” interchangeably.
Maturity has not made the discipline immune to faddishness. As Thomas Rasmussen & Dave Ulrich (2015) warn in their study of HR analytics in practice, the field has produced as many failed projects as celebrated ones, usually because the team built the model before they framed the question, or built the dashboard before they secured the decision owner. This module is organised around the discipline that prevents those failures. It begins with the meaning of HR analytics, the importance of the function, and the concrete operational link between metrics and decisions, before moving into frameworks and methods.
The visualisation lens runs through the module exactly as it ran through the previous one. An analytic that cannot be drawn in a way that the audience reads at a glance is an analytic that will not be acted on. The chart is not the analytic, but the chart is what the analytic produces, and the chart is what the audience remembers a week later.
TipThe HR-analytics overview contract
- Every HR analytic starts with a defensible business question and ends with a chart the audience can act on, not with a model output the audience has to interpret.
- The level of analytics chosen — descriptive, predictive, prescriptive — is matched to the question, not chosen for its sophistication.
- Every analytic earns a recurring decision moment: a recurring meeting, a triggered alert, a review cycle. Without the decision moment, the analytic is a one-off study, not a programme.
9.2 The Three Tiers of HR Analytics
Most useful definitions of HR analytics organise the discipline by three tiers: descriptive, predictive, and prescriptive. The tiers are nested. A predictive analytic depends on descriptive ground truth, and a prescriptive analytic depends on a predictive model that has earned trust. A function that promises prescriptive output without disciplined description underneath it is selling a structure that will not stand up.
TipThe Three Tiers of HR Analytics
| Tier | Question it answers | Example | Typical visual |
|---|---|---|---|
| Descriptive | What happened in the workforce | Attrition by business unit and tenure cohort | Trend lines, cohort charts, distributions |
| Predictive | What is likely to happen next | Probability of attrition for each employee in the next twelve months | Risk-distribution chart with model-confidence band |
| Prescriptive | What action should the organisation take | Recommended retention investment per high-risk segment | Scenario-comparison page with cost-and-outcome overlay |
TipThe tier ladder
Read the ladder upward and the questions become richer; read it downward and the data and trust requirements grow. A function that runs descriptive analytics well, with reliable definitions and a refreshed dashboard, has built the foundation a predictive layer can stand on. A function that has not done that work but tries to publish predictions has built the second floor on a missing first floor, and the building falls when the first model misses.
9.3 The Building Blocks of an HR Analytics Capability
Behind every HR analytics output sits a small number of building blocks. They are not glamorous, but they are what separate a function that publishes one defensible analytic each quarter from one that publishes a dozen and defends none. Each block has to be present and stable, and each block can be the place where the entire capability fails.
TipFive Building Blocks of an HR Analytics Capability
| Block | What it provides | What goes wrong when it is missing |
|---|---|---|
| Question framing | A defensible business question with a named decision owner | Models are built that nobody acts on |
| Data foundation | Clean, definition-locked data from HRIS and adjacent systems | Disputes about whose number is right derail every review |
| Methods literacy | Statistical and causal methods appropriate to the question | The team over-reaches with techniques it cannot defend |
| Visualisation craft | Charts, dashboards, and stories that the audience can read at a glance | Strong analyses are buried in slides nobody reads |
| Decision plumbing | The recurring meetings, alerts, and review cycles that turn output into action | Insights are produced and then forgotten |
TipThe order in which the blocks have to be built
The temptation is to start with methods literacy, because that is the part that looks like analytics. The order that actually works runs the other way. Question framing comes first, because without a question the rest is wasted. Decision plumbing comes second, because without a decision moment the output has nowhere to land. Data foundation and visualisation craft follow, because they are what makes the answer to the question reliable and readable. Methods literacy is the last block, because the technique should be chosen for the question, not the other way around.
9.4 Outputs: From Insight to Decision
The output of an HR analytic is not a number. It is a decision the organisation now takes differently. The discipline that converts a model run into a decision change has its own design pattern. Three forms of output appear most often: the recurring dashboard, the triggered alert, and the one-off study. Each has a different rhythm, a different audience, and a different visualisation.
TipThree Forms of Analytics Output
| Output form | When it fits | Cadence | Typical visual |
|---|---|---|---|
| Recurring dashboard | The decision is made on a regular cadence by the same audience | Daily, weekly, or quarterly | KPI page with model-driven tiles |
| Triggered alert | The decision is taken when a specific signal crosses a threshold | Event-driven | Status card with drill-through |
| One-off study | The decision is taken once, with high stakes, on a specific question | Ad hoc | Narrative report with supporting visuals |
TipChoosing the form
Most analytics failures begin with the wrong output form. A recurring decision served by a one-off study is forgotten by the next cycle. A one-off decision served by a recurring dashboard buries the unique question in a sea of routine pages. A signal worth alerting on lost inside a dashboard is a signal that fires after the moment for action has passed. Match the output form to the rhythm of the decision before building anything.
9.5 Visualising HR Analytics Outputs
The same analytic, drawn well, can change the conversation in the executive committee. Drawn badly, it changes nothing. Five design choices distinguish an HR analytics visualisation from a generic dashboard tile, and applying them consistently is what makes the analytics function legible to its audiences.
TipFive Visualisation Choices Specific to HR Analytics
| Choice | What it does on the page |
|---|---|
| Tier label | Every chart names whether it is descriptive, predictive, or prescriptive |
| Confidence band | Predictive charts show their uncertainty visually, not in a footnote |
| Comparison built in | Every chart shows benchmark, target, control group, or counterfactual |
| Decision call to action | The chart titles a decision, not a metric |
| Drill-through to evidence | The audience can click from output to the data and method behind it |
Summary
| Concept | Description |
|---|---|
| Why HR Analytics | |
| Metric versus analytic | Metrics describe; analytics interrogate and produce a decision change |
| From description to decision | An analytic that does not change a decision is a study, not a programme |
| Recurring decision moment | Every analytic earns a recurring meeting, alert, or review cycle |
| Question first, model second | Frame the business question before choosing the technique |
| Avoiding the management fad | Discipline that distinguishes lasting analytics practice from a passing trend |
| The Three Tiers | |
| Descriptive tier | What happened in the workforce, drawn as trends, cohorts, and distributions |
| Predictive tier | What is likely to happen next, drawn as risk distributions with confidence |
| Prescriptive tier | What action the organisation should take, drawn as scenarios and trade-offs |
| Tier ladder | Predictive depends on descriptive; prescriptive depends on trusted prediction |
| The Building Blocks | |
| Question framing | A defensible business question with a named decision owner |
| Data foundation | Clean, definition-locked data from HRIS and adjacent systems |
| Methods literacy | Statistical and causal methods appropriate to the question |
| Visualisation craft | Charts and stories the audience can read in seconds |
| Decision plumbing | The recurring meetings and review cycles that turn output into action |
| Order of building blocks | Question framing and decision plumbing come before methods literacy |
| Three Forms of Output | |
| Recurring dashboard | The decision is made on a regular cadence by the same audience |
| Triggered alert | The decision is taken when a specific signal crosses a threshold |
| One-off study | The decision is taken once, on a high-stakes specific question |
| Output-form match | Match output form to the rhythm of the decision before building anything |
| Threshold for an alert | The signal level above which the alert fires and a decision is owed |
| Narrative as a study output | One-off studies surface their findings in a written narrative with supporting visuals |
| Visualisation Choices | |
| Tier label on the chart | Charts name whether they are descriptive, predictive, or prescriptive |
| Confidence band | Predictive charts render uncertainty visually rather than in a footnote |
| Comparison built into the visual | Every chart shows a benchmark, target, control group, or counterfactual |
| Decision call to action | Chart titles describe the decision, not the metric name |
| Drill-through to evidence | The audience can click through from the output to the data and method |
| Behavioural-data visualisation | Charts that surface communication, collaboration, and movement signals |