People Analytics Starter

Use data on a people decision the way the field works at its best: start from the decision, pick few measures, test a change small before rolling it out, and keep hard ethical lines. Built for a team lead or a small business, not an analytics department.

People analytics is practice, not research; the founding articulation is a practitioner article (Davenport, Harris & Shapiro, 2010). The research behind specific people decisions is in the templates for those decisions: selection validity (4.16), engagement (4.20), appraisal and calibration (4.8). Thresholds on this page are rules of thumb.

Use it when:

• You keep making the same people decision on gut feel: who to hire, what onboarding contains, why people leave.
• You want to change something (an interview step, a meeting rhythm, an onboarding block) and want to know whether it helped.
• Someone asks for a people dashboard, and you want it to serve a decision rather than sit unread.

Skip it when:

• You cannot yet name the decision or the problem: start with 4.13 Team Health Scan.
• The question is about AI tools rather than people decisions: use 10.12 AI Benefit Tracker.

People analytics means running people decisions like other evidence-based decisions: write down what you believe, pick a few measures, test small and compare honestly. The field's founding articulation is a practitioner article (Davenport, Harris & Shapiro, 2010), and the method here is working craft, not research. The research sits one level down, in the specific decisions the method serves, such as selection or engagement, which this module sources in their own templates. The craft logic still has teeth. Starting from a decision rather than from available data avoids the commonest failure, dashboards nobody acts on. Pairing each outcome measure with a quality or experience measure stops you from optimizing what is easy to count. Writing the decision rule before the test is what separates analytics from reporting. The limit: at team scale the numbers are small and swing, so you are reducing uncertainty, not proving anything. And some things should never be measured, which is why the ethics gate comes before any collection.

Analytics that starts from available data produces dashboards; analytics that starts from a decision changes something. Write the decision first, then the belief you hold today, so it can be wrong.

Strong answer: "Decision: what the first 90 days contain. Belief: new starters leave because pay is below market. Existing evidence: exit notes mention a chaotic start more often than pay. If we knew: we would change onboarding, or argue for a pay adjustment."

Illustrative example: the hotel, roles and numbers are made up.

The head of housekeeping at a hotel in Bergen keeps losing new room attendants around month three.

Section 1: decision: what the first 90 days contain. Belief: "they leave because pay is below market." Exit notes mention a chaotic start more often than pay.

Section 2: 90-day retention over the last two years as baseline, time to cleaning a floor alone (supervisor's log), and a two-question pulse at day 30, shown only for the whole team.

Section 3: the next two starters get a structured first month with a named buddy and a fixed floor in week one; the two after them get the current start. Rule: adopt if both measures are better for both structured starters. Misleading if: the structured pair starts in the quiet season.

Section 4: all six lines ticked; the team hears about the pulse before it runs.

Section 5: after two rounds, the structured starters work alone sooner and rate day 30 higher; pay came up once. Structured onboarding becomes standard, and the pay belief, which would have cost a budget fight, is quietly retired.