25 Performance Analytics: Predicting Employee Performance

25.1 Why Performance Analytics Matters

Performance is the outcome that every other HR programme is trying to influence, and the one most often measured by means that the workforce no longer trusts.

Performance analytics sits at the centre of every workforce decision the firm makes. Selection systems are validated against performance. Training programmes are evaluated against performance. Pay structures are calibrated against performance. Yet the measurement of performance itself, in most organisations, has been carried out for decades through annual ratings whose reliability is modest, whose validity is contested, and whose acceptance by the workforce has steadily eroded. Performance analytics is the discipline that improves both the measurement and the use of performance evidence, and that renders both visibly to the audiences that depend on them.

The history of the field is one of accumulating evidence about what works and what does not. As Angelo S. DeNisi & Kevin R. Murphy (2017) set out in their centennial review of performance appraisal and performance management, the literature has produced consistent findings on rating-scale design, on rater training, on the limits of pure ranking systems, and on the role of feedback in driving improvement. The field’s challenge is no longer the absence of evidence; it is the gap between the evidence and the practices most firms continue to use. The analytics function is one of the few intervention points where that gap can be narrowed visibly, by surfacing rating quality, distribution patterns, and prediction calibration in ways the executive team cannot ignore.

The pivot toward performance management as a continuous discipline rather than as an annual event has reshaped what analytics is asked to do. As Herman Aguinis (2013) argues across his work on performance management, the most credible programmes treat performance as a system that runs continuously — goals, ongoing feedback, calibration, development, and reward — and the analytics function supports each part of the system rather than focussing only on the rating output. Performance prediction, where the function forecasts which employees will perform well in current or future roles, is one piece of that system; ratings analytics, distributional fairness analysis, and feedback-loop measurement are others.

The visualisation lens carries the discipline. A rating distribution is rendered with its target curve and forced or unforced choice surfaced. A calibration analysis is rendered as a manager-by-manager comparison with confidence intervals. A predictive-performance model is rendered with its calibration plot and its subgroup robustness panels. The page that surfaces all three lets the audience read performance as evidence rather than as opinion.

The performance-analytics contract

Performance is measured as a system, not as an annual event. The dashboard surfaces goal-setting, feedback, calibration, and outcome data on the same page.
Every rating chart is accompanied by a distribution view, a calibration view, and a fairness view. A single average rating is reporting, not analytics.
Predictive-performance models are evaluated on calibration, discrimination, and subgroup robustness, and the model’s recommended use is constrained to the inferences the evidence supports.

25.2 Performance as a System

Performance management, taken seriously, is not a rating. It is a system with five working parts: goals, ongoing feedback, calibration, development, and reward. Each part has measurable outcomes, and a function that surfaces only the rating output is reporting on the easiest part to measure rather than on the part that most determines whether the system actually moves performance.

The Five Parts of a Performance-Management System

Part	What it does	Example metric	Visualisation
Goal-setting	Translates strategy into individual objectives	Goal-cascade coverage, line-of-sight survey	Cascade map, goal-coverage heat map
Ongoing feedback	Provides feedback throughout the cycle	Feedback frequency, manager-conversation completion	Activity heat map, frequency distribution
Calibration	Aligns ratings across managers and units	Calibration-meeting attendance, rating shift after calibration	Pre-and-post rating distribution
Development	Acts on identified development needs	Development-plan completion, capability uplift	Cohort capability chart
Reward	Translates performance into pay and recognition	Pay-for-performance ratio, top-performer retention	Distribution of merit pay by rating

The arc of the performance-management system

flowchart LR
  A[Goals<br/>cascaded from strategy] --> B[Ongoing Feedback<br/>throughout the cycle]
  B --> C[Calibration<br/>across managers and units]
  C --> D[Development<br/>action on capability gaps]
  D --> E[Reward<br/>pay and recognition]
  E --> A
  style A fill:#E8F0FE,stroke:#1A73E8
  style C fill:#E6F4EA,stroke:#137333
  style E fill:#F3E8FD,stroke:#8430CE

The arc closes when the reward cycle feeds back into the next cycle’s goal-setting and informs strategic adjustments. A function that runs three of the five parts strongly and two of them weakly will see the weak parts dominate the audience experience and the dashboard’s credibility. Strengthening each part deliberately — rather than focusing only on the rating output — is the move that turns performance management into a system the workforce trusts.

25.3 Rating Quality and Calibration

Even the most-modernised performance system eventually produces ratings. The discipline of analytics is to surface the quality of those ratings — the reliability across raters, the distribution across managers, the consistency across units — and to support calibration meetings with evidence rather than impressions.

What Disciplined Rating Analytics Looks Like

Property	What the analytics renders	Why it matters
Inter-rater agreement	Agreement statistic with confidence interval	Two raters disagreeing on the same employee is the most fixable error
Rating distribution	Distribution by manager and by unit	Reveals leniency, severity, and central-tendency bias
Distribution shift after calibration	Pre-and-post calibration comparison	Shows whether calibration meetings actually move ratings
Subgroup distribution	Distribution by demographic, role family, tenure	Surfaces patterns that need investigation
Stability over cycles	Same employee’s rating across cycles	Surfaces volatility that calls the reliability into question

Calibration as the working discipline

Calibration meetings — where managers compare ratings across their teams to align standards — are the most consequential single moment in the performance cycle. As Angelo S. DeNisi & Kevin R. Murphy (2017) emphasise, the meta-analytic evidence supports calibration as one of the few interventions that consistently improves rating reliability. The analytics function’s job is to surface the pre-meeting distribution, the post-meeting distribution, and the rating shifts attributable to the meeting, so that the calibration is a defensible process rather than a closed-door negotiation.

25.4 Predicting Performance

Performance prediction is a specific class of HR-analytics work. It uses information available now to forecast performance in a future cycle, in a current role or in a different one. The methods are familiar from the previous chapters on selection, but the application is different: the model is asked to score current employees, not external candidates, and the action that follows the score has different implications for trust and fairness.

Performance-Prediction Use Cases

Use case	What the model predicts	Action that follows
Promotion readiness	Performance in a more senior role	Inclusion in a promotion slate, development sponsorship
Cross-role mobility	Performance in a different functional role	Internal-mobility recommendation, learning offer
Leadership pipeline	Performance in a future leadership role	Succession plan placement, executive coaching
At-risk performance	Probability of falling below target	Manager support, coaching, training
High-potential identification	Likelihood of accelerated growth	High-potential programme inclusion

Quality criteria for a prediction model

A performance-prediction model is evaluated on the same criteria as a selection model: discrimination, calibration, subgroup robustness, stability, and decision utility. The dashboard surfaces all five for every model, and the visual that anchors the page is the calibration plot — predicted score against realised outcome — with the perfect-calibration line drawn for reference. As Herman Aguinis (2013) argues, the credibility of prediction inside HR depends on the function’s willingness to expose its accuracy honestly, including the cycles in which the model missed.

25.5 Visualising Performance Analytics

The performance-analytics dashboard has to surface a system rather than a single output. Five design choices, applied consistently, hold the system together on a page that managers, leaders, and the workforce can all read.

Five Design Choices for the Performance Dashboard

Choice	What it does on the page
System view as headline	The page opens with the five-part system, not with the rating output
Distribution panel	Rating distributions are shown by manager, unit, and subgroup
Calibration before-and-after	Each cycle’s calibration impact is visible on the page
Prediction calibration plot	Every prediction model has a predicted-versus-realised chart
Action-and-outcome trail	Last cycle’s predictions, actions, and outcomes are recorded

Reading performance as evidence

A performance-analytics dashboard succeeds when the audience reads performance as evidence rather than as opinion. The page that combines goal-setting metrics, feedback metrics, calibrated ratings, prediction calibration, and action tracking is the page that anchors the executive committee’s discussion of capability, succession, and reward. The function earns the right to inform those decisions by presenting the evidence with the discipline this chapter sets out, cycle after cycle, even when the evidence is uncomfortable.

25.6 Hands-On Exercise: Building the Performance-Analytics Page

Aim, Scenario, Dataset, Deliverable

Aim. Build a performance-analytics page that surfaces rating distributions, calibration evidence, and a per-employee performance-prediction model on a single Power BI page.

Scenario. You are running performance analytics for an organisation. The chief people officer wants a single page that opens with the five-part system view and lets line managers and the executive committee read the same evidence at different levels of detail.

Dataset. Performance Metrics (Excel) from the HRMD library, including EmployeeID, PerformanceRating, Manager, Department, Goals Achieved, and related fields. Optionally enrich with the IBM HR Attrition dataset (JobInvolvement, JobSatisfaction, YearsAtCompany) joined on a synthetic key for predictor variables.

Deliverable. A Performance-Analytics.xlsx workbook with rating distributions, calibration evidence, and a prediction model, plus a Performance-Analytics.pbix Power BI file with the page described below.

25.6.1 Step 1 — Compute manager-level rating distributions

Load the Performance Metrics workbook into Excel as a Table named Performance. Build a pivot of PerformanceRating distribution by Manager and by Department. Compute the mean rating, the standard deviation, and the percentage in each rating band per manager.

Code

Excel Formula

Mean Rating per Manager     = AVERAGEIFS(Performance[PerformanceRating], Performance[Manager], <mgr>)
Distribution Skew per Mgr   = COUNTIFS(Performance[Manager], <mgr>, Performance[PerformanceRating], 5)
                            / COUNTIF(Performance[Manager], <mgr>)

The distribution panel reveals leniency, severity, and central-tendency bias by manager.

25.6.2 Step 2 — Compute inter-rater agreement

Where the dataset includes a second rating from a manager-once-removed or a calibration-meeting outcome, compute the correlation between the two ratings as a working agreement measure.

Code

Excel Formula

Inter-Rater r = CORREL(Performance[ManagerRating], Performance[CalibratedRating])

If a second rating column is not present, simulate one for the lab by adding random noise to the existing rating (=PerformanceRating + RANDBETWEEN(-1,1)) so the calibration discipline can be practised.

25.6.3 Step 3 — Compute the calibration before-and-after

Compute the rating distribution before the calibration meeting and the distribution after it. The shift in distribution shape — typically a tightening of the central peak and a removal of inflated 5s — is the evidence of the meeting’s impact.

25.6.4 Step 4 — Build the performance-prediction model

Use the Data Analysis ToolPak’s regression to predict PerformanceRating from JobInvolvement, JobSatisfaction, YearsAtCompany, and Department. Save the predicted rating per employee.

25.6.5 Step 5 — Build the calibration plot

Bin predicted ratings into deciles and plot decile-mean predicted rating against decile-mean realised rating, with the perfect-calibration diagonal as a reference. The plot is the central visual of the prediction layer.

25.6.6 Step 6 — Compute subgroup robustness

Repeat the calibration analysis separately for tenure cohorts (0–2 years, 3–5, 6+) and for departments. The page surfaces the subgroup-level calibration so the audience can read whether the model performs equally across populations.

25.6.7 Step 7 — Promote to Power BI

Load the data into Power BI. Build the rating distribution, calibration before-and-after, prediction calibration plot, subgroup-robustness panels, and an action-and-outcome trail panel.

25.6.8 Step 8 — Open with the five-part system view

Add a top-of-page summary that names the five parts of the performance-management system from Section 2 of this chapter — goals, feedback, calibration, development, reward — and links each part to its supporting visual on the page. The page opens with the system rather than with the rating, as the design choice in Section 5 requires.

25.6.9 Step 9 — Publish

Publish the report and add it to the quarterly performance-management review and the recurring talent committee. Confirm that the calibration before-and-after panel is opened during the calibration meeting itself.

Connect to the Visualisation Layer

The performance-analytics page sits beside the recruitment funnel of Chapter 22 and the bias-and-prediction page of Chapter 24. Together the three pages cover the lifecycle of a hire from selection through performance evaluation, and the prediction calibration on this page feeds the optimisation calculations of Chapter 28.

Files and Screen Recordings

Performance-Analytics.xlsx, Performance-Analytics.pbix, and ch25-performance-walkthrough.mp4 will be attached at this point in the published edition. The screen recording walks through Steps 1 to 9 with the Excel rating-and-prediction workbench and the Power BI performance page shown side by side.

Summary

Concept	Description
Why Performance Analytics Matters
Performance as central outcome	Performance is the outcome every other HR programme is trying to influence
Annual rating limitations	Annual ratings have modest reliability, contested validity, and eroding workforce trust
System over event	Performance management is a system with goals, feedback, calibration, development, reward
Continuous performance management	Mature programmes run performance management continuously rather than as an annual event
Predictive-model evaluation	Predictive-performance models are judged on the same criteria as selection models
Performance as a System
Goal-setting part	Translates strategy into individual objectives with cascade coverage
Ongoing-feedback part	Provides feedback throughout the cycle rather than only at year-end
Calibration part	Aligns ratings across managers and units in a defensible meeting process
Development part	Acts on identified development needs through capability uplift
Reward part	Translates performance into pay and recognition with a defensible curve
Closing the system arc	Reward feeds back to the next cycle's goals and informs strategy
Rating Quality and Calibration
Inter-rater agreement	Two raters disagreeing on the same employee is the most fixable rating error
Rating distribution by manager	Distribution by manager reveals leniency, severity, and central-tendency bias
Pre-and-post calibration shift	Pre-and-post comparison shows whether calibration meetings actually move ratings
Subgroup rating distribution	Distribution by demographic, role family, and tenure surfaces patterns to investigate
Stability across cycles	Same employee's rating volatility across cycles questions reliability
Calibration as a working discipline	Calibration is one of the few interventions that consistently improves reliability
Predicting Performance
Promotion-readiness prediction	Predicting an employee's performance in a more senior role
Cross-role-mobility prediction	Predicting performance in a different functional role
Leadership-pipeline prediction	Predicting performance in a future leadership role
At-risk-performance prediction	Probability of falling below target with action support
High-potential identification	Likelihood of accelerated growth, used for high-potential programmes
Quality Criteria for Prediction
Discrimination criterion for prediction	Does the model rank employees correctly, captured by ROC and lift
Calibration criterion for prediction	Do predicted scores match realised outcomes, captured by calibration plot
Subgroup robustness for prediction	Does the model perform similarly across protected groups
Stability for prediction	Does the model perform similarly across cycles
Decision utility for prediction	Does the model improve outcomes net of cost
Visualising Performance
System view as headline	The page opens with the five-part system, not with the rating output
Distribution panel	Rating distributions are shown by manager, unit, and subgroup
Action-and-outcome trail	Last cycle's predictions, actions, and outcomes are recorded on the page