Practical guide
How a people analytics leader can build the case for applied AI capability evidence
A people analytics leader can build a defensible case for an applied AI capability pilot by connecting a real work decision to a proportionate evidence plan. The role should not begin with a platform purchase, a workforce-readiness score, or a promised productivity return. It should help the organization understand the problem, alternatives, evidence gaps, safeguards, costs, and next decision..
A people analytics leader can build a defensible case for an applied AI capability pilot by connecting a real work decision to a proportionate evidence plan. The role should not begin with a platform purchase, a workforce-readiness score, or a promised productivity return. It should help the organization understand the problem, alternatives, evidence gaps, safeguards, costs, and next decision. Use the CASES method: change, alternatives, stakeholders, evidence, and staged decision.
Role basis and evidence limits
CIPD describes people analytics as using data about people and the business to inform organizational decision-making. Its evidence-based-practice framework combines organizational data with practitioner expertise, scientific literature, and stakeholder perspectives. The 2026 UK Green Book requires a case for change, business-as-usual comparison, objectives, options, evidence, risks, distributional effects, and transparent uncertainty. The current UK employer AI-upskilling guide recommends capability development linked to real work, roles, systems, governance, and responsible practice. Together these sources support the business-case method below; they do not prove that an assessment programme improves organizational outcomes or that a people analytics leader owns the investment decision. CIPD: People analytics guide CIPD Profession Map: Evidence-based practice The Green Book 2026 UK employer guide to AI upskilling
Sources: [1] [2] [3] [4]
C — Define the change case
State the organizational decision and the observed work problem before naming a solution. A useful change case identifies:
Avoid claims such as “employees lack AI skills” when the evidence is a low tool-usage rate. Usage may reflect access, policy, relevance, workload, or measurement configuration. Define the narrower question that a pilot can answer, such as whether a representative simulation and artifact review reveal actionable verification gaps for one workflow.
Create a theory-of-change chain but label each link:
programme input → participant activity → observed assessment evidence → learning or workflow action → later operational evidence
An assessment can directly inform only the observed-evidence link. Later links require separate implementation and evaluation evidence.
- the workflow, population, and operating context;
- current quality, control, access, or learning evidence;
- who experiences the problem and who carries the consequence;
- the current practice and its cost or limitation;
- the decision deadline and accountable owner; and
- what remains assumption, hypothesis, or missing evidence.
A — Appraise real alternatives
Compare at least:
Use an options table rather than a single weighted score.
| Option | Decision value | Evidence produced | Cost and effort | Rights and control implications | Key uncertainty | Exit or reversal | | --- | --- | --- | --- | --- | --- | --- | | | | | | | | |
Show trade-offs that a total score would obscure. A survey may be inexpensive and inclusive but cannot directly observe performance. Telemetry may show events but not whether the work was correct. A simulation may reveal process evidence but cannot by itself establish workplace transfer or business impact.
- continue current practice;
- repair access, policy, workflow, or manager support without an assessment;
- use surveys, interviews, artifact review, telemetry, or existing quality data;
- run a bounded applied assessment pilot; and
- defer action while gathering missing evidence.
S — Map stakeholders and decision rights
Classify each stakeholder as owner, influencer, reviewer, or collaborator for each decision.
| Decision or control | Accountable owner | People analytics role | Required reviewers or collaborators | | --- | --- | --- | --- | | Business problem and priority | Programme or business owner | Evidence adviser | Workflow leaders, affected staff | | Assessment purpose and construct | Assessment owner | Design and measurement adviser | Subject specialists, learning | | People-data use and retention | Privacy or information-governance owner | Specify evidence need | Legal, security, employee relations | | Learning response | Learning owner | Diagnose pattern and evaluate | Managers, workflow specialists | | Investment and scale | Named executive or budget owner | Present evidence and uncertainty | Finance, technology, operations |
Do not assume a title determines authority. Record the local allocation, challenge route, and who may pause or stop the pilot.
E — Specify evidence, safeguards, and economics
Define the minimum evidence needed for the next decision. Separate:
For each measure, document the construct, numerator, denominator, population, period, source, owner, missingness, confounds, and interpretation limit. Predefine critical failures that cannot be averaged away.
Build costs from stated quantities and ranges: design, subject review, participant time, accessibility, technology, administration, analysis, governance, learning response, and maintenance. Do not monetize benefits without a supportable causal and valuation route. Where evidence is too weak, present the item qualitatively or as an uncertainty rather than inventing precision.
Include purpose limitation, data minimization, access control, retention, accessibility, participant communication, review, and correction routes under the applicable organizational and legal requirements. Do not silently repurpose assessment evidence for an employment decision.
- eligibility and access;
- self-reported confidence or experience;
- observed applied capability in representative work;
- artifact quality and critical exceptions;
- required control operation;
- learning response; and
- later operational outcomes.
S — Stage the decision and evidence gates
Ask for the smallest reversible action that can resolve an important uncertainty. A pilot proposal should define:
Use decision gates:
Gate 0 — readiness: purpose, owners, source materials, access, safeguards, and task realism are sufficient. Gate 1 — pilot quality: delivery, participation, evidence completeness, and critical controls meet the protocol. Gate 2 — interpretation: findings are stable enough for the bounded decision and rival explanations are visible. Gate 3 — response: a named owner accepts a learning, workflow, or evidence action. Gate 4 — later evaluation: separate evidence examines transfer or operational outcomes.
Do not treat a pilot pass as automatic authority to expand.
- one workflow and a representative task;
- eligible population and recruitment route;
- comparison or baseline where appropriate;
- assessment and artifact criteria;
- process, rights, and critical-failure controls;
- analysis and independent review plan;
- stop, repair, proceed, and scale conditions;
- owner and date for each decision; and
- what the pilot cannot establish.
Worked case: a fictional customer-operations team
A people analytics leader is asked to justify an enterprise “AI readiness assessment” after a dashboard shows uneven use. CASES narrows the problem to one workflow: drafting evidence-based responses to complex service complaints. Quality reviews already show unsupported explanations and inconsistent source use, but the records do not distinguish whether problems arise from access, guidance, knowledge, or verification practice.
The alternatives review rejects a usage-only dashboard because it cannot show work quality. A survey remains useful for access and support needs. The team proposes a 30-person simulation pilot using synthetic case files, an artifact rubric, observed source review, AI interaction evidence, and independent quality review. Privacy and employee-relations reviewers confirm the bounded purpose and retention plan; the service owner owns workflow standards; learning owns any development response.
The cost range includes design, specialist review, participant time, accessibility review, analysis, and a repair allowance. The business case does not convert minutes saved into a benefit because no transfer or time evidence exists. Its immediate value is reducing uncertainty about which support action is warranted.
Gate 1 reveals that six participants lacked the approved reference system. Their sessions are reported as inaccessible, not low performance. Three accessible sessions contain unsupported citations, preserved as critical exceptions. The decision owner approves a targeted verification module and a repeated pilot after access repair. Expansion remains unapproved until the repeat and later workplace quality evidence are reviewed.
Business-case review checklist
[ ] The case begins with a decision and evidenced work problem, not a preferred product.
- [ ] Business as usual and non-assessment alternatives are compared fairly.
- [ ] Role ownership reflects local authority.
- [ ] Access, confidence, use, capability, quality, controls, and outcomes remain distinct.
- [ ] Costs use stated quantities, ranges, sources, and owners.
- [ ] Benefits are not monetized beyond the evidence.
- [ ] Rights, privacy, security, accessibility, employee-relations, and retention controls have owners.
- [ ] The pilot has critical failures, missingness rules, review, stop conditions, and an exit.
- [ ] Each decision gate states what the evidence can and cannot establish.
Two Prune boundary
Two Prune can provide realistic assessment simulations and evidence about how participants frame tasks, navigate supplied information, use AI, verify work, exercise judgment, and communicate deliverables. It reports observed assessment performance and evidence strength. It does not prove workforce readiness, productivity, transfer, return on investment, or workplace performance; it does not determine funding or make employment decisions.
Sources
CIPD: People analytics guide for people professionals
- CIPD Profession Map: Evidence-based practice
- HM Treasury: The Green Book 2026
- GOV.UK: Employer guide to AI upskilling
Sources: [1] [2] [3] [4]
Sources
- 1.CIPD: People analytics guide · prod.cipd.org
- 2.CIPD Profession Map: Evidence-based practice · www.cipd.org
- 3.The Green Book 2026 · UK Government
- 4.UK employer guide to AI upskilling · UK Government