Practical guide

An AI capability evidence agenda for a Chief Human Resources Officer

A Chief Human Resources Officer needs more than a count of AI-tool licenses or course completions. The useful agenda connects business priorities, changing work, observed capability, employee experience, governance, and a small set of decisions the people function can own or influence. Treat the CHRO as the owner of workforce capability and people-change evidence, a collaborator on enterprise AI.

By Two Prune

A Chief Human Resources Officer needs more than a count of AI-tool licenses or course completions. The useful agenda connects business priorities, changing work, observed capability, employee experience, governance, and a small set of decisions the people function can own or influence. Treat the CHRO as the owner of workforce capability and people-change evidence, a collaborator on enterprise AI governance, and a reviewer of workforce consequences. The exact mandate varies. Confirm which decisions belong to the CHRO, technology leadership, business-unit owners, legal counsel, employee representatives, and the board rather than assuming the title carries every authority.

Use the PEOPLE decision agenda

The PEOPLE method structures a recurring executive agenda:

Each agenda item should end with a decision, owner, date, and evidence needed next. Separate workforce planning, capability assessment, learning activity, employee listening, and operational outcomes. They inform one another but are not interchangeable.

  • P — Priorities and work changes
  • E — Evidence of observed capability
  • O — Operating model and ownership
  • P — Participation and people impact
  • L — Learning and role interventions
  • E — Evidence gates and executive decisions

P — Priorities and work changes

Start with two or three business workflows where AI is changing tasks now. For each, document:

Avoid beginning with a generic “AI skills” catalogue. Capability requirements should follow the real work: evaluating sources, checking a model-generated analysis, managing a human-AI handoff, or escalating an unsupported claim.

CIPD’s current AI skills-planning guidance asks people professionals to align skills work with organizational goals, examine job families and changing tasks, use internal evidence, shorten review cycles, and integrate monitoring into normal people processes. It also emphasizes collaboration with technology and business leaders. CIPD AI skills planning

  • the business objective and process owner;
  • tasks being augmented, redesigned, introduced, or removed;
  • human decisions and verification responsibilities;
  • affected job families and locations;
  • current tools, policies, and information boundaries;
  • the consequence if capability is missing;
  • the decision required this quarter.

Sources: [1]

E — Evidence of observed capability

Use multiple evidence layers and name what each can support:

Report denominators, missingness, assessment version, evidence strength, and relevant distribution—not only an average score. An assessment can show what participants demonstrated under stated conditions; it does not by itself establish job performance or business impact.

For Two Prune, a capability view can organize evidence across context framing, evidence navigation, AI orchestration, verification and risk control, judgment and synthesis, and communication and delivery. The CHRO agenda should connect those signals to the workflow without turning the report into a personnel ranking.

  • Access: approved tools and eligible populations.
  • Activity: training attendance, practice, and tool usage.
  • Observed capability: performance in a defined, realistic work task.
  • Transfer evidence: use of the capability in a real workflow.
  • Operational outcomes: quality, cycle time, rework, risk events, or another pre-defined measure.

O — Operating model and ownership

Map who owns the decisions and controls around AI-enabled work. A practical table is:

| Decision | Owner | CHRO role | Collaborators | Required evidence | | --- | --- | --- | --- | --- | | redesign a customer-service workflow | operations leader | collaborator on roles and change | technology, legal, service leads | task map, risk review, employee input | | define capability expectations | CHRO / learning leader | owner | process owners, AI governance | workflow evidence and assessment construct | | approve AI service and data use | designated governance authority | reviewer or collaborator | security, privacy, legal, procurement | system and control evidence | | change a development programme | learning leader | sponsor / owner | managers, participants | completion, observed capability, transfer evidence |

CIPD’s guidance on responsible AI use says people professionals should participate in governance wherever AI affects people, including skills, roles, employee data, change, and employee involvement. It frames this as cross-functional work rather than a people-team-only programme. CIPD responsible AI guidance

Sources: [2]

P — Participation and people impact

Place participation evidence beside capability evidence. Track who was invited, who started, who completed, who could not participate, and which groups are absent. Review access to practice, manager support, workload, accommodation, language, location, and employee concerns.

Do not infer low capability from missing participation. Do not use a small voluntary cohort to describe the whole workforce. Aggregate results at a level that protects people and supports the decision; restrict record-level access to the stated purpose.

The U.S. Department of Labor’s AI best-practices roadmap is a jurisdiction-specific government reference that emphasizes worker input, transparency, training, meaningful human oversight for significant employment decisions, worker rights, job quality, and data protection. Applicability must be assessed locally, but these topics are useful checks for the CHRO’s people-impact agenda. U.S. Department of Labor AI Best Practices

Sources: [3]

L — Learning and role interventions

Choose an intervention that matches the evidence gap. Options include:

Course completion is an activity measure. Define what participants should demonstrate afterward, when it will be assessed, and what transfer evidence will be collected. If the problem is a broken workflow or missing authority, more training alone is unlikely to solve it.

  • practice with a page-specific verification artifact;
  • manager coaching on challenge and escalation;
  • redesign of the workflow or review interface;
  • clearer source, data, or approval rules;
  • structured peer review;
  • a targeted work simulation followed by feedback;
  • specialist support for a high-consequence task.

E — Evidence gates and executive decisions

End the meeting with a decision register. Each row should contain:

Useful questions include:

  • decision: continue, change, constrain, investigate, or pause;
  • evidence supporting the decision;
  • evidence limitations and affected groups;
  • accountable owner and collaborators;
  • intervention and resource request;
  • review date and acceptance test;
  • communication and employee-involvement action.
  • Which workflow has enough evidence for the next controlled expansion?
  • Where is participation too incomplete to interpret capability?
  • Which gap is a skill issue, and which is a process or control issue?
  • What support do managers need to reinforce verification behavior?
  • Which role changes require deeper consultation or scenario planning?
  • What evidence must exist before the next investment or redesign decision?

Worked example: finance planning capability

A company introduces an AI assistant for monthly variance analysis. Tool telemetry shows 82% of the finance-planning team used it. Training records show 91% course completion. In a realistic assessment, 38 of 46 invited employees complete a task containing a stale forecast and a conflicting business-unit note. Most frame the executive question clearly, but 14 accept the stale forecast and nine fail to disclose the conflict.

The CHRO agenda does not label 82% usage as capability or describe the eight non-completers as weak performers. It reports each denominator and identifies a verification pattern in the completed cohort. The process owner also finds that the live workflow does not display the forecast timestamp prominently.

The decision is therefore shared: finance owns the interface and source-control repair; learning owns targeted practice on conflict handling; managers receive a review prompt; the assessment team repeats a defined challenge after the repair. The next gate requires a complete participation account, observed improvement on the challenge, and evidence that the live workflow exposes source dates. No productivity claim is made because a comparable operational baseline is absent.

Reporting boundary

Two Prune provides evidence from realistic AI-enabled work simulations. It is not an applicant-tracking or employment-decision system, and it should not issue a recommendation about a person’s employment. The CHRO and other accountable leaders decide how aggregate assessment evidence informs development, workforce planning, role design, and governance, subject to applicable policy, law, consultation, and human judgment.