Practical guide

How a functional practice leader can build the case for applied AI capability evidence

A functional practice leader should not begin with a generic claim that the workforce needs more AI training. The stronger business case starts with a practice standard, a consequential work decision, and an evidence gap that current methods cannot resolve. It then compares realistic options, defines the smallest useful pilot, and names who has authority to act on the result. This leader is.

By Two Prune

A functional practice leader should not begin with a generic claim that the workforce needs more AI training. The stronger business case starts with a practice standard, a consequential work decision, and an evidence gap that current methods cannot resolve. It then compares realistic options, defines the smallest useful pilot, and names who has authority to act on the result. This leader is usually an owner of practice quality and capability, an influencer on investment, and a collaborator with learning, technology, risk, and business leaders. Budget authority varies by organization and should be mapped, not assumed.

Frame the case around a practice decision

Choose one decision that matters to the function, such as:

Then write the current evidence gap. Usage logs can show activity. Course completion can show participation. Manager observation can show selected workplace behavior. Work samples can show an output. None automatically supplies comparable, task-bounded evidence of how different people framed the same realistic assignment, navigated the same materials, used AI, checked the result, and communicated a decision.

The UK Government’s GovS 001 is a public-sector standard, not a universal private-sector operating model. It is still a useful primary-source example because it connects functional leadership with governance, planning, monitoring, assurance, capability, advice, service delivery, continuous improvement, and knowledge management. GovS 001: Government functions

  • whether a new AI-enabled workflow is ready for wider use
  • which verification behavior needs targeted practice
  • whether post-training evidence is useful to managers
  • where review burden has shifted after AI adoption
  • whether a common work standard needs repair

Sources: [1]

Build the CASE map

CASE forces the proposal to connect the practice decision to an evidence option and an accountable next step.

### C: Consequence and current standard

Describe the work standard and what happens when it is not met. Use operational evidence such as rework, missed constraints, unsupported claims, escalation failures, reviewer time, or inconsistent handoffs. Separate observed facts from hypotheses about AI’s contribution.

Record the standard in observable terms. “Use AI responsibly” is not testable. “For a sourced recommendation, identify material constraints, distinguish source fact from inference, verify every decision-relevant claim, and escalate unresolved conflicts” is closer to an assessable work standard.

### A: Alternatives and authority

Compare at least four options: improve manager observation, strengthen work review, use a knowledge check, add a realistic simulation, or make no change. For each, record what evidence it can produce, its burden, comparability, limitations, and the decision it could support.

Map the sponsor, practice owner, evidence reviewer, participant manager, learning partner, technology or risk reviewer, and budget authority. The UK Cabinet Office’s implementation guide says standards should be applied proportionately and highlights senior-leader accountabilities plus the role of functional leadership. Use it as an example of proportionality and role clarity, not proof that every organization should copy a government structure. Implementing functional standards: a guide for senior leaders

### S: Smallest useful evidence test

Define a pilot that can change a real decision. A bounded capability-evidence test should state:

Do not promise validation, productivity improvement, lower risk, or transfer to real work. The pilot asks whether the evidence is useful enough for a named decision under stated limitations.

### E: Economics, expiry, and executive ask

Estimate internal preparation, participant time, reviewer time, scenario adaptation, platform cost, governance effort, and follow-up work. Compare those costs with the current review burden and the cost of remaining uncertain. Use ranges where the inputs are immature.

Finish with a specific ask, evidence window, owner, and expiry. A request for a bounded pilot is different from a request for enterprise rollout. If no decision owner will use the result, do not run the pilot.

  • the practice standard and target behavior
  • the synthetic or explicitly approved non-sensitive task
  • the participant group and inclusion rule
  • the comparison or baseline, if any
  • the deliverable and available workflow evidence
  • the interpretation protocol and reviewer
  • accessibility, security, fairness, and data-handling gates
  • the proceed, repair, hold, and stop conditions

Sources: [2]

Copyable decision and evidence map

| Practice decision | Current evidence | Missing evidence | Options compared | Proposed test | Decision owner | Expiry | | --- | --- | --- | --- | --- | --- | --- | | whether sourced recommendation work needs targeted verification practice | review notes and manager observations | comparable evidence of source checking under the same constraints | observation, work review, knowledge check, simulation, no change | small synthetic simulation with human-reviewed evidence | named practice leader | date |

Attach a practice-standard card, alternative comparison, authority map, pilot protocol, cost range, claim ledger, and result-use plan. Keep the business case separate from the eventual result so that the success criteria cannot be rewritten after the evidence is seen.

Worked example: proposal quality practice

A proposal practice leader sees more reviewer comments about unsupported statements and missed requirements after teams adopt drafting tools. The available data includes four months of quality-review notes, tool usage counts, course completion, and manager interviews. The leader cannot infer that AI caused the defects because work mix, deadline pressure, reviewer behavior, and source quality also changed.

Consequence and standard: the team defines a narrow standard for a synthetic compliance-red-team task. Participants must identify mandatory requirements, link corrections to supplied evidence, flag unsupported claims, and communicate a prioritized correction plan. The leader owns the practice standard, not employment decisions about participants.

Alternatives: manager observation is inexpensive but inconsistent. Live-work review is relevant but uses incomparable bids and sensitive material. A knowledge quiz is comparable but weak on applied behavior. A synthetic simulation creates a consistent opportunity to observe work but may not transfer to live performance. The no-change option preserves current review burden and uncertainty.

Smallest test: twelve volunteers complete the same 30-minute synthetic exercise after accessibility and security review. Two trained reviewers interpret task-bounded evidence using a frozen protocol. The test does not rank people or set a pass threshold. It asks whether the evidence identifies actionable practice questions that managers can use without disproportionate review time.

Economics and ask: the case estimates scenario preparation, participant time, double review, calibration, and a decision meeting. The ask is approval for one evidence test, not broad purchase. Proceed requires acceptable reviewer agreement on evidence location, manageable interpretation time, no unresolved data-handling issue, and at least one predeclared practice decision that the evidence can inform. Otherwise the disposition is repair or stop.

Objections the case should answer

“We already have usage analytics.” Treat them as adoption evidence, not direct proof of judgment, verification, or work quality.

“Managers already know who is capable.” Manager observation may be valuable. Test whether a consistent task adds information, exposes missing evidence, or simply duplicates current judgment.

“A simulation is not real work.” Correct. State the transfer limitation and use the result as one bounded evidence source for human interpretation.

“We could build this ourselves.” Compare the real internal work: scenario design, source materials, workspace operation, rubric definition, reviewer training, missingness handling, accessibility, reporting, and maintenance.

Two Prune creates controlled work simulations and structured assessment evidence for human interpretation. It does not prove workplace performance, promise business outcomes, rank participants, or make automated employment decisions.

Sources

UK Government Functional Standard GovS 001: Government functions

  • UK Cabinet Office: Implementing functional standards, a guide for senior leaders

Sources: [1] [2]