Practical guide

AI-assisted proposal checklist: free template and guide

An AI-assisted proposal is ready for review only when a decision maker can trace the need, compare credible options, inspect the evidence and assumptions, understand costs and risks, and see who has authority to approve the next step. Polished prose is not a substitute for that decision record. This copyable template uses PROOF: purpose, routes, origins, operating case, and final review.

By Two Prune

An AI-assisted proposal is ready for review only when a decision maker can trace the need, compare credible options, inspect the evidence and assumptions, understand costs and risks, and see who has authority to approve the next step. Polished prose is not a substitute for that decision record. This copyable template uses PROOF: purpose, routes, origins, operating case, and final review.

Source basis and limits

HM Treasury's current guidance on developing business cases asks reviewers to examine objectives, options, benefits, risks, constraints, dependencies, commercial and financial arrangements, governance, and evaluation. Its October 2026 proportionality guidance says the depth of analysis should reflect scale, cost, complexity, and risk, while preserving information needed for a sound decision. The US Federal Acquisition Regulation requires proposal evaluation against disclosed factors and documentation of strengths, weaknesses, deficiencies, and risks. These are useful control principles even when a proposal is not a public-sector business case or procurement submission; the receiving organization's own rules remain authoritative. HM Treasury business-case guidance Proportionality guidance FAR 15.305

Sources: [1] [2] [3]

Copyable PROOF checklist

### Document control

| Field | Entry | | --- | --- | | Proposal, version, and decision date | | | Decision requested | | | Author, accountable owner, reviewers, approver | | | Evidence cut-off and source folder | | | Applicable evaluation criteria | | | Confidentiality and tool-use restrictions | |

### P — Purpose and decision

Decision line: Approve / decline / request evidence for [bounded action], by [owner], before [date], within [limits].

### R — Routes and alternatives

| Option | What changes | Benefits | Whole-life cost or effort | Risks and dependencies | Evidence confidence | Reason retained or rejected | | --- | --- | --- | --- | --- | --- | --- | | Business as usual | | | | | | | | Minimum change | | | | | | | | Proposed route | | | | | | | | Credible alternative | | | | | | |

### O — Origins of claims and numbers

| Claim or number | Source ID and date | Calculation or extraction | Assumption | Owner | Verification status | | --- | --- | --- | --- | --- | --- | | | | | | | |

### O — Operating case

| Element | Required evidence | | --- | --- | | Deliverables and acceptance | Observable output, acceptance rule, and accepting owner | | Plan and capacity | Milestones, dependencies, skills, available people, and estimate basis | | Cost and affordability | Timing, whole-life view, funding source, sensitivity, and exclusions | | Risk and control | Exposure, control owner, test, escalation trigger, and residual uncertainty | | Governance | Decision rights, specialist reviews, change authority, and reporting cadence | | Measurement | Baseline, leading and lagging indicators, missingness, comparison, and review date | | Transition | Pilot boundary, support, fallback, exit, data handling, and lessons capture |

### F — Final independent review

Ask a reviewer who did not draft the proposal to answer:

Record each issue as accepted, repaired, accepted with condition, or unresolved. Preserve the prior text and the reason for change.

  • [ ] State the problem or opportunity in one testable sentence.
  • [ ] Name the people affected and the decision this proposal supports.
  • [ ] Separate the desired outcome from the proposed solution.
  • [ ] Define scope, exclusions, constraints, and the consequence of doing nothing.
  • [ ] State what approval would authorize now: discovery, pilot, procurement, rollout, or another bounded step.
  • [ ] Record what the proposal does not establish.
  • [ ] Use the same criteria for every option.
  • [ ] Include a baseline and minimum-change route rather than comparing only preferred suppliers or ideas.
  • [ ] Explain why excluded options are not decision-relevant.
  • [ ] Test the ranking when a major cost, benefit, delay, or adoption assumption changes.
  • [ ] Link every material claim to an accessible source, not a model-generated citation.
  • [ ] Distinguish observed facts, estimates, forecasts, opinions, and AI suggestions.
  • [ ] Reproduce totals from source values and retain units, denominators, and time periods.
  • [ ] Record missing, conflicting, or stale evidence rather than smoothing it away.
  • [ ] Check whether quoted evidence supports the precise proposition made.
  • [ ] Mark assumptions with an owner and a date for confirmation or expiry.
  • [ ] Confirm that named owners accepted their responsibilities.
  • [ ] Check privacy, security, accessibility, legal, finance, commercial, and operational review where relevant.
  • [ ] Describe how adverse evidence can pause, narrow, or end the work.
  • [ ] Keep an alternative route available when uncertainty is material.
  • Can I identify the exact decision and authority requested?
  • Could I reconstruct the important claims and calculations?
  • Were alternatives assessed on comparable criteria?
  • Which conclusion changes under a plausible counter-scenario?
  • Are major exceptions visible in the main proposal?
  • Are controls assigned, testable, and timed?
  • Does the recommendation exceed the evidence?

Bounded AI workflow

Use AI for inventory, structure, challenge, and editing. Provide approved source identifiers; ask it to extract candidate requirements and gaps; require source IDs beside claims; and ask for counterarguments and missing alternatives. Then reopen the original sources, reproduce calculations, confirm owner commitments, and have the appropriate people make the judgment.

Do not paste restricted information into an unapproved tool. Do not let AI invent performance history, customer evidence, legal conclusions, prices, commitments, citations, or approvals. A fluent draft with inaccessible sources remains unverified.

Worked example: a fictional support-triage assistant pilot

A service manager proposes a six-month rollout of an AI assistant after a two-week demonstration. The first draft claims a 25% handling-time reduction, lists only the preferred tool, and describes privacy and accessibility review as post-approval activities.

PROOF reframes the immediate decision as a six-week, 40-case pilot with no autonomous customer communication. The routes table compares business as usual, a revised knowledge-base search, and the bounded assistant pilot. The 25% figure is relabelled as a vendor scenario because the team has no local baseline. Cost now includes review time, configuration, accessibility testing, incident handling, and exit work.

The origins ledger finds that three cited links discuss generic contact-centre automation rather than the proposed workflow. They are removed. The operating case names a service owner, privacy reviewer, accessibility reviewer, and frontline quality reviewer; defines acceptance as accurate routing with evidence links and recorded human disposition; and retains the existing triage path as fallback.

A counter-scenario shows that if review takes more than four minutes per case, the pilot does not release capacity even when drafting is faster. The proposal therefore asks to learn whether assisted triage is usable and controllable, not to approve full deployment or claim productivity. The approver can now choose a bounded experiment with visible stop conditions.

Adaptation guidance

For a low-cost internal change, shorten the narrative but retain the decision, alternatives, source ledger, ownership, risks, and review. For a regulated or high-consequence proposal, add the organization's formal control mapping and specialist sign-offs. For a supplier response, map every section to the buyer's stated evaluation factors and keep unsupported marketing claims out of the evidence table.

Version history

| Version | Date | Material change | Evidence affected | Author | Reviewer | | --- | --- | --- | --- | --- | --- | | | | | | | |

Two Prune boundary

Two Prune can assess whether a participant frames a proposal decision, navigates supplied materials, uses AI thoughtfully, verifies evidence, identifies uncertainty, and produces a decision-ready deliverable in a realistic simulation. It reports observed assessment evidence. It does not approve proposals, validate business cases, certify controls, accept risk, replace specialist review, or make employment decisions.

Sources

HM Treasury: Guidance on developing business cases

  • HM Treasury: Proportionality guidance for business cases
  • Acquisition.gov: FAR 15.305 Proposal evaluation

Sources: [1] [2] [3]

Sources

  1. 1.HM Treasury business-case guidance · UK Government
  2. 2.Proportionality guidance · UK Government
  3. 3.FAR 15.305 · U.S. Acquisition.gov