Practical guide
AI-assisted project plan template and guide
An AI-assisted project plan should make delivery logic, authority, evidence, uncertainty, and control visible. It should not be a confident schedule assembled from incomplete notes. AI can help inventory source material, detect omissions, compare versions, draft structured fields, and challenge assumptions. People remain accountable for objectives, commitments, estimates, dependencies, risk.
An AI-assisted project plan should make delivery logic, authority, evidence, uncertainty, and control visible. It should not be a confident schedule assembled from incomplete notes. AI can help inventory source material, detect omissions, compare versions, draft structured fields, and challenge assumptions. People remain accountable for objectives, commitments, estimates, dependencies, risk decisions, and approval. This template uses PLAN-R: purpose, lineage, activities, navigation and control, and review.
Source basis and limits
The current UK Government Functional Standard for Project Delivery sets expectations for governance, roles, planning, control, and solution delivery across the lifecycle. The Orange Book integrates risk and uncertainty with governance and decision-making. The AQuA Book calls for traceable inputs, assumptions, limitations, changes, decisions, and proportionate assurance. The current Government Functional Standard for Digital also requires evidence, user needs, benefits, uncertainty, risks, opportunities, assumptions, dependencies, and responsible AI considerations in relevant digital decisions. These sources support the controls in this template; they do not approve a plan or make it suitable for every organization. GovS 002: Project Delivery The Orange Book The AQuA Book GovS 005: Digital
Sources: [1] [2] [3] [4]
Copyable project plan
### Document control
| Field | Entry | | --- | --- | | Project and plan ID | | | Version and status | | | Planning horizon and data cut-off | | | Author, owner, independent reviewer, approver | | | Approved source snapshot | | | Next review date and change threshold | |
### P — Purpose and authorized outcome
Decision context: What authorized decision or commitment does this plan support? Problem or opportunity: What evidence establishes the need? Outcome: What change should exist, for whom, and by when? Outputs: What accepted products or services enable that outcome? Success and acceptance: What observable conditions, measures, and approvers determine acceptance? Scope: What is included? Exclusions: What is explicitly outside the project? Constraints: Time, cost, quality, policy, data, accessibility, security, technology, and resource constraints. Non-claims: What will completion not prove?
Do not substitute activity for outcome. “Run training” is an output; it is not evidence that participants can perform a representative task or that workplace results changed.
### L — Lineage, assumptions, and decision rights
| ID | Source, requirement, or assumption | Owner | Date/version | Plan consequence | Verification or expiry | | --- | --- | --- | --- | --- | --- | | | | | | | |
| Decision or control | Accountable owner | Contributors or reviewers | Authority limit | Escalation route | | --- | --- | --- | --- | --- | | | | | | |
Label assumptions rather than smoothing them into facts. Record conflicts between sources and route them to an authorized owner. An AI-generated reconciliation is not authority to choose between conflicting requirements.
### A — Activities, milestones, and resources
Build the plan from accepted outputs backward to the work and evidence needed.
| Work package | Output and acceptance evidence | Owner | Inputs and dependencies | Estimate basis | Start/finish or range | Resources | Status | | --- | --- | --- | --- | --- | --- | --- | --- | | | | | | | | | |
Use ranges or scenarios when a single date would imply false precision. Separate effort, elapsed time, contingency, and dependency waiting time. Show how accessibility, assurance, transition, support, and disposal work fit the plan rather than appending them after delivery.
For every milestone, name the decision or accepted evidence it represents. “Draft complete” is ambiguous unless the expected artifact, quality rule, reviewer, and disposition are known.
### N — Navigation, dependencies, and control
Maintain linked control records:
| ID | Type | Statement | Evidence | Exposure | Owner | Response | Trigger/date | Residual uncertainty | | --- | --- | --- | --- | --- | --- | --- | --- | --- | | | Risk / issue / dependency / opportunity / decision | | | | | | | |
| Change request | Baseline affected | Evidence and reason | Options and effects | Authority | Decision | Downstream updates | | --- | --- | --- | --- | --- | --- | --- | | | | | | | | |
Define tolerances for time, cost, scope, quality, benefits, risk, and evidence confidence. State what can be managed within the project and what requires escalation. Do not mark a risk closed merely because a mitigation action was assigned; verify whether the control operated and whether residual exposure remains.
### R — Review, assurance, and learning
| Review | Question | Reviewer | Evidence needed | Timing | Disposition retained | | --- | --- | --- | --- | --- | --- | | | | | | | |
Include proportionate reviews for plan quality, subject matter, accessibility, privacy, security, finance, commercial, operations, and benefits where relevant. Record accepted, repaired, accepted with condition, or not verified; preserve dissent and limitations.
At each review point ask:
- Are objectives, outputs, and current work still aligned?
- Have source requirements, assumptions, or dependencies changed?
- Can milestones and forecasts be reproduced from evidence?
- Are key exceptions hidden by aggregate status?
- Does the plan still provide a viable transition, fallback, and support route?
- What should stop, repair, continue, narrow, or be reauthorized?
Bounded AI workflow
Never paste unapproved confidential information into an AI tool. Do not ask the model to make commitments on behalf of people, accept risk, or silently update the approved baseline.
- Inventory: provide approved source IDs and ask AI to list requirements, dates, owners, gaps, and conflicts.
- Structure: draft PLAN-R fields one section at a time, retaining source IDs beside material entries.
- Estimate challenge: ask for missing assumptions, dependency waits, range drivers, and counter-scenarios, not an invented precise duration.
- Control challenge: search for orphan outputs, missing owners, contradictory dates, untested acceptance rules, and unplanned transition work.
- Human verification: reopen sources, confirm commitments with owners, reproduce calculations, obtain specialist review, and record dispositions.
Worked example: fictional knowledge-base migration
A service team plans to migrate 1,800 knowledge articles to a new platform. The first AI draft produces a six-week schedule from meeting notes and assumes that a bulk import feature, a premium entitlement, and two content owners are available.
PLAN-R begins with the authorized outcome: agents can retrieve approved, accessible guidance without losing current version history. The outputs are a reconciled article inventory, accepted information architecture, tested migration tooling, accessible templates, trained owners, a fallback route, and an operational handover. Completion does not prove faster service or better customer outcomes.
The lineage table finds that the premium entitlement is not in the contract, 320 articles have no confirmed owner, and accessibility testing covers only the landing page. The plan records these as an unresolved commercial dependency, a content-ownership issue, and an assurance gap. It does not let AI infer owners from authorship metadata.
The schedule becomes a range. A four-week pilot migrates 120 lower-risk articles after contract confirmation and keyboard and screen-reader testing. The full migration remains conditional on duplicate resolution, ownership, failure-rate evidence, support capacity, and rollback rehearsal. A content-freeze change requires the service owner's approval; privacy and accessibility specialists retain authority within their mandates.
During review, AI flags inconsistent article counts across the inventory and prior report. Human reconciliation finds archived duplicates rather than silently choosing the larger number. The decision record approves the pilot only, retains the legacy platform as fallback, and schedules a gate using measured migration exceptions and task-based accessibility evidence.
Plan QA checklist
[ ] Purpose, outcome, outputs, scope, exclusions, constraints, and non-claims are distinct.
- [ ] Requirements and assumptions have source IDs, owners, dates, and expiry or verification routes.
- [ ] Decision rights and authority limits are explicit.
- [ ] Work packages connect accepted outputs to inputs, dependencies, estimates, resources, and evidence.
- [ ] Forecast ranges show their drivers.
- [ ] Risks, issues, dependencies, opportunities, and decisions are distinguished.
- [ ] Tolerances, escalation, fallback, transition, support, and disposal are planned.
- [ ] AI-generated claims, dates, citations, classifications, and calculations were independently checked.
- [ ] Reviews preserve repairs, conditions, dissent, and unresolved uncertainty.
- [ ] Version history shows who approved each baseline change.
Version history
| Version | Date | Change and evidence | Author | Reviewer | Approver | | --- | --- | --- | --- | --- | --- | | | | | | | |
Two Prune boundary
Two Prune can assess whether a participant frames a project, navigates supplied records, uses AI thoughtfully, identifies assumptions and conflicts, controls risk, and produces a decision-ready plan in a realistic simulation. It reports observed assessment evidence. It does not approve projects, provide delivery assurance, validate estimates, accept risk, replace specialist review, or make employment decisions.
Sources
UK Government Functional Standard GovS 002: Project Delivery
- HM Treasury: The Orange Book
- GOV.UK: The AQuA Book
- UK Government Functional Standard GovS 005: Digital
Sources: [1] [2] [3] [4]
Sources
- 1.GovS 002: Project Delivery · UK Government
- 2.The Orange Book · UK Government
- 3.The AQuA Book · UK Government
- 4.GovS 005: Digital · UK Government