How to develop and assess scope control in AI-enabled work

Treat scope as a controlled agreement

NASA defines a scope baseline as the approved version of a project's goals, scope statement, work breakdown structure, requirements, and minimum product definition. Its programme guidance distinguishes the approved baseline from later changes and requires rebaselining when important goals, requirements, milestones, or design decisions change. The US Government Accountability Office similarly explains that a baseline supports comparison with current work and that changes should be documented and controlled so performance is not rewritten after the fact.

Knowledge work rarely needs the same governance as a large programme, but the control principle transfers. Before work begins, the worker should be able to show the agreed outcome, boundaries, evidence set, decision owner, deadline, and acceptance conditions. A new idea is not automatically part of the job. It becomes a proposed change whose benefit and impact can be assessed.

AI makes this discipline more important because it can produce plausible extra analyses, audiences, formats, and recommendations faster than a person can evaluate them. The worker remains responsible for deciding whether each addition helps the agreed outcome or changes it.

Keep a SCOPE control record

Use a short SCOPE record beside the work plan:

| Field | Control question | Evidence to retain | | --- | --- | --- | | Statement | What outcome and artifact were agreed? | One-sentence outcome, deliverable, and owner | | Constraints | Which time, source, policy, audience, and format limits apply? | Constraint list with source or authority | | Outside | What is explicitly excluded or deferred? | Out-of-scope list and reason | | Proposals | Which new requests or discoveries could change the baseline? | Change request with requester and date | | Effects | What would each proposal change in value, evidence, risk, effort, and timing? | Impact note and affected sections |

End each proposal with one disposition: accept, defer, reject, or escalate. If accepted, update the baseline, identify who approved it, and mark which earlier assumptions or work products are now stale. Do not erase the original agreement. The difference between the original and current scope is useful evidence about why the work changed.

The record can be lightweight. A one-hour analysis may need five lines in a note, while a multi-team programme may need formal change control. The important behavior is that the boundary and the authority to move it remain visible.

Worked example: a synthetic policy-summary request

Imagine a synthetic operations analyst asked to produce a two-page summary of a new travel policy for regional managers by Friday. The approved evidence is the policy document, finance guidance, and an implementation email. This example is invented and does not describe a Two Prune customer.

The analyst asks an AI assistant to outline the summary. The assistant suggests adding a reimbursement calculator, country-by-country tax guidance, manager training slides, and a comparison with competitor policies. Each idea sounds useful, but none is automatically in scope.

The analyst records the baseline:

The accountable owner accepts the exception table and defers the calculator. The revised scope record names the added table and preserves the calculator as a follow-on decision. The analyst then prompts AI only for work inside the updated boundary and checks every rule against the supplied documents.

This is not rigid refusal. The scope changed once because the proposed exception table improved the agreed outcome at acceptable cost. The calculator was not hidden or discarded. It was separated from the current assignment because its evidence and verification needs were different.

• Statement: a two-page decision aid explaining what managers must do differently from the effective date;
• Constraints: use only the approved internal evidence, distinguish rules from unresolved questions, and deliver by Friday;
• Outside: tax advice, tool development, competitor research, and employee training materials;
• Proposals: finance asks for a short exception table, while one manager asks for a calculator;
• Effects: the exception table can replace a lower-value background paragraph with no deadline change; a reliable calculator needs rules and testing that are not in the evidence pack.

Practice with controlled change cards

Give a learner a bounded brief and ask for a SCOPE record before using AI. During the exercise, introduce three change cards:

Require a written disposition for each card. The learner should update affected assumptions and sections when a change is accepted, retain excluded work, and identify when escalation is necessary. Add a tempting AI-generated idea that has no route to the required outcome. A strong learner will classify it rather than quietly adding it.

The practice artifact should include the original baseline, change log, current baseline, and a short impact trace. This makes scope judgment reviewable instead of relying on a final document that happens to be short.

• a necessary correction to an input;
• an attractive request from someone without decision authority;
• a new requirement from the accountable owner that affects the deadline.

Assess observable scope-control behavior

In an assessment, use a task with a clear deliverable, limited time, several plausible extensions, and at least one legitimate scope change. Observe whether the participant:

Do not score scope control by page count or by the number of requests refused. A concise artifact can omit necessary work, and a longer artifact can remain controlled after an approved change. The evidence is the participant's treatment of boundaries and change, not a preference for doing less.

• states the agreed outcome and exclusions before expanding the task;
• separates source corrections from new requirements and optional ideas;
• identifies who can approve a material change;
• assesses effects on evidence, risk, effort, timing, and existing work;
• updates the baseline and traces stale assumptions after acceptance;
• uses AI inside the current boundary and challenges irrelevant additions;
• preserves deferred ideas without presenting them as completed work;
• delivers an artifact that still answers the approved assignment.

Verify with a baseline-delta review

Ask a reviewer who did not perform the work to compare the original SCOPE record, accepted changes, and final artifact. The reviewer should be able to explain every material addition or removal, identify its authority, and trace it to an impact assessment. Unexplained differences are control failures even when the final prose is polished.

Then inspect whether accepted changes invalidated earlier evidence, calculations, prompts, or conclusions. Re-run the affected checks rather than assuming the new material fits. Confirm that excluded items did not reappear through AI-generated text and that deferred work is not described as if it were complete.

Scope control is task-dependent. It does not establish that every baseline was well chosen or that a person will control scope in another context. In a Two Prune assessment, the baseline, change decisions, AI interactions, revisions, and final artifact can contribute bounded evidence about Context Framing, AI Orchestration, Verification & Risk Control, and Communication & Delivery. They do not make an employment decision.

Sources

NASA procedural requirements for project baselines and rebaselining

• US GAO Schedule Assessment Guide

Sources: [1] [2]