Do these 3 things before closing this tab:
1Fix the driver behind crashes, sound loss and screen glitches2Clear out junk files and repair common Windows errors3Scan for outdated or missing drivers - takes under a minuteScope creep hurts an AI automation project’s margin when new work is accepted without anyone re-pricing, rescheduling or re-testing the original plan. No source we reviewed establishes an average margin loss for commercial AI projects, so any article quoting a universal percentage is guessing. What the evidence does support is the mechanism, and it points to a set of controls you can apply before the first change request arrives.
Why AI projects are unusually easy to over-extend
The UK government’s AI Risk Management Toolkit (September 2026) notes that AI projects may integrate commercial solutions, drive adoption across wide user groups, build models in-house, or support internal and external operations. Those are very different jobs. “Build the AI automation” can therefore mean a model, an integration, a rollout, or an ongoing service, and a client and a vendor can each assume a different one.
As an Amazon Associate I earn from qualifying purchases.
The World Bank’s report on AI in the public sector adds that there is no single project-management approach for every AI project. Processes depend on the type, scope and timeline. A generic template will not fix a vague boundary, so the boundary has to be written down for each engagement.
How scope change turns into lost margin
The arithmetic is simple. Fixed-fee or capped work has a planned margin. Every unpriced addition raises cost while revenue stays flat.
#1 Best Overall
Hypothetical illustration (not data): a project priced at 100 units with a planned cost of 60 has a 40% margin. If unpriced additions add 15% to the cost, cost rises to 69 and margin falls to 31%. Nothing dramatic happened in any single request, which is why the loss tends to go unnoticed.
AI work has extra places for this to happen. GAO’s 2026 review of federal AI acquisitions found that agencies struggle to access technical experts and to understand AI-related costs. It also found that leaving out AI-specific contract terms may raise the risk of unanticipated cost growth and operational problems such as model drift. Officials described difficulty choosing tests for diverse AI systems and a need for robust, continuous evaluation. This is government procurement evidence. It shows which kinds of work get under-scoped, but it is not a commercial cost model.
What the numbers do and do not show
Historical federal IT evidence
GAO’s 2008 survey estimated that about 48% of the federal government’s major IT projects had been rebaselined. Among the reasons given, changes in requirements, objectives or scope were cited for 55% of projects, and changes in funding stream for 44%. Of the rebaselined projects, 51% had been rebaselined at least twice and about 11% four or more times.
Quick wins for a faster PC:
Scan for outdated or missing drivers - takes under a minuteDriver Scan →Repair Windows errors before they cause bigger problemsFix Now →Fix the driver behind crashes, sound loss and screen glitchesFind Drivers →These are 2008 federal IT figures. They are not AI figures, not commercial figures and not margin figures. GAO also noted that rebaselining can be valid when circumstances change, but it can mask cost overruns and schedule delays. That caution is the useful part: a re-approved baseline can hide the fact that the original one slipped.
Rank #3
- book
- A Guide to the Project Management Body of Knowledge (PMBOK Guide) – Seventh Edition and The Standard for Project Management (ENGLISH)
Recent public-sector practice data
The OECD’s Digital Government Outlook 2026 reports that only 10 of 36 OECD countries (28%) report any financial or non-financial impact measurement of AI use cases in government. Fourteen of 36 (39%) require pre-deployment risk assessments, and 11 of 36 (31%) conduct post-deployment audits. These describe national governments, not company project economics. They do show that “AI implemented” is often treated as the finish line, without measuring whether it delivered value.
Practitioner guidance
The PMI and NASSCOM CoE playbook on data science and AI projects drew on interviews and surveys of leaders from 25 organizations. It argues that these projects need tailored, fit-for-purpose management. It is not a source for a failure rate.
Rank #4
- Harvard Business Review Project Management Handbook: How to Launch, Lead, and Sponsor Successful Projects
- Harvard Business Review Press
- BLANK BOOK
Write a scope baseline that can be defended
The World Bank report says: “Project managers help mitigate risk and counteract scope creep by coordinating and elucidating the requirements and steps necessary for projects during the planning phase.” The checklist below is a practical synthesis of that guidance and the GAO and UK material. It is not a legal standard or a contract template.
- Process and task: the specific task being automated, and what stays human-led.
- Systems and people: workflows, user groups, systems, data sources, integrations and environments that are included.
- Acceptance criteria: measurable thresholds and who signs them off.
- Lifecycle work: whether data preparation, security and privacy review, vendor or model evaluation, testing, deployment, training, adoption, monitoring and maintenance are in or out.
- Exclusions and assumptions: client dependencies, required access, data quality expectations.
- Change rule: how a request is assessed against price, schedule, quality and risk, or traded against existing work.
GAO’s accountability framework groups AI oversight under four lenses: governance, data, performance and monitoring. Use them as a quick audit of the baseline. If you cannot say who owns the data, what “good performance” means, or who watches the system after launch, those gaps are where unpriced work will appear. GAO lists clear goals and stakeholder engagement among its governance practices.
Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.Handle each change request the same way
- Describe the request in one or two sentences, and name who asked for it.
- State its intended value so you can tell a real need from a convenience.
- Estimate the effect on cost, schedule, testing and risk. New data sources and new user groups usually add test and evaluation work as well as build work.
- Offer options: approve with a price or date change, swap for existing scope, defer to a later phase, or decline.
- Record the decision and keep the original baseline visible next to any new one.
The last step answers GAO’s warning about rebaselining. It lets you tell a justified change in circumstances from silent additions, and it leaves a record of why the targets moved.
Three delivery choices that shape creep risk
The sources do not rank these options or give outcome rates. They are axes for comparing approaches, and the second is an editorial reading of the UK toolkit’s project categories.
| Choice | Option A | Option B | What to compare |
|---|---|---|---|
| Build or integrate | In-house model development | Integrating a commercial AI service | Control, integration burden, evaluation needs, ongoing obligations |
| Pilot or rollout | Limited workflow and user scope | Broad adoption | Evidence gained, change-management effort, exposure to more systems and user dependencies |
| Baseline handling | Bounded deliverables with change approval | Open-ended iteration | Predictability, flexibility, visibility of price and schedule trade-offs |
Define “done” by outcome, not by deployment
Because ongoing evaluation and monitoring are part of AI delivery, decide up front who is responsible after launch and for how long. Tie acceptance to measured impact where you can, such as time saved, error rates or cost per transaction. Otherwise “make it better” requests keep arriving after the agreed work is complete, and nobody can say whether the system has met its goal.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
