CFO Review of Fixed-Price AI Consulting Contracts
On this page
Article Summary
A fixed-price AI consulting contract needs a CFO review before signing because testing, retries and post-launch support can erase the margin while the fee stays fixed. In the hypothetical case, a $120,000 project with $78,000 of direct cost earns $42,000, or 35%, but $25,000 of data cleanup, repeated evaluation and acceptance delay cuts contribution to $17,000, about 14.2%. Price the accepted outcome and post-launch obligations, separate usage from the build fee, sell paid discovery for uncertain scope, and tie cash milestones to controllable evidence.
A $120,000 AI implementation contract can look like a 35% margin project until testing, retries, and post-launch support add another $25,000. The client still owes $120,000. Your delivery promise hasn't changed, but the margin has.
For a $1M–$10M consultancy selling AI projects, the CFO review should happen before the statement of work is signed. The question is how much variable delivery the fixed price actually includes.
Price the accepted outcome, the conditions needed to deliver it, and the obligations after launch. A technically impressive prototype doesn't establish any of those costs.
Follow the work past the first demonstration
A delivery budget should include data preparation, integration, evaluation, human review, deployment, documentation, and the agreed support period. Model paid usage during development and testing as well as production.
NIST's Generative AI Profile identifies risks specific to generative AI and proposes risk-management actions. It does not price a consulting project. Its practical implication for this model is that evaluation and oversight need explicit delivery resources.
A client may expect an “AI assistant” to handle many kinds of work. The contract needs narrower acceptance criteria: supported inputs, output requirements, agreed test cases, dependencies, and the process for resolving failed tests.
Finance should not invent the technical threshold. It should require the delivery lead to estimate the cost of meeting it.
A margin bridge before signing
This hypothetical $120,000 project has the following delivery estimate:
| Direct project cost | Base case |
|---|---|
| Engineering and integration labor | $42,000 |
| Data preparation | $8,000 |
| Model and infrastructure use during build | $5,000 |
| Evaluation and human review | $9,000 |
| Deployment and documentation | $6,000 |
| Defined post-launch support | $8,000 |
| Total direct project cost | $78,000 |
Project gross contribution is $42,000, or 35%, before sales expense, shared overhead, interest, and tax.
Now assume poor source data adds $10,000 of labor, repeated evaluation adds $7,000, and a delayed acceptance cycle adds $8,000 of support. Direct cost rises to $103,000. Contribution falls to $17,000, or about 14.2%.
A contract that appeared healthy under the base case has little room for another failure. The response may be a higher price, narrower scope, paid discovery, different acceptance conditions, or a variable usage component.
Don't bury usage inside an unlimited promise
Separate the build fee from ongoing consumption where the commercial arrangement permits it. Specify who owns the production provider account, who pays usage charges, and what is included in any recurring support fee.
If the consultancy carries the usage bill, model completed tasks, calls per task, retries, context size, and the cost of human escalation. Use current supplier rates and a credible high-use case. Don't assume every customer task costs the blended average from a demonstration.
A consumption ceiling should include an action: notification, customer approval, or a commercial adjustment consistent with the agreement. A threshold that nobody enforces is just a spreadsheet note.
If fixed-price work is growing but project margin is falling, a free 20-minute Profit & Tax Leak Check can help identify which delivery assumption to examine first. Rough project figures are enough; no documents are required.
Charge for uncertainty while it is still uncertain
Paid discovery can turn an unbounded promise into a priced scope. The useful output is evidence: data condition, integration constraints, representative evaluations, delivery estimate, and acceptance criteria.
Do not sell discovery as a guaranteed successful implementation. The client is paying to reduce uncertainty, which can legitimately lead to a decision not to proceed.
For the hypothetical project, a discovery result showing poor source data should change the implementation quote before work starts. Quietly absorbing the cleanup cost after signature turns discovery into theater.
When a client declines discovery, decide explicitly whether to price a larger risk allowance or reduce the guaranteed scope. Technical uncertainty doesn't disappear because procurement wants one fixed number.
Cash milestones should follow controllable evidence
A profitable contract can still need funding if most payment waits until final acceptance. Model the actual deposit, progress billings, payment terms, and dispute conditions.
Avoid assuming that delivering a demonstration triggers cash if the agreement requires a separate written acceptance. Identify client dependencies, including access, data, reviewers, and response dates, and have counsel validate how delay affects the obligations.
Use the project-pricing framework for the broader commercial structure. The scope-creep model helps track additional work. This AI-specific review adds variable processing, evaluation effort, and post-launch behavior to those existing controls.
Ask for a contract decision
A fractional CFO review should produce a base and downside margin, a dated funding need, and the conditions that require a revised quote. The delivery lead validates effort. Counsel validates enforceability. The owner approves the remaining commercial risk.
For consulting firms, repeatable profit begins with repeatable boundaries. Take the next fixed-price AI proposal, list the obligations that continue after the demonstration, and bring that estimate to a Profit & Tax Leak Check before committing the team.
Frequently asked questions
What should a fixed-price AI delivery budget include?
Include engineering and integration labor, data preparation, paid model and infrastructure use during build and testing, evaluation, human review, deployment, documentation, and the defined post-launch support period. A technically impressive prototype does not establish any of those costs, so the delivery lead should estimate each one.
What is the base margin on the fixed-price AI project example?
A $120,000 fee less $78,000 of direct project cost leaves $42,000 of contribution, or 35%. That figure is before sales expense, shared overhead, interest, and tax, so it is a project gross contribution rather than operating profit for the consultancy.
What happens to AI project margin in the downside example?
Poor source data adds $10,000 of labor, repeated evaluation adds $7,000, and delayed acceptance adds $8,000 of support, a $25,000 increase. Direct cost becomes $103,000 and contribution falls to $17,000, about 14.2%, leaving little room for another failure while the client still owes $120,000.
Should AI production usage be included in a fixed-price contract without a limit?
Only if the business has deliberately priced and accepted the exposure. Separate the build fee from ongoing consumption where possible, and specify provider-account ownership, who pays usage charges, what recurring support includes, and the action at an agreed consumption ceiling, such as notification or customer approval.
What should paid discovery deliver on an AI consulting project?
Evidence about data condition, integration constraints, representative evaluations, a delivery estimate, and acceptance criteria. Discovery is not a guaranteed successful implementation; it may properly lead to a decision not to proceed, and poor data findings should change the implementation quote before work starts.
Does a successful AI prototype establish final delivery cost?
No. A demonstration may omit difficult inputs, human review effort, repeated evaluations, production usage, and ongoing support obligations. Price the accepted outcome, the conditions needed to deliver it, and the obligations after launch rather than the cost of the first demonstration.
Why do AI project acceptance terms affect cash flow?
Final payment may depend on separate written acceptance rather than a demonstration. Client dependencies such as access, data, reviewers, and response dates can delay approval and extend the period the consultancy funds delivery, so model the actual deposit, progress billings, payment terms, and dispute conditions.
What should a CFO review of a fixed-price AI contract approve?
A base and downside margin, a dated funding need, and the conditions that require a revised quote. The delivery lead validates effort, counsel validates enforceability, and the owner approves the remaining commercial risk before the statement of work is signed.