How to tell a useful AI strategy proposal from a sales document

AI strategy proposals look alike. Discovery, interviews, an assessment, a roadmap, a presentation to the steering committee. The differences that decide whether the result is usable are rarely on the front page. They show up in the answers to a few direct questions.

These are the ones worth asking before you sign.

How is it priced, and what does the provider sell next?

Fixed price or day rate. Framing paid by the day rewards more framing. A fixed price agreed before the work starts, scaled to scope, puts the incentive on finishing.

What happens after the document. If the provider resells licences, earns referral fees from a platform, or has a build team waiting for the work, that is not disqualifying, but it is a bias you should know about. Ask directly. Then ask whether the recommendation would change if they were not allowed to bid for the build.

Whether the document works without them. It should be specific enough for another supplier or your own team to execute. A strategy that only works if its author builds it is a sales document.

Who will they talk to?

The people who execute, not only the people who decide. The gap between how a manager describes a process and how the team actually runs it is usually where the opportunity is. A proposal that interviews only leadership will describe the company leadership believes it has.

Separately. Executives and staff in the same room produce the official version. The useful version comes out in separate conversations.

With evidence. Ask whether the provider will work from actual volumes, cycle times and error rates, or from what people estimate in interviews. The two disagree more often than not.

And who exactly will be doing the talking. The people who present the proposal are not always the people who run the interviews and write the document. Ask for the names of those who will, and meet them before you sign. The quality of a strategy depends far more on who asks the questions than on the method printed in the proposal.

What will the document refuse to say?

Does it reject anything? A good strategy is mostly a record of what not to do. Ask how a rejections section looked in previous work — anonymised, but the structure should exist.

Does it include the case against? The strongest argument for doing nothing, or for solving the problem another way. A provider who has never written one has never tested a recommendation.

Can it conclude that the answer is not AI? Ask what happens if the finding is a process fix and a data cleanup. If the answer is hesitant, the conclusion was reached before the work started.

Infographic

Before you sign a strategy proposal

• Fixed priceAgreed before it starts, scaled to scope, not billed by the day.
• Declared interestsYou know what the provider sells after the document, and to whom.
• Executors interviewedNot only leadership, and separately from leadership.
• Evidence, not estimatesActual volumes, cycle times and error rates.
• Written rejectionsWhat was considered, what was rejected, and why.
• The case againstThe strongest reason to do nothing, or to do it differently.
• Buildable without the authorAnother supplier or your own team could start from it.

What a strong proposal sounds like

It is short on method and specific about scope. It names who needs to be involved on your side and for how long — typically a few hours each from two to six people — and what evidence it will need. It gives the duration in weeks, not months; much longer and it stops being framing and becomes a programme. And it describes the output as a decision: a ranked shortlist with owners and numbers, rather than a maturity score and a roadmap. It also says what you get if, halfway through, the problem turns out to be somewhere else — and does not treat that as a change request to be priced separately.

Weak proposals are long on method and vague about output. They promise a framework, a maturity assessment and a list of opportunities. All of it will be accurate, and none of it will tell you what to do on Monday.

A trap on your side

The questions above test the provider. One tests you: do you want the answer, or a particular answer? If the sponsor already knows what they want built and needs a document to support it, any good provider will either decline or write something the sponsor does not like. It is cheaper to notice that before the proposal stage than after the final presentation.

Our own AI strategy engagement is built to pass these questions: a fixed price agreed before it starts, separate interviews with the people who decide and the people who execute, a document another supplier could build from, and a case against for every recommendation. We do not resell licences or platforms, so there is no margin behind the recommendation.