Negotiating the AI Purchase: Getting the Promise Into the Contract
Most AI spend is not a technology decision, it is a commercial one, and it is being made without the commercial rigour any other purchase this size would get. Seats bought for everyone and used by few, consumption pricing nobody modelled, capability claims nobody wrote down. Here is how to negotiate an enterprise AI purchase properly, and why the single most valuable thing you can do is get the vendor's promise into the contract. Written by people who sold these deals from the vendor side.
AI is being bought in a hurry, under pressure to be seen to be doing something, and that urgency is where the money leaks. The technology conversation gets all the attention, the boards and the pilots and the use cases, while the commercial conversation, the part that decides what this actually costs and whether the vendor is on the hook for any of what they promised, gets waved through. Treat the AI purchase as a technology project and you will overpay and under protect yourself. Treat it as what it is, a large commercial commitment with unusually soft claims attached, and the whole thing changes.
C4C is an independent, vendor neutral advisory firm that negotiates the AI purchase on the buyer's side: the right technology, on the right commercials, with the vendor's promises actually written into the contract. We are not here to deploy AI or to sell you a model. We spent years inside the major vendors selling exactly these deals, so we know which claims survive contact with the paperwork and which quietly evaporate after signature. We use that to get our clients a fair price and enforceable commitments, with nothing of our own to sell.
The AI purchase is a commercial event, not a technology project
Every other purchase of this scale in the business goes through procurement rigour: a competitive process, a modelled cost over the term, defined outcomes, remedies if the supplier misses. AI spend routinely skips all of it, because it arrives dressed as innovation rather than as a contract, and innovation gets a pass that a data centre refresh never would. The result is that organisations are signing multi year commitments on the strength of a compelling demo and a confident roadmap, with almost none of the protection they would demand anywhere else. The technology may well be excellent. That is not the question. The question is what you are paying, what you are actually getting, and what happens when the reality falls short of the pitch.
Where AI spend quietly overruns
The overspend is rarely a single bad decision. It is a handful of soft assumptions that all resolve in the vendor's favour. The common ones:
- Seats for everyone, value for few. AI assistants are frequently bought at full price for the whole workforce on the theory that everyone will use them. Real adoption is usually concentrated in a fraction of users. Paying a full seat price for fractional usage across thousands of people is one of the largest and quietest sources of waste in AI spend today.
- Consumption pricing nobody modelled. Usage based AI pricing looks cheap at the pilot and can scale in ways finance never forecast. Without a modelled ceiling and clear visibility, a consumption line can run well past anything that was approved, and the first anyone hears of it is the invoice.
- AI uplifts folded into bundles. An AI capability added into a wider suite renewal, where it cannot easily be priced separately or removed, is a classic way to raise the floor of a deal permanently. If you cannot see what the AI element costs on its own, you cannot judge whether it is worth it, which is precisely the point.
- Capability claims nobody verified. The demo shows the best case on curated data. The productivity gain is asserted, not evidenced. Unless the claim is tested against your data and your workflows before you commit, you are buying a promise, and a promise with no measurement attached is a promise with no accountability.
Contracting the promise
This is the part almost nobody does, and it is the most valuable thing in the whole negotiation. In the room, AI vendors promise a great deal: adoption levels, productivity gains, specific outcomes, a roadmap of features arriving soon. Almost none of it survives into the contract unless someone deliberately puts it there. The paperwork commits you to pay. It rarely commits the vendor to deliver. Closing that gap is what we mean by contracting the promise, and it has three parts.
- Measurable commitments. Turn the pitch into numbers in the contract: the adoption the vendor says you will reach, the outcomes they say you will see, the capabilities they say are coming and when. If a claim is real, the vendor can stand behind it in writing. If they will not, that tells you what the claim was worth.
- Remedies for a miss. A commitment with no consequence is decoration. Tie the misses to something that matters: service credits, price reductions, the right to hand back seats, a renegotiation trigger. The remedy is what turns a promise into an obligation.
- Exit rights that keep leverage alive. The moment you sign a multi year AI deal with no way out, your leverage is gone until the very end of the term, exactly when the vendor knows you are stuck. Build in review points, off ramps and the ability to scale down, so the relationship stays honest after signature and not just before it.
Vendors promise deliverables in the room. Almost none of them survive into the paperwork unless someone who knows the game puts them there. That is the single message to carry into any AI negotiation.
What to fix before you sign
Before an AI contract goes through, get honest answers to these, and get the answers into the document rather than the meeting notes.
- What is the realistic adoption across this population, and are we paying for seats or for use?
- If pricing is consumption based, what is the modelled cost at expected volume, and where is the ceiling?
- What does the AI element cost on its own, separate from the bundle it is folded into?
- Which of the capabilities we were shown are available now, and which are roadmap, and what happens if the roadmap slips?
- What can we measure, what is the vendor committing to, and what is the remedy if they miss it?
- How do we scale down or exit, and at what points, without waiting for the end of the term?
There is also a risk dimension that runs alongside the purchase, because AI tools change where your sensitive data goes. That is a governance question in its own right, and we cover it separately in Securing Enterprise AI. Here the point is narrower and commercial: buy the capability on terms that hold.
How C4C helps
We negotiate the AI purchase from the buyer's side, using knowledge built on the vendor's. We spent years selling enterprise technology deals of exactly this shape, so we know how AI is packaged, where the pricing bends, which claims a vendor will actually commit to in writing and which they will only ever say out loud. We model what you are really likely to spend rather than what the pilot suggested, right size what you are buying to what will genuinely be used, and turn the promises made in the room into commitments, remedies and exit rights in the contract. We do it without wrecking the vendor relationship, because you will work with them for years, and independence is not the same as hostility. The best time to bring us in is before you commit, while you still have the leverage that signature takes away.
About to sign an AI deal, or renew one?
Send us what is on the table and we will give you an independent view: what it should really cost, what you are actually getting, and how to get the vendor's promises into the contract with remedies and a way out. Independent, with nothing of our own to sell. We sold these deals from the vendor side for years.
Prefer email? Reach us directly at hello@c4cgroup.co.uk.
Frequently asked questions
How do you negotiate an enterprise AI purchase?
Treat it as a commercial commitment, not a technology project. That means modelling the real cost over the term rather than trusting the pilot, right sizing what you buy to what will actually be used, keeping a credible alternative in play for leverage, and turning the vendor's spoken promises into written commitments with remedies. The technology conversation gets all the attention, but the price and the protection are decided in the commercial one, and that is the one usually left unmanaged.
Why do AI licences and Copilot deals often cost more than expected?
Because of a few soft assumptions that all favour the vendor. Seats are bought for everyone at full price when real adoption is concentrated in a fraction of users. Consumption pricing looks cheap at the pilot and scales past what finance forecast. AI uplifts get folded into bundles where they cannot be priced or removed. And capability claims are asserted rather than tested. Each is quiet on its own, and together they can make an AI deal far more expensive than the headline suggested.
What does contracting the promise mean for AI vendors?
It means turning what the vendor promised in the room into obligations in the contract. AI vendors pitch adoption levels, productivity gains and a roadmap, and almost none of it survives into the paperwork unless someone deliberately puts it there. Contracting the promise has three parts: measurable commitments for what was claimed, remedies if the vendor misses, and exit rights that keep your leverage alive after signature. Without it, the contract commits you to pay while committing the vendor to very little.
Should we buy Copilot or an AI assistant for everyone?
Rarely at full price on day one. Adoption of AI assistants tends to concentrate in a fraction of users, so buying a full seat for the entire workforce usually means paying for a lot of use that never happens. A better approach is to size the initial purchase to realistic adoption, measure actual usage, and expand on evidence, with pricing that lets you scale down as well as up. Paying for seats is not the same as paying for value.
How do you avoid overpaying for AI consumption pricing?
Model it before you sign, not after the invoice. Get the vendor to price the expected volume, agree a ceiling or a clear alert and cap mechanism, and make sure you have real time visibility of consumption rather than discovering it at the end of the month. Consumption pricing is not bad in itself, but it shifts the cost risk onto you, so it needs a modelled forecast and a hard limit, both written into the contract, before it is safe to accept.
What should be in an enterprise AI contract?
Beyond price, it should contain the vendor's promises made real: measurable adoption or outcome commitments, remedies such as service credits or the right to hand back seats if they miss, clarity on which capabilities are live versus roadmap and what happens if the roadmap slips, a modelled ceiling on any consumption pricing, and exit or scale down rights at defined points rather than only at the end of the term. If those are absent, the contract protects the vendor and not you.