The 90 Minutes That Decide Whether Your AI Program Ships.

When a Fortune 100 company signs a $40 million AI services contract, they typically congratulate themselves on three things: a competitive price, a strong-looking team named in the proposal, and rigorous reference checks. Six months later, most of those programs are in crisis. The price was fine. The team was wrong. The references answered different questions than the buyer asked.

What separates the programs that survive from the ones that stall is not better procurement, better strategy, or better technology. It is a single ninety-minute meeting that most buyers never put on the calendar — and that most sellers will never volunteer.

This post is about what that meeting is, why it works, and how to ask for it.

The 70 percent problem

McKinsey’s long-running research on enterprise transformations has placed the failure rate of major change programs at around 70 percent for more than a decade. The exact rate varies by program type and methodology, but the conclusion is stable enough for procurement: most large enterprise programs underdeliver, and when they do, the cause is often organisational rather than technical.

The standard responses — more change management, thicker statements of work, executive sponsors named in the contract — help at the margins. None of them addresses what I have come to believe is the single most decisive translation step in the procurement-to-production lifecycle. It happens, or fails to happen, before contract.

The step is a conversation between the buyer’s executive sponsor and the seller’s senior technical leadership. Most buyers, in my experience, never ask for it. The ones who do see outcomes that are noticeably different from those who do not.

What I observed

For most of the last decade, I have been in a senior leadership capacity inside organisations that scope, deliver, and occasionally rescue enterprise AI and cloud programs at scale. From that seat, I watched many large engagements unfold from first conversation to final retrospective.

Customers fell into two groups.

The first group is the one most practitioners already know about. Customers who pushed back on technical specifics during proposal review had better outcomes than customers who delegated those reviews entirely to procurement. This is unsurprising. Engaged technical leadership on the buyer’s side predicts engagement quality. It is in every textbook.

The second group is the interesting one. Among the customers who did engage their senior technical leadership in the buying process, there was a sharp internal split — between those who asked to meet with the seller’s senior technical leadership before signing, and those who did not.

The customers who held the meeting renegotiated something material in the SOW the majority of the time. The renegotiations were almost always sensible:

• An overly aggressive timeline pulled back.
• A named delivery lead replaced because the original was rolling off another engagement.
• A governance requirement added that the SOW had not anticipated.
• A phase-two scope item promoted to phase one because both sides agreed it was load-bearing.

None of these would have surfaced through the account team. The account team’s incentives do not surface them. This is not a criticism of account teams; it is a description of the role they are paid to perform.

The customers who skipped the meeting signed proposals that looked structurally similar to the renegotiated versions — and ran into renegotiation-shaped problems three to nine months later, by which point the cost of changing course had multiplied. Some programs absorbed the correction. Some did not.

Across the engagements I have observed, this single meeting has been the strongest pre-signing predictor of program survival I have encountered. The variance in outcomes between the two sub-groups was wider than the variance produced by any other procurement signal — references, technical specifications, price negotiation, even the specific account team assigned.

Two incentive structures, one missing room

The mechanism is what economists would call an incentive asymmetry, and it is hiding in every services-company proposal you have ever read.

Every services company has two technical voices, and they are paid to want different things.

The account team is paid to close the deal on the terms proposed. Their commission, their pipeline, and their reputation depend on the signature. They are not, structurally, accountable for what happens after signing — that accountability flows to delivery.

The seller’s senior technical leadership has the opposite incentive shape. They are accountable for delivery. They know what the proposal is asking them to deliver. They know, often quite precisely, what the proposal is not asking that will become load-bearing by week three.

In a room with the buyer’s procurement team and the seller’s account team present, the senior technical leadership cannot raise these concerns without endangering the deal. In a quieter room with only the buyer’s technical sponsor, they can. In my experience, they almost always do — when the conversation is framed correctly.

The framing matters. It is not what is wrong with this proposal. It is closer to:

If your team and our team were already two weeks into delivery, and the SOW had not yet been written, what would the engagement actually look like?

This single question reframes the entire conversation from sales to design. Both sides move into joint problem-solving, almost reflexively.

Four asks before you sign

If you are scoping an enterprise AI program — particularly one where compliance, audit, governance, or strategic optionality will matter — your procurement process is almost certainly not designed to surface this conversation. It evaluates proposals on cost, on timeline, on references, and on technical specification. It does not surface the gap between what the proposal says and what the seller’s senior technical leadership would honestly say in a quieter room.

You are responsible for asking. Four asks, specifically:

1. A sixty- to ninety-minute meeting with the seller’s senior technical leadership, before contract. Not a sales engineer. Not a solutions architect named on the proposal. The person on the seller’s side accountable for what gets delivered across the company’s portfolio.
2. No procurement, no sales, in the room. This is the load-bearing constraint. The meeting cannot do what it needs to do with either commercial team present.
3. A design-framed question. The two-weeks-into-delivery framing, or a thoughtful variant.
4. A willingness to renegotiate. If the conversation surfaces a material change — and it usually will — your procurement flow must absorb it. The meeting is worth less than nothing if anything that emerges from it is treated as an attack on a closed proposal.

The seller’s senior technical leadership will almost always accept the meeting if your CIO or CTO requests it personally. The seller will not usually offer it by default. Their account team’s incentives do not point in that direction.

The agenda to send

Send the agenda before the meeting so the room knows it is a design review, not a procurement ambush.

1. First 10 minutes: restate the outcome. Buyer sponsor names the business result, the operating constraint, and the risk that would make the program not worth doing.
2. Next 25 minutes: rewrite the first thirty days. Seller technical leader describes what the team would actually do first if the SOW were not already drafted.
3. Next 25 minutes: name the hidden load-bearing work. Both sides list governance, data, integration, staffing, evaluation, and handoff assumptions that are implicit in the proposal.
4. Next 20 minutes: decide what changes. The room writes down what belongs in phase one, what can move later, who must be named, and what decision must go back to procurement.
5. Final 10 minutes: confirm the artifact. Someone leaves owning a written amendment, risk register, or no-change note.

The output is not a feeling of alignment. The output is a change record.

What “good” sounds like in the meeting

The seller’s senior technical leader should be able to answer questions like these directly:

• Which two assumptions in this SOW would you change if you could?
• Who on your team would you most want for this engagement, and is that who is named?
• What is the riskiest decision we are making with this scope?
• What does your team typically need from us in the first month that proposals do not ask for?

If the answers are candid and specific, the meeting is doing its job.

If the answers are vague, deflective, or read as sales language, that signal is itself useful. The same evasion will appear in delivery.

What this means for sellers

If you are running a services company, a model lab, a platform vendor, or any organisation that sells AI delivery to enterprises, your senior technical leadership is one of the most under-used commercial assets in your company.

The customers most likely to renew, expand, and refer are the ones who had the meeting. The customers who escalate during delivery, fight contract terms after signing, or refuse to renew are disproportionately the ones who did not.

The implication is not to put your CTO in every sales meeting. The role’s bandwidth is limited, and the meeting only does what it does in a constrained format. The implication is to make the meeting a deliberate, formal, well-prepared stage in your enterprise sales process — gated by deal size or strategic importance, with a written framing of what the meeting is and is not for. Treat it as the most decisive ninety minutes of the relationship.

2018 versus 2026

The cost of skipping this meeting is rising.

Procurement-to-production cycle time is shorter in AI than it was in cloud. The translation between proposal language and production reality is foggier, because model capabilities change quarterly and the SOW cannot anticipate them. Governance stakes are higher, audit consequences arrive faster, and the regulatory frame is moving while contracts are being signed.

A program that got the wrong answer in a cloud contract in 2018 had years to absorb the correction. A program that gets the wrong answer in an AI contract in 2026 has months.

The meeting that decided whether a cloud program would survive is the same one that decides whether an AI program will. What has changed is the time available to recover from not having had it.

When to talk to us

A program that gets the wrong answer in an AI contract in 2026 has months to recover, not years. The ninety minutes are the cheapest insurance there is. If you are scoping a program — or running a services business that should be offering the meeting — talk to us before the SOW is signed. The fractional CAIO model is the leadership shape that complements the same caution on the buyer side.

The ninety minutes are the cheapest renegotiation you will ever do. Months in, it is the most expensive.

Request a consultation

References

1. McKinsey & Company, Why do most transformations fail? A conversation with Harry Robinson (2019–2024). mckinsey.com