Fixed Price or Hourly? How to Buy AI Consulting

Two proposals arrive for the same AI implementation. One quotes a daily rate and an estimated range of fifteen to twenty-five days. The other quotes a single number and a delivery date. The first looks more flexible and more honest about uncertainty. In practice it is usually the more expensive and the more risky of the two, and it is worth understanding exactly why.

What each structure actually does to incentives

Set aside the numbers for a moment and look at what each arrangement rewards.

Under hourly billing, the supplier's revenue rises with the time taken. Nobody reputable pads deliberately, and that is not the issue. The issue is subtler: hourly billing removes the pressure to be efficient. An extra workshop, a more thorough options paper, another round of stakeholder consultation, all of these feel like diligence, all of them are billable, and none of them has to justify itself against a fixed budget. Scope creeps because nothing is pushing back on it.

Under fixed price, the supplier's margin rises as the work is completed faster. This creates its own risk, which is corner-cutting, and it is a real risk you should manage. But it also means the supplier has a direct interest in your decision-making being efficient, in the project not drifting, and in going live rather than continuing to explore. For a category where the dominant failure mode is drift, that alignment is valuable.

Who should carry the estimation risk

Every project has uncertainty. The commercial question is which party absorbs it.

Hourly billing places it on the buyer. You commit to paying for whatever duration the work turns out to require, based on an estimate produced by someone else, for work you cannot assess in detail. That is a poor allocation of risk, because you have neither the information nor the control to manage it.

Fixed price places it on the supplier, who has done this work before, knows where it typically slows down, and can price that in. They will include a margin for uncertainty, and that margin is a real cost to you. It is also the cost of transferring a risk you were otherwise carrying blind.

The general principle is unremarkable in every other kind of procurement: risk should sit with whoever is best placed to understand and control it. Professional services is one of the few categories where the opposite arrangement is treated as normal.

When hourly is the honest structure

Fixed price is not universally correct, and a supplier who insists on it for genuinely unbounded work is either mispricing or planning to cut scope later.

Time and materials is the more honest structure when:

• The work is genuine research. If nobody knows whether the approach will work, there is no defensible fixed number, and any quoted one is either padded heavily or a promise someone will break.
• You want ongoing advisory access rather than a defined deliverable. Retainers and day rates fit this well.
• The scope depends on findings that only emerge during the work, and you genuinely want the freedom to redirect as you learn.
• You are supplementing your own team, where the supplier is providing capacity under your direction rather than delivering an outcome.

Note that a first AI rollout on a commercial provider is none of these. Choosing a provider, buying at the right tier, configuring an admin console, setting spending limits, connecting identity and training staff is well-trodden work with a knowable shape. If a supplier tells you that scope cannot be priced, that is information about the supplier rather than about the work.

Managing the real risk of fixed price

Fixed price does create pressure to finish quickly, and the way that goes wrong is a supplier who technically delivers while leaving you with something thin. Three protections handle this well:

A specific deliverable list. Not "AI implementation" but the actual items: provider selected with written rationale, licences at named tier, admin console configured with retention and sharing set, hard spending limit in place, single sign-on connected, rollout plan document, staff guide. Each one is either present or it is not.

A definition of done a non-technical person can verify. Can you open the billing page and see a hard cap? Can you disable a test account and confirm access is gone? Can a randomly chosen employee complete a named task without help? These are checkable by you, without expertise.

Handover as an explicit deliverable. The setup should be documented and your own people should be able to change settings without calling anyone. An arrangement that only works while the consultant is reachable has not been handed over.

Questions that expose a weak proposal

1. What exactly do we have at the end? A list of artefacts, not a description of activities. Activities are what you get billed for; artefacts are what you keep.
2. What is explicitly out of scope? A supplier who cannot answer this has not thought about scope, which means it will be negotiated later at your expense.
3. Who actually does the work? Particularly relevant with larger firms, where the person in the room and the person on the project are often different people at different levels of experience.
4. Are provider licence fees included? They usually are not, and they should not be hidden. You want them stated separately and sized honestly.
5. What happens if we are slow? Fixed price depends on your availability too. A fair contract says what happens if your side causes the delay.
6. What does handover mean? Documentation, admin access, and someone internal briefed. Get it written into the deliverables.
7. What if we want to stop halfway? Understand the exit position before you need it rather than during a disagreement.

The number is less important than the structure

Buyers tend to compare headline figures across proposals, which is understandable and often misleading. A fixed price that is twenty per cent higher than the midpoint of an hourly estimate is frequently the cheaper outcome, because hourly estimates are systematically optimistic and the overrun lands on you.

Compare on four things instead: what you have at the end, when you have it, who carries the risk of it taking longer, and whether you can run the result yourself afterwards. A proposal that is strong on those four is worth paying more for than one that is merely cheaper per hour.

Frequently asked questions