7 min read - Fixed Fee vs Time & Materials for AI Projects: Risk Matrix

Someone will ask for a fixed fee.

It usually sounds reasonable: “Just tell me what it costs.” But AI projects break the assumptions that make fixed fee feel safe. Quality depends on evaluation. Data access is messy. Integrations reveal surprises. And “one more workflow” turns into ten.

So the real question is risk allocation: who absorbs uncertainty, and how do you prevent that uncertainty from turning into conflict?

What you'll learn

• A simple risk matrix to choose fixed fee, T&M, or hybrid
• What must be “known” before fixed fee is safe
• How to write scope boundaries that procurement can accept
• Contract guardrails that matter for AI: evaluation, data boundary, and change control

TL;DR

Fixed fee works when scope and dependencies are stable. Time & materials works when the work is exploratory and the true scope will emerge during delivery. For AI projects, most teams land on a hybrid: fixed fee for a short discovery or build sprint, then T&M or a capacity retainer once the backlog and evaluation thresholds are clear. Use a risk matrix based on scope certainty, data readiness, and integration risk.

The risk matrix (pick your commercial model)

Use three questions to decide:

1. Scope certainty: Do we agree on deliverables and a definition of done?
2. Data readiness: Do we have access, permissions, and examples to evaluate quality?
3. Dependency risk: How many teams/systems do we rely on (security, legal, platform, vendors)?

Now map them:

• High scope certainty + high data readiness + low dependencies: fixed fee can work.
• Medium scope certainty or medium data readiness: fixed fee with strong change control, or hybrid.
• Low scope certainty or unknown data boundary: start with T&M or a fixed-fee discovery, not a fixed-fee build.

If you're buying, fixed fee feels safer. If you're delivering, fixed fee without guardrails is how teams burn out. The matrix is your shared language to avoid that fight.

Common scenarios (and the model that usually fits)

If you want to move fast, map your project to a scenario instead of debating contracts abstractly.

This isn’t legal advice. It’s an operating truth: pick the model that matches uncertainty.

When fixed fee is a good idea (and how to make it survivable)

Fixed fee can be great when:

• You are improving a known workflow (not inventing a product).
• You have access to data and the ability to run evaluation early.
• The integration surface area is small or well understood.
• Stakeholders can make decisions quickly.

To make fixed fee survivable, you need three non-negotiables in the SOW:

1. Acceptance criteria that include quality. Not just “feature shipped,” but “meets threshold on the golden set.”
2. Explicit exclusions and a change-control path. If it changes the data boundary or adds a new workflow, it is a change request.
3. Client responsibilities. Access, subject matter time, and decision cadence.

When time & materials is the honest choice

T&M is the right model when the work is genuinely exploratory:

• The organization cannot yet define “done.”
• The data boundary is unclear.
• Evaluation needs to be invented.
• Dependencies (security, procurement, platform) will shape scope as you go.

The trick is making T&M feel safe for the buyer. You do that with:

• A weekly demo cadence
• A visible backlog with estimates
• A cap or checkpoint (“not to exceed” spend before the next decision)
• A short decision log

Copy/paste: the “checkpoint cap” language buyers accept

You don’t need legal magic. You need a clear decision point.

Example language (plain English):

• “Phase 1 is capped at X hours/days. We will run weekly demos and maintain a visible backlog. At the end of Phase 1, you can: (a) stop, (b) continue on T&M, or (c) convert the next phase to fixed fee once scope is clearer.”

This protects the buyer from runaway spend and protects the delivery team from pretending uncertainty doesn’t exist.

The hybrid model most teams actually end up with

If you want the “best of both,” use hybrid:

• Fixed-fee discovery (1 to 2 weeks) to map scope, boundary, and eval.
• Fixed-fee build sprint (2 weeks) to ship a thin slice.
• Retainer or T&M for iteration and maintenance once the backlog is real.

This keeps procurement happy without forcing the delivery team to pretend uncertainty doesn't exist.

Copy/paste: the risk matrix + guardrails you can add to your SOW

Use this in proposals and internal reviews.

Risk matrix inputs:
- Scope certainty: high / medium / low
- Data readiness: high / medium / low
- Dependency risk: high / medium / low

Recommended model:
- Fixed fee / Hybrid / Time & materials

Guardrails:
- Acceptance criteria includes: evaluation threshold + latency target + cost cap
- Change control triggers: new workflow, new data source, new security boundary
- Client responsibilities: access, SME time, decision cadence
- Reporting cadence: weekly demo + change log + risk register

Common failure modes (and how to avoid them)

• Fixed fee without a data boundary becomes an argument about access and scope. Fix it early.
• T&M without checkpoints becomes an argument about “what did we get?” Add demos and caps.
• No evaluation definition becomes an argument about “quality.” Create a golden set.

Negotiation scripts (what to say out loud)

Most negotiations go off the rails because people argue about price instead of uncertainty. These scripts keep it concrete.

If you're the buyer and you want a fixed fee:

• “We can do fixed fee if we agree on the data boundary, acceptance criteria, and change triggers. If those move, it becomes a change request. Can you propose both: a fixed-fee discovery and a fixed-fee build sprint?”

If you're the delivery team and the buyer demands fixed fee:

• “We can price fixed fee for what we can actually define today. The fastest path is a short discovery to lock the boundary and evaluation. Without that, we’d be guessing, and you’d pay for the guess either way.”

If you're stuck on T&M skepticism:

• “Let’s cap the first phase. Weekly demos, a visible backlog, and a ‘not-to-exceed’ checkpoint. If we can’t show progress, you can stop.”

When a retainer is the cleanest answer

Some AI projects don’t fit fixed fee or classic T&M because the “project” never really ends. You ship, then you maintain quality, handle model/provider changes, and keep the workflow evolving.

That’s when a capacity retainer can be the most honest model:

• you sell a fixed amount of delivery capacity
• you define service levels (incidents, evaluation cadence, reporting)
• you keep scope boundaries explicit

It’s often the easiest way to avoid renegotiating every month while still protecting both sides from “unlimited work.”

Price uncertainty honestly

Pricing decisions get easier when you stop trying to “pick the best contract” and start trying to price uncertainty honestly. Use the risk matrix, install guardrails, and make evaluation part of “done.” You'll spend less time fighting about scope and more time shipping. Need help structuring your AI project pricing? Let's talk.