Fixed-Bid or Time-and-Materials for an AI Agent Engagement?

Quick answer

Fixed-bid pricing works when the scope is well-defined and the use case is one the agency has built before; it protects the client's budget but requires the agency to price in a risk buffer. Time-and-materials works better for genuinely novel scope or exploratory phases, since AI agent behavior is harder to estimate precisely than typical software work. Many agencies split the engagement: time-and-materials for discovery, fixed-bid for build once scope is locked.

Pricing an AI agent engagement is harder than pricing a typical web or software project, because the deliverable is not just a feature set. It is a system whose quality depends on how well it performs against real conversations, which are not fully knowable until after some of the work is already built.

When fixed-bid works

Fixed-bid pricing is appropriate when the agency has built a similar agent before (a customer support deflection bot on a known platform, for example) and the client's requirements are genuinely stable. It gives the client budget certainty, which is often a hard requirement for procurement approval. The tradeoff is that the agency must price in a risk buffer for the unknowns that always surface once the agent meets real traffic, similar to the cost-overrun risk covered in /blog/ai-agent-development-cost-2026, and it should be sized against a realistic team allocation like the one covered in /blog/team-sizing-budget-ai-agent-engagement.

When time-and-materials works better

Time-and-materials fits better when the use case is novel for both parties, when the client's requirements are still being discovered, or when the engagement includes an open-ended tuning phase against production data. The risk shifts from the agency's estimate to the client's budget oversight, which means the client needs the internal capacity to actively manage the engagement rather than treating it as a black box, similar to the internal capacity gap discussed in /blog/data-preparation-for-ai-agent-onboarding.

The hybrid structure most agencies land on

A common and defensible structure splits the engagement into two phases. Discovery and requirements gathering, the process of mapping real conversation volume and auditing underlying data before any build commitment, is billed time-and-materials because its entire purpose is reducing scope uncertainty. Once discovery produces a locked scope document, the build phase is bid fixed-price against that document, with an explicit acceptance criteria and a change-order process for anything the client adds afterward.

What the contract needs regardless of model

Whichever pricing model is used, the contract should define acceptance criteria in measurable terms (a target containment rate on a defined test set, not "the client is happy with it"), a cap on included tuning iterations, and a change-order process for scope added after signoff. Clients evaluating multiple agencies should treat the presence or absence of these terms as a signal of agency maturity, the same signal covered in /blog/evaluate-ai-agency-before-signing-contract.

A pricing mistake to avoid on both sides

The most common failure is treating the fixed-bid number as covering unlimited tuning until the client is satisfied. Because subjective satisfaction with agent quality can always be pushed slightly higher with another round of tuning, an open-ended "until it's good" clause turns any fixed bid into an effectively unbounded time-and-materials engagement without the pricing to match. Cap the included iterations explicitly, and price additional rounds as a known, pre-agreed rate.

FAQ

Q: Why is AI agent scope harder to estimate than typical software projects?

Because agent quality depends on iteration against real conversation data, not just feature completeness. A chatbot that passes every scripted demo can still need several rounds of prompt and retrieval tuning once it sees live traffic, and that tuning effort is hard to size precisely up front.

Q: Should discovery be billed separately from build?

Yes, in most cases. Bidding a fixed price for build before discovery is complete forces the agency to either pad the estimate heavily or eat the risk of scope surprises. A separate, smaller time-and-materials or fixed-fee discovery phase produces the requirements the build-phase fixed bid can actually be priced against.

Q: How do agencies protect themselves on a fixed-bid AI engagement?

By defining an explicit acceptance criteria and change-order process in the contract before work starts, and by capping the number of tuning iterations included in the fixed price, with additional iterations billed as change orders.