The Short Answer
Time-and-materials pricing is an admission that the work cannot be estimated, which happens when a vendor is building your platform from scratch. On ibl.ai the base already exists: you own all the code and the data, run it model-agnostic across any LLM, and pay with no per-seat pricing — so the engagement scopes integration against your systems rather than open-ended construction, and can deploy anywhere.
Federal acquisition regulation is unusually blunt about time-and-materials contracts, and the commercial market has never repeated what it says.
FAR 16.601 permits a T&M contract only when it is not possible at the time of award to accurately estimate the extent or duration of the work, or to anticipate costs with any reasonable degree of certainty.
Then it explains why that permission is grudging: a time-and-materials contract provides no positive profit incentive to the contractor for cost control or labor efficiency.
Because of that, the regulation requires government surveillance to give reasonable assurance that efficient methods and effective cost controls are being used.
Read the two together and the picture is complete. T&M is a formal statement that nobody can size the work, attached to a structure that pays more the longer the work takes.
What is a time-and-materials contract actually saying?
That the buyer will carry the cost risk, because the seller cannot quantify it either.
That is an honest position. In genuinely exploratory work — novel research, unproven feasibility — it is the only honest position, and pretending otherwise produces a fixed price with a large risk premium buried inside it.
What makes it worth examining in AI specifically is how routinely it appears for work that is not exploratory at all.
An organization wanting an AI assistant over its own documents is not commissioning research. It is commissioning retrieval, evaluation, guardrails, access control, audit logging, and model routing — a well-understood list that thousands of organizations need identically.
The reason that engagement is still priced by the hour is not that the problem is mysterious. It is that the particular vendor has not built it yet.
Why is the estimate unreliable in the first place?
Because the vendor is constructing the platform, and construction of something not yet designed genuinely cannot be scoped.
This is the part worth sitting with. The unestimability is real — FAR is not wrong to treat it as a legitimate basis for T&M. But it is a property of the starting point, not of AI as a technology.
Split any AI engagement into two portions and the difference is immediate. Building permissions-aware retrieval, an evaluation harness, guardrails, RBAC and audit logging is hard to estimate.
Connecting a finished platform to a named list of systems — your SIS, your EHR, your identity provider, your document store — is ordinary integration work that an experienced team scopes routinely.
The first portion is where estimates fail. It is also the portion that is identical across organizations, which is the uncomfortable arithmetic: the industry keeps funding the same infrastructure, one client at a time.
Accenture alone reported cumulative advanced-AI bookings of $11.5 billion through Q1 FY2026 across more than 1,300 clients and 11,000 projects. A substantial share of those projects needed the same foundational layers built again.
What does hourly billing actually reward?
Duration. That is not an accusation; it is arithmetic, and FAR states it plainly.
Most integrators are not exploiting the structure. Good ones work hard to finish. But no amount of professionalism changes which direction the incentive points, and the regulation's response — mandatory surveillance — exists precisely because good faith is not a control.
The contrast with a flat platform licence is stark. A licence plus a bounded integration engagement pays the same whether delivery takes eight weeks or six, so the incentive is to finish and move on.
The practical test for any contract model is simple: what does it reward when nobody is watching? Judge the structure, not the intentions of the people inside it.
Does fixed-price solve it?
It relocates the problem and charges you for the relocation.
Under fixed-price the vendor carries scope risk, which sounds like the answer.
But the vendor cannot estimate the work any better than you can, so the price includes a premium for that uncertainty, and the contract's incentives shift toward defending the scope rather than improving the outcome.
The predictable result is a scope document written before anyone saw how the model performs on real data, and a year of change orders arguing about it.
Both models are pricing the same underlying unknown. Neither removes it. The only thing that removes it is reducing how much remains unbuilt — which is why we set the two structures side by side in Fixed-Price AI Builds vs an Owned Platform.
What changes when the platform already exists?
The unestimable portion disappears, and with it most of the argument about contract type.
A platform already running in production has a known cost, because someone else already incurred it and amortised it across every customer. What remains is integration against systems you can name, which is estimable enough that a fixed price is not a gamble for either party.
That is the shape of the ibl.ai engagement. The platform is in production with 1.6M+ users from 400+ organizations and ships with the full source code under a perpetual licence, so forward-deployed engineers arrive to integrate and extend rather than to construct.
It is also why the work is comparatively fast and cost-effective: the customization is real, and it happens on top of something that already works instead of standing in for the thing that should already have existed.
For public-sector buyers this has a direct consequence.
When a platform can be licensed with source rights, the Determination and Findings asserting that no other contract type is suitable becomes considerably harder to write — which we cover in FAR 16.601 and Time-and-Materials for AI Services.
How should you read a T&M proposal now?
As information about the vendor, not about the difficulty of your problem.
An hourly proposal for well-understood AI capability is telling you that the seller will be building the foundation during your engagement. That may still be the right choice — for genuinely novel requirements it often is — but you should price it knowing what you are funding.
Three questions separate the cases quickly. Which of the capabilities I need do you already have running in production? What proportion of the proposed hours goes to those foundational layers versus to work specific to my organization?
And what do I hold at the end — including whether I can operate and modify it without you?
A vendor with a platform answers all three easily. A vendor without one answers the first with a roadmap.
The arithmetic behind the second question is worked through in our T&M vs owned-platform calculator, and the broader contracting picture in Time & Materials for AI Infrastructure.