How AI Is Forcing Engineering Services to Rethink the Way They Price Value?

Forty minutes into a discovery call, the CTO stopped me mid-sentence.

"Your engineers use AI. So why am I still paying for their hours?"

I gave him a reasonable answer. Something about senior oversight, review cycles, and the fact that a model does not carry accountability for what ships. All true. None of it landed. He nodded the way people nod when they have quietly moved on.

Two weeks later, a procurement lead on a different account sent back our draft SOW with one comment. She wanted a clause naming which parts of delivery would be AI-assisted, and what that did to the rate. Not whether we used AI. She assumed that. She wanted it priced.

Same question, opposite ends of the org chart. That shift is something we are seeing directly in enterprise engineering conversations at Cubet, and it is changing how we price work.

The logic that broke

Time and materials worked because everyone accepted a useful fiction. Hours stood in for value. More hours, more work, more output. Nobody loved the model. Everybody could audit it, which mattered more.

AI broke the proxy. Google's 2025 DORA report, drawn from nearly 5,000 technology professionals, put AI adoption among software development professionals at 90 percent, a 14 point jump in a single year, with more than 80 percent reporting productivity gains. My clients read the same research. Once a buyer believes your engineer ships on Wednesday what used to land on Friday, a rate card stops being a price and becomes a position you defend.

Cost had already loosened its grip. Deloitte's Global Outsourcing Survey 2024, polling more than 500 business and technology leaders, found only 34 percent now name cost reduction as their primary outsourcing driver, down from 70 percent in 2020. When cost stops being the headline, hours stop being the unit.

Three things that changed in the cycle

Pricing shows up on the first call. I used to save it. Build the case, prove the capability, talk commercials in week five. That sequence does not survive a buyer who has already decided the pricing model is the qualifying question. So the commercial hypothesis comes into the first conversation now. It gets pulled apart fairly often. Better then than in month three.

The room is fuller and more technical, earlier. Head of platform, security lead, sometimes an internal AI owner, by the second call. They want to know where the models sit, whether our engineers push their code into public tools, and what happens to their data when we part ways. Finance arrives sooner too, because outcome-linked terms are a finance decision before a procurement one. AI compresses the delivery timeline and stretches the buying timeline at once.

Proposals look different. Rate card and CV pages are dead weight. Tech Mahindra recently signed a healthcare engagement priced against roughly 40 percent fewer tickets, 20 percent lower mean time to resolution, and a 30 to 35 percent reduction in technical debt. That is the shape of a current proposal.

What I would say now

I never got to re-answer that CTO. So here is the answer.

He was not really asking about hours. He was asking who carries the risk if the work does not land. Hours put that risk on him, which was tolerable when nobody could tell how much of an estimate was padding. So the answer is not a better defence of the rate card. It is a different structure, and the parts are worth naming.

There is a base fee that covers the cost of delivering. Below that, no vendor can staff the work properly, and a supplier quietly running at a loss is a supplier about to become your problem. Above the base, a portion of the fee rides on an agreed metric, usually between a tenth and a quarter of the total. Enough to be felt on both sides, small enough that nobody is betting the engagement on one number.

We measure at least two things that pull against each other. Release frequency on its own invites a vendor to ship faster and worse. Pair it with defect escape rate and that route closes. A single metric is an invitation to game it, and any buyer who has run one of these will tell you the same.

If a client cannot produce ninety days of clean data, we do not open with outcome terms. We open with a short instrumentation phase priced on time and materials, and the outcome clauses begin once a baseline exists. That is the unglamorous reason we sometimes look slower to commit than the firm down the road. You cannot quote a result you cannot predict, and you cannot predict a result you do not measure.

Access is a contract term, not a favour. You cannot be held to telemetry you are not allowed to see. If a client cannot grant read access to the relevant systems, I would rather say so in week one than discover it in month seven.

All of that only works if a firm can predict its own delivery, which is why the internal work matters more than anything I say on a call. Our on-prem private LLM stack lets us answer the data question without hedging, and standardised internal tooling turns throughput into something we measure rather than something particular engineers happen to be good at. Cubet has been building since 2007 and 78 percent of our client relationships run multi-year. You do not put your own fee at risk as a twelve-week stranger.

What my pipeline is telling me

I set myself an aggressive new-business target this year. The number stays between me and my leadership. The shape of it is the part worth mentioning.

Several of my active enterprise conversations are structured around outcome-based or value-linked terms rather than hours. No names, no sizes, no sectors. I raise it only because eighteen months ago I was the one arguing for those terms. Now they turn up in the buyer's first email.

This is not confined to my accounts. IDC's FutureScape: Worldwide Services 2026 expects 30 percent of all contractual engagements with service providers to be outcome-based by 2029, driven by agentic AI. Cognizant has said 45 percent of its BPO contracts are now signed on outcome-based commercial terms.

Worth being honest about the pace. Most industry revenue is still time and materials or fixed price, and Infosys has said publicly that client interest has not yet turned into volume. This is a direction of travel, not a completed transition.

Where this goes next

Hybrid, mostly. Saurabh Gupta of HfS Research describes the new deal shape as subscription for the platform, consumption for the work, and a performance component on top. Time and materials survives where scope genuinely is not knowable, because pretending otherwise is how services firms get hurt. Outcome and consumption models take the repeatable work, which over time is most of it.

If you are rethinking how you buy engineering delivery this year, the four questions above are where I would start. Send me a message for the full set we run before committing to an outcome, or see how we structure engagements at cubettech.com