The AI Efficiency Trap, Buyer's Edition

A Romanian IT services leader posted something unusually honest on LinkedIn this week. His point: most CEE software houses are already caught in what he calls the "AI Efficiency Trap". They sell efficiency but bill time. The better their delivery gets, the less they invoice for the same outcome. A team that ships in half the hours, on a time-and-materials contract, bills half the hours. Productivity up, revenue down.

A reply from the CEO of a listed software group was more measured but reached the same place. The industry grew for twenty years on a straight line between headcount, hours and revenue. AI is cutting that line. His answer: move from selling capacity, to selling capabilities, to selling outcomes.

Both of them are right. Both of them are also writing for other suppliers.

Read the same thread from the buyer's chair and it turns into something else: a map of exactly where your leverage sits for the next 18 months, and a warning about where it goes if you do nothing.

The trap only exists if you stay on T&M

Here is the part the supplier side does not say out loud. The "trap" is a trap for them only as long as you keep buying hours. The moment they move you to fixed price, the trap flips.

Think about how a fixed-price quote gets built. Someone estimates effort. That estimate comes from historical velocity, from the last five projects, from the senior architect's gut. All of that is pre-AI data. If the team now ships the same scope in 60 percent of the hours, a fixed price built on last year's estimate hands the other 40 percent straight to the supplier's margin. You pay 2024 prices for 2026 work and you never see the difference, because nobody is counting hours anymore.

This is why the move to fixed price is being sold to you as a favour. Certainty, risk transfer, no surprises. In reality it is the cleanest way to stop the productivity gain reaching your invoice.

The same logic applies to the advisory firms, with one twist: they rarely even bother moving you off the rate card.

A rate card that did not move

Earlier this year I was running the sourcing for a Big Four advisory engagement on a banking licence application. Crucial work, regulator-facing, no room for a weak firm. The rate card arrived and it was indistinguishable from the one I had seen three years earlier, except that every workstream now had "AI-enabled" in its description.

So I asked the obvious question. Show me, workstream by workstream, what AI does inside your delivery. Which tasks, what share of the hours, and what that does to the staffing mix. Not because I expected a discount on the spot, but because the answer tells you everything about whether the firm has actually changed how it works or has just changed the brochure.

The answer I got was a version of "AI is embedded in our methodology". No number. No pyramid. No change to the analyst-heavy mix that makes these engagements expensive in the first place. Which is itself an answer: the efficiency is real, it just lives in their margin, and nobody is going to volunteer it.

That is not unique to one firm. Clients of the big accounting networks are now openly asking for a share of the AI efficiency, and some are writing double-digit AI efficiency discounts straight into contracts. Analysts tracking the outsourcing market are seeing buyers reopen deals within 24 months of signature rather than wait for renewal, because contracts priced in 2024 were priced before agentic delivery changed the economics. The window is open. It will not stay open.

If the rate will not move, the days must

Reopening a rate card is hard. The firm has defended it for a decade, the partner's leverage model depends on it, and "our seniority costs what it costs" is an argument you can lose for hours. So do not fight there first.

Fight on effort. The supplier's own marketing says AI makes delivery materially faster. Take them at their word. If the same engagement was 40 man-days before AI augmentation, the rate stays flat and the estimate becomes 20. You pay 20 days, not 40. The rate card is their number; the number of days is yours to challenge, line by line, task by task, and every "AI-enabled" label in their proposal is evidence on your side.

This reframes the whole conversation. You are no longer asking for a discount. You are asking them to deliver at the speed they advertise. A firm that insists on 40 days for a workstream it calls AI-enabled is telling you one of two things: the AI claim is marketing, or the gain is being kept. Either answer is useful.

In practice the ask is simple: same rate, half the estimated effort as the opening position, settled somewhere between 50 and 70 percent of the pre-AI estimate depending on how much of the work is commodity. For a client, 40 days at a flat rate versus 20 days at a flat rate is the same 50 percent saving a rate cut would deliver, and it is far easier to get.

Selling capacity, capabilities, outcomes: use their ladder against them

The supplier narrative gives you the roadmap. Each rung is a different negotiation, and you should run all three at once, on different parts of the scope.

Where they still sell capacity (T&M, staff augmentation, run and maintain): stop accepting a flat rate card. Put a productivity ratchet in the contract: a declining blended rate, or a re-baselined hours-per-unit every two quarters, or an annual efficiency credit. BPO buyers have had these clauses for twenty years. Software and advisory buyers somehow forgot to ask.

Where they want to sell capabilities (fixed price per project or feature): refuse any fixed price built on effort estimates you have not seen. Demand the estimate basis, the assumed AI tooling and the velocity it implies. Ask for the staffing pyramid, not the rate card. If the pyramid still shows the 2023 analyst-to-partner ratio, the AI gain is either not real or not yours.

Where they are ready to sell outcomes (SLA-based support, milestone-based regulatory work, per-resolved-ticket): take it, and take it first on the commodity parts. Gap analyses, policy drafting, document review, regression testing, L2 support. Pay the premium only on judgement: regulator relationships, architecture decisions, the call on what is worth building at all.

Six moves before your next renewal

1. Hold the rate, halve the days. Open every estimate at 50 percent of the pre-AI effort and make the supplier justify every day above it. This is the lever that works when the rate card will not move.
   

2. Re-baseline now, not at renewal. If the contract was priced before mid-2025, treat it as mispriced. Open it. Suppliers are still unsure of their own velocity numbers, which is the only moment you will get a straight answer.
   

3. Change the unit of measure. Hours are the supplier's unit. Story points, features, function points, resolved tickets, approved regulatory submissions are yours. Once the unit is output, the AI gain becomes a margin question for them and a benchmark for you.
   

4. Ask for a gainshare instead of a fight. Baseline velocity, split the savings above it. Suppliers resist pure pass-through but sign gainshares, and it aligns them to actually deploy the tooling rather than hide it.
   

5. Write in transparency rights. Disclosure of AI tools used on your codebase or your data, throughput reporting, and the right to audit time logs against delivered output. In a regulated environment this is also an IP, confidentiality and data-residency question, and your compliance team will thank you for asking before the auditor does.
   

6. Benchmark against the boutique. The real threat to a Big Four rate card is not another Big Four. It is the eight-person ex-regulator boutique running the same work on an AI-first stack at a third of the price. Mentioning a competing proposal does more than any question about methodology.

The uncomfortable bit

Most suppliers, in CEE and the GCC alike, will respond to all of this by quietly protecting rates and absorbing the gain. That is rational. It is also temporary. Within two years the market rate for AI-assisted delivery resets, every supplier prices it in, and the gain is gone for buyers who did not move early.

The supplier post that started this ends with a line worth stealing: in the end, nobody will be paid for how much they work, but for the value they create. Fine. Then the buyer's job is to make sure that value is defined, measured and priced in the contract, and not left to the supplier's margin to discover.

The full Procurement Union playbook, "Negotiating Professional and IT Services in the AI Era", covers the clause bank, the re-baselining process, the data request list and the tactics table in detail. Community members can pull it straight from the Companion.

Procurement Union | Community-Powered Procurement Intelligence | procurementunion.com