10 Signs Your AI Contractor Is Overcharging You
Everyone is putting "AI" on the invoice now — and marking it up. The pattern is simple: slap an AI label on standard work, wrap it in buzzwords, and charge a premium the client cannot evaluate. The good news is that the tells are consistent. Here are ten, drawn from documented 2026 tactics, plus the single question that defeats most of them.
They build the infrastructure before proving the engine
The classic setup: a big "platform build" comes first, the actual working capability later. By the time you find the core does not deliver, you are financially committed — the money already spent pressures you to keep going.
The fix: demand a small paid pilot that proves the core use case before any infrastructure spend. This is prove-then-pay, the best defense against sunk-cost entrapment.
The proposal says "transformation," not deliverables
"Transformation" is the word consultants reach for when they do not want to commit to specific, measurable deliverables. Count how many times it appears versus how many concrete shipped workflows are named.
The fix: ask for specific shipped workflows with before-and-after numbers. Vague scope is where the padding lives.
Implementation is billed hourly, open-ended
Hourly implementation pricing creates a perverse incentive: the longer the work takes, the more they earn. Senior practitioners charge fixed-fee for implementation — their value is speed of shipping, not hours logged.
The fix: require fixed-fee per shipped workflow. Advisory can be hourly; the build should not.
"Proprietary AI" that is really a wrapper
A great deal of "proprietary AI engine" is a thin layer over a public API you could nearly call yourself. Not always wrong — but it completely changes what a fair price is.
The fix: ask plainly which foundation model or tools it calls. The answer resets the number.
The demo dazzles — in their sandbox
Implementing in your actual business, with your messy real data, is meaningfully harder than a generic demo. The hard part often reappears later as a separate "integration contract."
The fix: make a working integration into your systems the test of whether the pilot succeeded.
The impressive case study has no name and no reference
"A Fortune 500 client saw 300% ROI — but we cannot say who." An unverifiable number is a marketing asset, not evidence. The more impressive and less specific, the more skeptical you should be.
The fix: ask for a reference you can actually call.
The expert who pitched will not be the one building
The senior architect wows you in the sales meeting; delivery quietly goes to juniors learning on your budget. Classic bait-and-switch.
The fix: get the named people from the pitch written into the contract.
You do not own the code, the data, or the exit
"At the end, everything stays on our platform." That is lock-in: leave, and you lose everything — so the monthly fee becomes forever. Some vendors even charge you to export your own data.
The fix: require code and data ownership, plus exit and export terms, in writing on day one.
There is no baseline and no way to measure success
"You will just feel the improvement." Without before-and-after metrics, you can never prove it failed — which is precisely the point. Unmeasurable success is unfalsifiable, and that protects them.
The fix: write baseline and target metrics into the contract.
It has been "90% done" for a month
The eternal 90%: always almost-finished, always one more payment and a few more weeks. This is sunk-cost milking — the last 10% never quite arrives.
The fix: tie every payment to a demonstrated, working milestone — not a percentage they assert.
The through-line: sunk-cost entrapment
Notice the pattern. Almost every tactic above is a version of the same move: get you spending on scaffolding before the core is proven, then use what you have already spent to pull you further in. Economists call it escalation of commitment; at the poker table it is being pot-committed. The defense is always the same — prove-then-pay: a small paid pilot, fixed-fee per shipped workflow, ownership in writing, controls on usage, before you are in deep.
See if you can navigate an AI project without getting fleeced.
We turned these tactics into a free simulation — a vendor works you across 20 turns while you make the calls, and it tracks the money, the red flags, and how pot-committed you get. Then take the free FAIG assessment to see where your AI governance actually stands.