You can usually tell whether an AI services proposal is built to deliver or to extract inside the first reading. The signal is in the sentences. Below are six that show up in extraction-grade engagements, what each one is doing, and the one question to ask in writing before you sign.
These are real sentences from real decks people have brought us for a second opinion (Class C — observable in the proposal documents themselves; we are describing patterns in the artifacts, not motives).
Tell 1: "We will integrate AI across your organization."
Across, integrate, organization — three abstractions in one sentence, and not a single deliverable. The scope is unbounded, which means the budget is unbounded, which means the success criterion will be defined by the vendor at the end. That is the structural problem.
Ask in writing: Give me the three smallest deliverables in this engagement, each with a one-week falsifier I can check. If the vendor cannot reduce the scope to three concrete artifacts and three failure conditions, the scope is decoration.
Tell 2: "Your team will be enabled through our platform."
Translation: the deliverable is a seat license on a tool the vendor resells, not a system you own. When the contract ends, the work ends with the subscription.
Ask in writing: At the end of the engagement, what runs on infrastructure I control, and what stops working if I cancel every subscription you resell? The honest answer might be "most of it stops" — that is fine as long as it is disclosed before signing, not after.
Tell 3: "Our proprietary methodology accelerates time-to-value."
Proprietary methodology is the part of the proposal you are not allowed to inspect. The "value" is whatever the vendor reports at the end. Both halves of the sentence are unfalsifiable by design.
Ask in writing: Show me the methodology document. Show me last quarter's value report for another client, redacted as needed. A vendor with a real method can show the method. A vendor with a real track record can show a redacted artifact.
Tell 4: "We will optimize the workflow on an ongoing basis."
This is the retainer treadmill, named in the proposal. "Ongoing" has no end condition, and "optimize" has no measurement. The product is the monthly invoice.
Ask in writing: What is the metric, what is its current value, what is the target, and what happens to the retainer when the target is hit? If "the retainer continues regardless" is the honest answer, you now know what you are buying.
Tell 5: "Trust the process — early results may be counterintuitive."
This is the pre-emptive excuse for the period when the engagement does not work. It is being written into the proposal before the work has started, because the vendor expects to need it.
Ask in writing: What is the written falsifier that would end this engagement in week four, and who decides it has fired? If the vendor does not want a falsifier in the contract, you are signing for a year of "trust the process."
Tell 6: "Our consultants embed with your team for cultural alignment."
Embedded consultants for cultural alignment is the longest version of "we will be here every day forever." It sounds collaborative; it is structural lock-in.
Ask in writing: What does the engagement look like in month seven when your consultants are not in the building? Show me the runbook a new hire of mine would follow. If the answer is that the consultants need to stay, the engagement is the consultants — not their work.
What hiring managers are actually paying for
The pattern in these six tells is the same: the vendor is selling presence, not artifacts. Jay Kumar Chimata, who runs a 22k-user AI talent platform, sat down with Themesis on what employers in 2026 are actually willing to pay for versus what candidates are studying: Meet Jay Kumar Chimata: JobFirst.ai and the Real AI Job Market (Class E — primary-source interview with the operator of a hiring marketplace).
Our one-line read, in our own words: managers are paying premiums for people who can ship a working, inspectable artifact and walk a colleague through it — not for people who can sit in a chair on a retainer. The buying side and the hiring side are converging on the same standard, and proposals that fail the six tells above are out of step with what the people writing the checks actually want.
The standing test
If you have a proposal on your desk right now, write the six questions in an email, send them to the vendor, and wait. The reply itself is the test. A vendor who can answer in writing, with specific artifacts and a falsifier, is worth a second meeting. A vendor who answers with another deck is showing you what month seven will look like.
We use the same six questions as our own internal pre-sign checklist. If a scope we are about to write fails any of them, we rewrite the scope.