Back to Blog
A four-week selection runbook for manufacturers choosing an AI consulting partner
November 10, 2026 6 min read Shayntech Engineering

How to Choose an AI Consulting Partner: What Manufacturers Should Know

Most manufacturers choose an AI consulting partner the way they choose a machine: three demonstrations, the most impressive one wins, then the price gets negotiated. That works for equipment: a machine either produces the part or it does not. An AI engagement is different. The deliverable is a change in how work gets done, and the evidence arrives weeks after signing.

Why the Selection Slips

The failure is rarely a bad partner. It is a vague brief. When nobody writes down which process is being fixed, what the current cycle time is, and who owns the outcome afterwards, every proposal looks interchangeable and the deciding factor becomes presentation quality.

The process below is the one we walk clients through before any work is quoted. It takes four weeks of calendar time, most of it waiting, and costs far less than a pilot that was never going to land.

Week One: Write the Brief Before You Talk to Anyone

One page, four blanks. That page is the strongest filter you have, and it must exist before the first call.

  • Process: the specific step that costs money — quoting, glass takeoff, dispatch scheduling — not a department name.
  • Volume: how often it happens each week, and how long one instance takes today.
  • Owner: the person accountable for the result after the consultants leave.
  • Definition of done: the number that must move, by how much, within what period.

Send that page to every candidate before the first meeting. A partner who cannot restate your brief in their own words has not read it. A partner who restates it more sharply than you wrote it has shown you how they think.

Week Two: Screen for Fit, Not for Slideware

Two calls are enough to cut four candidates down to two. Ask each one the same four things: name a factory of similar size and process that you have worked with, name the engineer who will actually be on site, describe the first two weeks in detail, and state what our team will own at the end. Vague answers to the third question are the most reliable warning sign.

Ask also who does the data work. On a manufacturing site the integration is usually the whole job: a fifteen-year-old ERP, a planning sheet one person maintains, and a quality log in a shared folder. A partner who assumes somebody else will handle that will discover otherwise in week six, at your expense.

Week Three: Buy a Paid Scoping Workshop

Free discovery calls are sales meetings with a technical accent. A paid scoping workshop, usually one to three days on site, is the cheapest way to see how a partner works before committing to a project.

Demand three outputs: a map of the current process as it runs, a ranked list of where the hours are lost, and a fixed-price pilot proposal with a written acceptance test. If the workshop ends with a slide deck and a large number instead, you have bought useful information cheaply and should stop there.

W3

What the workshop should cost

Expect a fixed fee in the low thousands for three days, credited against the pilot if you proceed. A partner who will not credit it is pricing the workshop as profit.

Week Four: Compare Proposals on the Same Four Numbers

Insist that every proposal states the same four things, in the same units, so they can be read side by side.

  • Acceptance test: the measurable condition that ends the pilot, agreed before it starts.
  • Pilot price: one fixed number, not a range with assumptions attached.
  • Time to first live result: stated in weeks, not in phases.
  • Exit terms: what you keep, and what you owe, if you stop.

A partner who will not put an acceptance test in writing is selling effort rather than outcome. Effort is hard to audit, impossible to walk away from cleanly, and always renewable.

Weeks Five to Ten: Buy a Pilot, Not a Platform

A pilot runs on real work, inside one process, with two or three users, for a fixed number of weeks. Keep it small enough to stop without a board decision and real enough that the output reaches a customer.

  1. Week five: connect one data source and run the first live job with the partner watching.
  2. Week six: put it in front of the real users and log every override they make.
  3. Weeks seven and eight: fix what the overrides reveal; the override rate is the honest quality score.
  4. Weeks nine and ten: measure the agreed number against the week-one baseline, then decide.

Run the pilot on work you were going to do anyway. A parallel test dataset shows how a model behaves with clean data, not how it behaves on a Tuesday when the dispatch sheet arrives late.

What Belongs in the Contract

Three clauses matter more than the rate card, and all three are routinely missing from the first draft a manufacturer receives.

  • Data ownership: operational data, and anything trained on it, stays yours and is exportable.
  • Knowledge transfer: named documents and a handover session, priced into the engagement rather than offered as a favour.
  • Exit: a defined way to stop, with configurations handed over and no dependency on a subscription you cannot cancel.

The exit clause is the one most factories skip and later regret. If the only person who can adjust the prompts is on the consultant payroll, you have bought a subscription wrapped in a project invoice.

The Signals That You Picked Wrong

Most bad selections become visible in the first month, before the invoice for phase two. The partner asks for more data than the brief implied. The weekly update turns into a status call with no numbers in it. The acceptance test migrates from a cycle time to a successful deployment. The engineer who ran the workshop is replaced by a project manager who has never seen your floor. None of that is fraud. All of it is an early exit, and the contract from week four is what makes the exit possible.

What the Four Weeks Actually Buy

A disciplined selection costs a fraction of a failed engagement and produces three things that cannot be bought afterwards: a brief your own team believes in, a shortlist you can defend to the board, and a partner who already knows the number they will be measured on. Factories rarely fail at AI because the technology was wrong. They fail because nobody wrote down what right looked like before the work began.

Key takeaways

  • Write the one-page brief before the first call; it does most of the filtering.
  • Pay for a scoping workshop and expect a process map, a waste ranking and a fixed-price pilot.
  • Compare proposals on acceptance test, pilot price, weeks to value and exit terms.
  • Contract data ownership, handover and exit before the pilot starts.

Ready to transform your workflow?

Book a free 15-minute demo and see how AI Consulting works for your business.

Book a Free Demo