Choosing an AI consulting partner is consequential. The right partner accelerates your business. The wrong one wastes months and budget while delivering a strategy deck you can’t execute.

Here’s what actually matters when evaluating potential partners.

Have They Shipped?

This is the single most important question. Not “have they advised companies that shipped?” Have they personally written code, made architecture decisions, and deployed AI products to production?

The difference is enormous. Someone who’s shipped knows the difference between demo-quality and production-quality. They know that data is always messier than expected. They know that the last 20% of an AI project takes 80% of the effort.

Ask for specifics. Not “we helped a Fortune 500 company with their AI strategy.” What did they build? What stack? What were the hard problems? What failed the first time?

How Do They Work?

The engagement model tells you a lot. Red flags:

  • Long discovery phases. If they need 8 weeks to “understand your business” before writing a line of code, they’re optimizing for billing, not outcomes.
  • Separate strategy and implementation teams. The people who design the solution should be the people who build it.
  • Deliverables measured in documents. If the primary output is a PDF, you’re buying analysis, not implementation.

Green flags:

  • Embedded with your team. They sit with your engineers, attend your standups, review your PRs.
  • Short iteration cycles. Working software every few weeks, not a big reveal after three months.
  • Knowledge transfer built in. The explicit goal is to make themselves unnecessary.

What’s Their Track Record?

Not their client list — their shipping record. How many engagements resulted in production software? What percentage of projects actually delivered the promised outcome?

Most consulting firms won’t give you these numbers because the numbers aren’t good. That’s informative in itself.

Are They Honest About AI?

The best AI consultants will sometimes tell you not to use AI. If your problem is better solved with a rules engine, they should say so. If your data isn’t ready, they should say so. If the ROI doesn’t justify the investment, they should say so.

Beware of consultants who see AI as the answer to every question. That’s a sign they’re selling a hammer, not solving your problem.

What Does the Engagement Look Like?

Good engagements have:

  • Clear scope and deliverables. You know exactly what you’re getting.
  • Fixed timelines. 4-12 weeks, not “ongoing advisory.”
  • Defined success criteria. Measurable outcomes, agreed upfront.
  • A handoff plan. What happens when they leave? Your team should be able to maintain and extend what was built.

The Decision Framework

Rank potential partners on three dimensions:

  1. Practitioner depth. Have they personally shipped AI products? How many? How recently?
  2. Engagement model. Will they embed with your team and write code? Or deliver documents and leave?
  3. Honesty signal. Are they willing to tell you not to use AI? Will they scope small and deliver fast?

Weight practitioner depth highest. Everything else follows from it. A consultant who’s shipped 20 products will naturally prefer embedded engagements, honest assessments, and measurable outcomes — because that’s what works.