How to choose an AI consultant (even if you don't pick us)

You are shopping a shortlist. Three or four firms, all confident, all using the same words: enablement, transformation, partnership, outcomes. The decks look similar. The references are all glowing. And you have to pick one to spend real money with.

This guide is meant to be useful even if you do not pick us. We would rather you choose the right partner than the wrong one and conclude AI does not work. So here is how we would evaluate the field if we were in your seat.

The three categories of provider

AI service firms are not interchangeable. They cluster into three types, and the most common buying mistake is hiring one type when you needed another.

Advisors sell thinking. Strategy, roadmaps, vendor selection, maturity assessments. The deliverable is a recommendation, and a good advisor sharpens your decisions. The trap: a recommendation is not a result, and advisory engagements end before anything is built.

Builders sell software. They integrate models, ship the system, make the thing work. The deliverable is functioning technology. The trap: a working system aimed at the wrong outcome, or one nobody adopts, produces nothing, and builders are usually not accountable for adoption or revenue.

Outcome owners sell results. They include strategy and build, but they are accountable for the metric moving, which means they also own the go-to-market application and the adoption work. The trap here is the opposite: these engagements are more involved and cost more up front, and they are overkill if you genuinely just need a second opinion.

None of these is better in the abstract. The skill is matching the category to your actual need. If you have a capable internal team and just need direction, an advisor fits. If you know exactly what to build and have nobody to build it, a builder fits. If you have tried and stalled at the gap between capability and revenue, you need an outcome owner.

Ten questions to ask every vendor

Run every shortlisted firm through these. The pattern of answers tells you more than any single response.

1. What specific outcome are you accountable for? Listen for an outcome, not a deliverable. "A strategy" and "a working integration" are deliverables. "Cost per qualified lead down 20%" is an outcome.
2. How will you measure success against our current baseline? If they cannot describe how they will establish your starting numbers, they cannot prove they helped.
3. Who, by name, will actually do the work? Agencies that sell with seniors and deliver with juniors are common. Ask who is on your account.
4. What will we own at the end? You should own your data, your integrations, and the understanding of how it all works. Black boxes lock you in.
5. How do you handle data governance and privacy? For Canadian buyers especially, ask specifically about PIPEDA and Quebec's Law 25. A blank stare here is disqualifying.
6. What will you not do? Firms that claim to do everything usually do nothing well. A good answer here signals focus and honesty.
7. Walk me through a comparable engagement, including what went wrong. Everyone has a war story. A vendor who claims none is either inexperienced or not telling you the truth.
8. What does the first 30 days look like? You want to hear "ship something small and real," not "discovery phase one of three."
9. What is the smallest version of this we could start with? Willingness to start small is a sign of confidence; insistence on a big up-front commitment is a sign they need the lock-in.
10. What happens if it does not work? How they answer this, with defensiveness or with a clear-eyed plan, tells you who you are dealing with.

Red flags

Some signals are reliable enough to weight heavily:

• Deliverables defined as documents. If the contract's deliverables are all decks and reports, you are buying advice priced like outcomes.
• No baseline measurement. A partner who will not measure where you start cannot honestly claim where you ended up.
• Transformation in weeks. Real AI enablement ships fast in small pieces, but "transform your business in 30 days" is a sales line, not a plan.
• No named team. Reluctance to tell you who does the work usually means the answer would not impress you.
• Percentage-only case studies. "We lifted conversion 300%!" with no starting number could mean one sale became four. Always ask for the baseline.
• Proprietary black boxes. Tools you cannot understand or own are leverage for the vendor, not value for you.

What a good engagement looks like

The shape of a good engagement is consistent regardless of provider type:

It starts small, with a scoped pilot rather than a multi-quarter commitment. It ships something real fast, with usable output inside about a month instead of after a long discovery phase. It measures against a baseline you both agreed on up front. It leaves you owning the system and understanding it. And it builds trust before scale by proving the model on one use case before expanding.

If a provider is comfortable working this way, they are confident in their work. If they need a big up-front commitment and a long runway before any result, ask why.

A note on honesty

The reason we wrote a guide that helps you evaluate competitors is that the AI services market is full of overselling right now, and overselling poisons the well. Buyers who get burned by an oversold engagement conclude that AI does not work for their business. That is wrong, and bad for everyone doing the work properly.

So use this guide on us too. Ask us the ten questions. If our answers do not hold up against another firm's, hire the other firm.

If you are still defining the category itself, start with what is AI enablement.