How to Choose an AI Consultant: A Buyer's Checklist
You are choosing between firms that will all claim the same things. The differences that matter are structural: whether they measure a baseline, whether they name a metric, whether they define failure, and whether their fee structure rewards your outcome or their hours. Seven questions separate the real ones from the resellers.
What should you look for in an AI consultant?
Look for someone who asks about your processes before proposing anything, who quotes workflow cost in dollars rather than capability in features, and who offers a diagnostic cheap enough that declining the rest is genuinely easy. A consultant who wants a large commitment before they have costed anything is selling, not advising.
The seven questions to ask before you sign
- 1. What is the annual cost of the workflow you would automate first? They should be able to tell you before proposing a scope.
- 2. What metric should move, and what is its baseline today? No baseline means the work cannot be evaluated.
- 3. What would make you tell me this failed? Anyone who cannot answer has not considered your downside.
- 4. Is the fee fixed, and what is excluded? Fixed fees force scope discipline. Ask specifically about data preparation, which commonly adds 30 to 50 percent.
- 5. Who does the work? In boutique firms the senior person sells and delivers. In larger firms the senior person sells and a junior team delivers.
- 6. What is the cheapest way to start? A real advisor offers a small, useful first step. A reseller pushes the full engagement.
- 7. What happens at handoff? The goal is your team running it. Ask what documentation and training is included.
Red flags
- A proposal with no baseline measurement and no named metric.
- Hourly billing on an open-ended scope, which pays them more when work takes longer.
- A large build proposed before any diagnostic, which means the scope was guessed.
- Guarantees of specific percentage savings without having seen your data.
- A senior partner on the sales call and an unfamiliar team on delivery.
What a good engagement looks like
| Stage | What happens | Price shape | Your protection |
|---|---|---|---|
| Diagnostic | Workflows documented and costed, use cases ranked | Under $2,500 fixed | You can decline everything after it |
| First build | One workflow rebuilt, deployed, instrumented | About $10,000/month | 90-day measured result |
| Expansion | Additional workflows, optimization, tuning | $1,000/month and up | Renewed against documented results |
Three things make that structure work for the buyer: the entry cost is small enough that the decision is low risk, the first build is measured against a baseline recorded before work began, and every later stage is justified by a result rather than a promise.
Frequently asked questions
Frequently Asked Questions
- What should I ask an AI consultant before hiring?
- Ask for the annual cost of the workflow they would automate first, the metric that should move and its baseline, what would constitute failure, whether the fee is fixed, who actually does the work, the cheapest way to start, and what the handoff includes.
- How do I avoid hiring an AI tool reseller?
- A reseller sells licenses and training and describes features. A consultant describes workflow cost in dollars, insists on a baseline, and offers a small first step. If there is no baseline and no named metric, you are talking to a reseller.
- Is a fixed fee or hourly billing better for AI consulting?
- Fixed fee. Hourly billing pays the consultant more when work runs long, which inverts your incentive, and it lets the scope stay undefined. Fixed pricing forces both sides to agree what is being delivered before work starts.
- How cheap should a first AI consulting step be?
- Cheap enough that declining the rest is easy, which in practice means under $2,500 for the diagnostic. Its purpose is to tell you where the money is before you commit to a build. If the entry point is large, you are buying a build, not a diagnosis.