Every agency added "AI" to its homepage in the last two years, which makes hiring genuinely hard: from the outside, a firm that ships production AI systems and a firm that resells hype look identical. The websites use the same words.
The good news is that the difference shows up quickly under direct questioning. Here are the seven questions we'd ask any AI consulting firm — including us — before signing anything.
1. What have you actually shipped to production?
Not piloted. Not demoed. Shipped, and still running. The gap between an impressive demo and a production system is where most AI projects die — we've written about why AI demos fail in production — so evidence of systems that survived contact with real users and real data is the single strongest signal you can get.
Listen for specifics: what the system does, what it replaced, what broke along the way, and what it costs to run. Vague case studies ("we helped a client leverage AI to transform operations") are a red flag dressed as an answer.
2. Will the people selling me also build it?
The classic consulting failure mode: a senior partner runs the sales conversations, then delivery lands with a rotating cast of juniors. In AI work this is worse than usual, because the judgment calls — which model, what guardrails, when the output is good enough to trust — are exactly where seniority matters.
Ask directly who will do the work, and how much of their time you get. Firms built around senior, hands-on delivery will answer happily; firms built around leverage will get vague.
3. How will we know if it worked?
A serious firm proposes success metrics before the contract is signed: hours saved per week, error rates, resolution time, cost per processed document. If the answer is engagement-level fluff ("adoption", "AI maturity"), you're buying activity, not outcomes. Our guide to measuring AI ROI covers what good measurement looks like — any firm you hire should be comfortable with that level of rigour on its own work.
4. What happens to our data?
Ask where your data goes, which third-party model providers see it, what the retention terms are, and how the firm handles anything sensitive. Competent consultants answer this fluently because they've had to solve it before; if you get hand-waving, walk. (For what to check yourself, see our guide to protecting company data when using AI tools.)
5. What would you do if AI is the wrong answer?
This is the honesty test. Plenty of business problems are better solved with a simple integration or a process change than with a language model. A firm that also does IT consulting or systems work can tell you that; a firm that only sells AI has every incentive not to. The best answer you can hear in a sales conversation is some version of: "We'd tell you, and here's a cheaper thing that would work."
6. How do you scope and price the first engagement?
Be wary of firms that want a large upfront commitment before anything has been proven. The sensible pattern is a small, well-defined first project — a scoped AI pilot with clear success criteria — that lets you judge delivery quality before you commit serious budget. On numbers: API-based integrations typically run $2,000–15,000 to set up, and custom-built AI systems $10,000–50,000+ depending on complexity. Anyone quoting a precise price and a guaranteed outcome before seeing your data is guessing, and you'll pay for the guess.
7. Strategy, implementation, or both?
Some firms only produce strategy; some only build to someone else's spec. The handoff between the two is where context, momentum, and accountability get lost. If you can, hire for both halves in one place — AI strategy consulting that ends in an implementation plan, and AI implementation carried out by the same people who wrote it.
The quick red-flag list
If the seven questions feel like too much for a first call, screen for these:
- Guaranteed results before seeing your data. Nobody honest can promise that.
- Slideware economics. A long strategy phase, priced generously, with implementation "to be scoped later".
- The team switch. Seniors in the sales meetings, juniors on the delivery roster.
- One answer for every client. If every problem gets the same chatbot, the diagnosis was the pitch.
- No interest in your existing systems. AI that ignores your current workflow and data will not survive contact with either.
Where Clear Frame AI fits
Clear Frame AI is built around the answers we'd want to hear to these questions: senior people who both scope and build, AI consulting that starts with the business problem, honest advice when AI isn't the right tool, and small first engagements with success metrics agreed upfront.
If you're evaluating AI consulting firms, get in touch — the first conversation is free, and you're welcome to put all seven questions to us.