The First 10 Minutes Tell Me Everything

In the first ten minutes of a conversation, I can usually tell whether we can genuinely help a company with AI — long before we talk about models, budgets, or timelines. That sounds like a sales line, but it isn't. It's pattern recognition from a lot of discovery calls, and it comes down to listening for a handful of signals that have almost nothing to do with technology. The technology is the easy part. Whether a project will actually work depends on things you can hear in how someone describes their problem.

I want to walk through what those signals are, partly because it explains how we qualify work at Crux Digits, and partly because if you recognise them in your own thinking, you'll get far more out of any AI conversation — with us or anyone else.

Why the first ten minutes matter more than the brief

Most companies arrive with a brief: "we want a chatbot", "we want to use AI for X", "our competitor is doing something with LLMs". The brief is rarely the real story. What tells me whether we can help is the conversation underneath it — how clearly the person understands the problem they're actually trying to solve, and whether AI is a means or just a buzzword they feel pressure to be seen using.

By the time we're discussing tech stacks, the important decision has usually already been made. A well-scoped problem with a clear owner and usable data will succeed with fairly ordinary technology. A vague ambition with no owner and messy data will fail with the most sophisticated model on the market. So I spend those first minutes on the problem, not the solution.

What I'm actually listening for

There are four questions running in my head during any first call. I rarely ask them this bluntly, but the answers come out anyway.

Can you describe the problem without mentioning AI?

This is the big one. If someone can describe a concrete, painful problem — "our support team answers the same forty questions every day and it's eating fifteen hours a week" — then we have something to work with. AI may or may not be the answer, but there's a real problem to aim at. If the only way they can describe what they want is by naming a technology, that's a sign the technology came first and the problem is still being reverse-engineered to fit it. That ordering rarely ends well.

Is there a number attached?

Genuine problems usually come with a number, even a rough one. Hours lost, deals dropped, error rates, response times, cost per case. When someone can tell me what the problem costs them today, two good things follow: we can size whether a solution is worth building, and we'll be able to tell afterwards whether it worked. "We want to be more innovative" has no number, and a project with no number is a project no one can ever call a success. This is the same discipline behind a proper AI implementation — outcomes first, tools second.

Who owns this if it works?

AI isn't a project you finish; it's a system you operate. So I listen for whether there's a named person who will live with the thing after launch — monitor it, correct it, own its results. When the answer is "we'll figure that out later", I get cautious. A model with no owner drifts, quietly degrades, and eventually gets switched off. The most successful clients already know whose job this becomes.

How do you talk about your data?

You learn a lot from how someone describes their data. The ones who say "it's all in one system and it's reasonably clean" are usually ready. The ones who go quiet, or cheerfully assume the data is fine without having looked, are telling me there's data engineering work to do first. That's not a deal-breaker — it's often the most valuable thing we end up doing — but it changes the shape and sequence of the whole engagement.

There's a fifth thing I notice, harder to name. It's whether the person is curious or just looking for reassurance. Curiosity sounds like questions back at me — "what would that actually take?", "where does this usually go wrong?". Reassurance sounds like someone who has already decided and wants permission. The curious ones build better systems, because they stay engaged when the work gets unglamorous, which it always does. A model is only as good as the team willing to keep tending it after the launch buzz fades.

The green flags that make me lean in

Some signals make me genuinely excited about a project. They have a clear, specific problem they can describe in plain language. They're curious rather than defensive when I push on the details. They've thought about what success would look like and can put a number on it. And — this one matters more than people expect — they're open to hearing that AI might not be the right tool yet. Paradoxically, the clients most ready to build well are the ones least desperate to build.

I also pay attention to whether someone treats the first conversation as a two-way diagnosis or a pitch they need to win. The best engagements feel like two people trying to figure out the truth of a situation together. You can usually sense that within minutes.

The red flags that make me slow down

And some signals make me want to slow things down — not to walk away, but to reset expectations before anyone spends money. The clearest one is hype with no underlying problem: a strong desire to "do something with AI" that evaporates the moment you ask what specifically is broken. Another is a fixed solution looking for a justification — they've already decided exactly what to build and just want a supplier to nod. A third is unrealistic certainty about data that no one has actually checked.

None of these mean a project is doomed. They mean the honest first deliverable isn't a model — it's clarity. Sometimes the most useful thing we do is help a company turn a vague ambition into a sharply defined problem, which is exactly what a short, fixed-scope audit is for. Occasionally the right answer really is "not yet", and saying so early is the most valuable thing a vendor-neutral partner can do. If you operate in Europe, getting that sequencing right also keeps you on the correct side of governance — the EU AI Act ties obligations to how and where a system is used, so the scoping conversation and the compliance conversation are really the same conversation.

What this means if you're about to book a call

If you're thinking about talking to us — or any AI partner — the most useful preparation has nothing to do with technology. Come ready to describe the problem in plain words. Bring the number it costs you. Know who would own the solution. And be honest about the state of your data. Do that, and the first ten minutes will tell both of us something real, fast.

That's deliberately how we work. We'd rather spend an honest half-hour establishing whether there's a project worth doing than sell you one there isn't. You can see how that plays out in our case studies, and our transparent pricing starts with a low-commitment audit precisely so the first real deliverable is clarity, not a contract. Review our pricing, or book a free consultation and we'll map your first use case together.

Frequently asked questions