I Don't Try to Convince People to Work With Us

I don't try to convince people to work with us. That probably sounds odd coming from the founder of an AI consultancy — surely the whole job is persuasion? But when someone books an AI consultation with Crux Digits, my goal on that first call isn't to close them. It's to tell them the truth about whether AI will actually move the needle for their business — even when the honest answer is "not yet", "not like this", or "honestly, not us". I've found that the less I push, the better the partnerships turn out. Here's why I run things that way, and what it means if you're the one on the other end of the call.

Persuasion is the wrong tool for an AI project

Most AI projects don't fail because the technology wasn't clever enough. They fail because nobody agreed on what success meant, the data wasn't there, or the business never really wanted the thing in the first place — it was sold on it. The numbers back this up: an MIT study reported in 2025 found that the overwhelming majority of generative-AI pilots delivered no measurable return, and analyst estimates have long put AI project failure rates north of 80%. A big share of that waste traces back to one root cause: a project that existed to satisfy a pitch rather than a need.

If I talk you into a proof of concept you don't believe in, I haven't won anything. I've started a clock on a project that your own team will quietly resist, that will stall the moment the initial enthusiasm fades, and that will end up as another line in the "we tried AI once" graveyard. A signature I had to extract is a liability, not a win. So I don't extract them.

What a no-pressure AI consultation actually looks like

Here's the shape of a first call with us. I ask far more than I tell. What's the actual problem — not the AI feature you read about, but the bottleneck that's costing you time or money? Who feels that pain today, and how do they work around it? What does "this worked" look like in numbers you'd recognise six months from now? Most of the call is me trying to understand your business well enough to give you an honest read, not me running through slides.

By the end, you'll get one of three answers. Sometimes it's "yes, there's a clear, fundable use case here, and here's roughly how I'd scope it." Sometimes it's "maybe, but you've got a data or process problem to fix first, and rushing the AI would waste your money." And sometimes it's "you don't need us for this — here's the simpler thing you could do instead." All three are useful. Only one of them is a sale, and I'm genuinely fine with that. If you want a sense of how we'd structure the work when the answer is yes, our guide on how to scope an AI proof of concept walks through it.

Defining success before discussing solutions

I won't talk architecture until we've agreed what we're trying to achieve. That's not consultant theatre — it's the single biggest predictor of whether an AI project survives contact with reality. Frameworks like the NIST AI Risk Management Framework exist precisely because "we built something clever" and "we solved the business problem responsibly" are very different outcomes, and the gap between them is where budgets disappear. Pinning down a measurable goal early is the most honest thing I can do, because it's also the thing that lets you hold me accountable later.

Self-selection beats persuasion every time

The clients who turn into the best partnerships are the ones who chose us with clear eyes — not the ones I convinced. When someone arrives already understanding the trade-offs, ready to commit their own people's time, and aligned on what they want, the project has a spine. When someone has been talked into it, every obstacle becomes a reason to question the whole thing, because they never fully owned the decision.

So my job on a first call is less like selling and more like helping you decide for yourself. I'd rather you walk away clear-eyed — even if that means walking away — than sign up half-convinced and regret it in month two. Removing the pressure isn't a soft tactic. It's how you end up with partners instead of customers who feel cornered.

The hidden cost of being convinced

There's a real price to being sold something. When a project starts from persuasion rather than conviction, you carry a sunk-cost weight from day one. The internal sponsor who was lukewarm stays lukewarm. The team that wasn't consulted drags its feet. And when the inevitable hard moment arrives — the data is messier than hoped, the first model underwhelms, the timeline slips — there's no shared belief to fall back on. The project doesn't fail loudly. It just quietly never ships, and becomes the reason that organisation is wary of AI for the next three years.

I've seen this pattern enough to want no part in starting it. It's also why we publish transparent pricing rather than bespoke "let's discuss budget" quotes: when you can see the numbers up front, you're deciding, not being managed toward a close.

What I'm actually doing on that call

If I'm not selling, what am I doing? Diagnosing fit. I'm trying to work out three things quickly. First, is there a real, valuable problem here that AI is genuinely the right tool for — or is it a process fix, a better spreadsheet, or a clearer policy in disguise? Second, do you have the raw materials — usable data, a willing owner, a realistic timeframe — for this to work? Third, are we the right team for it, or would you be better served by an in-house hire or a different specialist?

That last question matters more than it sounds. Being vendor-neutral and engineering-led means I'm not reselling anyone's licences and I don't have a quota of seats to fill. I have no product to push you toward, which means I can tell you when the honest answer is "build this on tools you already pay for" or "this doesn't need a consultancy at all." Our piece on how to choose an AI consultant goes deeper into what that neutrality should look like from the buyer's side.

When I tell people not to work with us

This isn't theoretical. I end calls with "I don't think you need us" more often than you'd expect. A few of the recurring reasons:

• The data isn't ready. If the information an AI system would need to learn from is scattered, inconsistent, or doesn't exist yet, the responsible move is to fix that foundation first — not to bill you for a model that will learn the wrong lessons from bad inputs.
• The problem isn't an AI problem. Plenty of "we need AI" requests are really "our process is broken" or "our team is under-resourced." AI layered on top of a broken process just gives you a faster broken process.
• The timing is wrong. If a reorganisation, a system migration, or a regulatory shift is about to change everything, it's often smarter to wait a quarter than to build on ground that's about to move.
• The scope only makes sense for someone in-house. Some work is so continuous and embedded that you'd be better off hiring than retaining a studio. I'll say so.

Turning down work I could have taken isn't generosity. It's the only way the word of a vendor-neutral consultancy means anything. If I'll say yes to everything, my "yes" is worthless to you.

Regulation makes honesty even more important

There's a compliance dimension to all this too, especially in Europe. Under the EU AI Act, the obligations attached to an AI system scale with its risk — and a consultancy that oversells you into a high-risk use case without flagging what that entails is setting you up for a problem that's far more expensive than a missed sale. This is general information, not legal advice — but part of an honest AI consultation is mapping not just whether something is possible, but whether it's wise and compliant in your context. Pressure-driven sales tends to skip that conversation. We start with it.

How this changes the relationship

When you remove the pressure, something better takes its place: a working relationship built on the assumption that we're both trying to get to the truth. You're not bracing for a pitch; I'm not steering toward a contract. We can disagree, scope things down, kill a bad idea early, and trust each other's read. That's the soil good AI work actually grows in — usually a working prototype by the second call, an audit before any big spend, and a clear path from proof of concept to production if and only if the proof of concept earns it. You can see how that tends to play out in our case studies.

It's a slower way to grow a consultancy. I'm fine with that. I'd rather have twenty clients who chose us on purpose than fifty who were talked into it and tell their network we oversold them.

If you're evaluating an AI partner, watch for the pressure

Let me turn this around, because it's genuinely useful whoever you end up working with. When you're choosing an AI consultancy, the way they sell tells you how they'll deliver. A few signals worth noticing: Do they ask about your problem before pitching their solution? Are they comfortable telling you something isn't a fit? Is the pricing visible, or does every answer route back to "let's get you on a call"? Do they manufacture urgency, or respect that this is a serious decision? A partner who pressures you into starting will pressure you into scope creep, into ignoring red flags, into shipping before it's ready. The sales conversation is the trailer for the whole film.

A call I'm glad I lost

A while back I spoke with a founder who was certain he needed a custom AI assistant trained on his company's documents. He'd read about retrieval systems, he had budget approved, and he wanted to start the next week. It would have been the easiest yes of my month. Instead I asked what the assistant was supposed to save him. As we talked it became clear that his real problem wasn't search — it was that three teams kept the same information in three different places, and none of them trusted the others' version. An AI assistant trained on contradictory documents would have confidently served up contradictory answers, and he'd have blamed the AI.

I told him so. I suggested he spend a fraction of his budget getting those three sources into one place first, and that if he still wanted the assistant after that, we could talk. He was, briefly, annoyed — he'd come ready to spend. A few months later he emailed to say he'd consolidated the data, the "AI problem" had mostly evaporated, and the slimmed-down assistant he eventually built was genuinely useful because it had clean ground to stand on. I lost a bigger project and kept a relationship. That trade is one I'll make every time.

An audit is not a sales call

This is where the no-pressure approach gets concrete rather than philosophical. When there's enough substance for it, the honest next step usually isn't "sign a build contract" — it's a paid audit. For a few thousand euros we map the actual use case, look hard at your data and constraints, and tell you whether a proof of concept is worth funding. The deliverable is a clear-eyed recommendation, and that recommendation is sometimes "don't build this yet." A sales call wants you to commit; an audit wants you to know. The two have opposite incentives, and I've deliberately built our process around the second one.

That's also why our pricing ladder is public and gradual: an audit to map the right use case, a proof of concept only once the audit justifies it, and production only when the proof of concept earns it. You're never asked to leap straight to the expensive end on the strength of a good meeting. Each step has to prove itself before the next one starts, which means the pressure to "decide now" simply has nowhere to live. You can see the shape of that on our pricing page and what the work looks like in AI implementation.

Education is part of the deal

A surprising amount of a good first call is just helping someone understand their options well enough to make a decision they won't regret — what AI can and can't do for their specific situation, where the real costs hide, what "good" looks like. I'd rather you leave more knowledgeable than when you arrived, whether or not you hire us, because an informed buyer makes better decisions and better partners. If that means I've spent an hour teaching someone who then does the work in-house, fine. They'll remember who was straight with them.

The questions I wish more buyers asked me

If you're going into any AI consultation — with us or anyone — here are the questions that cut straight to whether you're being levelled with:

• "What would make you tell me not to do this?" Anyone who can't answer that has nothing they won't sell you.
• "How will we know in six months whether this worked?" If the answer is vague, the project has no spine yet.
• "What could we do without you?" A confident partner will happily tell you the cheaper, simpler path — and why it might not be enough.
• "Where does this typically go wrong?" Honest practitioners have scars and will show them; salespeople have only success stories.
• "What does this cost, roughly, before we go further?" If you can't get a ballpark without three more meetings, that's a signal in itself.

None of these are gotchas. They're just the questions that make pressure-based selling impossible to sustain — which is exactly why a good partner welcomes them.

It helps that we'd rather build than sell

Part of why this comes naturally is the kind of team we are. We're engineers first, not a sales organisation with a delivery arm bolted on. The work I actually enjoy is the building — scoping a tricky use case, getting a model to behave on real data, shipping something a client uses on a Monday morning. Selling is just the necessary doorway to that work, not the point of it. When the satisfying part of your job is the delivery rather than the close, you've got no reason to drag someone through a door they shouldn't walk through. A bad-fit project isn't a win you booked; it's months of doing work you don't believe in for a client who'll be unhappy. I'd genuinely rather have the empty slot.

So — I won't try to convince you. If you've got a problem you think AI might solve, book a consultation and we'll work out together whether it's real, whether it's us, and what the honest first step looks like. If the answer is that you don't need us, you'll leave knowing that too — and that's a perfectly good outcome.

Frequently asked questions