The Question Nobody Else Asked

The call was supposed to be routine. A mid-sized logistics firm — 300 staff, three UK sites, a compliance officer who'd clearly been burned before — wanted a second opinion on an AI proposal they'd already received. Ben had seen the type before he'd even opened the deck. Chatbot for customer service. Document summarisation for the ops team. A dashboard with the word "insights" on it somewhere.

"It looks good," the compliance officer said, not quite convinced by her own sentence. "I just don't know what I don't know."

That was the whole meeting, really. She'd stopped worrying about whether the AI was good somewhere around the second vendor pitch. What she wanted to know was what happened after the demo ended. Where did the data go. Who could see it. What happened if it quietly started doing the wrong thing for three weeks before anyone noticed.

Ben had spent years before Inzen answering exactly that kind of question, just not about AI — designing, deploying, and supporting infrastructure for organisations across hundreds of countries and thousands of users, including backup architecture, data analytics, and discovery projects. In that world, "we'll figure out recovery when we need it" was never acceptable, because by the time you need it, it's too late to figure out. You learn to ask about the failure before you ask about the feature.

So instead of walking through the proposal slide by slide, he asked a different set of questions. Where does the data actually sit once it leaves your systems? Is there a log of what the AI queried and why? Who has access to the outputs, and is that access ever reviewed? What happens the day the model gives a wrong answer with total confidence — is there a human checkpoint, or does it just go out?

The compliance officer went quiet. Then: "Nobody's asked us those questions. The other consultancy just told us it was secure."

That's usually how it goes. Most AI consultancies today came from the model outward — brilliant at prompting, at frameworks, at getting an agent to do something genuinely impressive in a demo. What they haven't spent years doing is the unglamorous work of keeping a system alive and accountable long after the room has stopped being impressed. Logging. Observability. Evaluation that keeps running after launch, not just before it. Access control that's actually reviewed. It isn't that they're bad at AI — most are genuinely good at it. It's that building a capable model and running one safely inside a real business are two different disciplines, and most firms have only mastered one.

Ben didn't try to sell the logistics firm on Inzen being smarter about AI. He just answered the questions honestly, including the ones where the honest answer was "that depends on how you want to scope it." Halfway through, the compliance officer stopped taking notes. "Can you send over what a proof of concept with you would actually look like?"

Trust didn't show up because of a confident pitch. It showed up in the specific, checkable details — the kind that are easy to skip past and very hard to fake.

If that's the kind of conversation you've been waiting to have — where someone actually answers what happens after the demo — get in touch for an initial consultation. No hard sell. Just a straightforward, governance-aware look at where AI genuinely fits in your organisation.

Not magic. Just momentum. Let's build something.

Inzen