Nobody here will connect their inbox to an AI

A publisher I work with said something on a call last week that explains more failed AI projects than any report I have read. We were talking about tools that sit on top of your email and do something useful with it, and he said he knows business owners who simply would not connect their inbox to one, whatever it could do for them. Then he had a go at why. People are nervous because they do not understand the language around it, and the ones who are confident are the ones who can follow the words. I think it sits one layer below that. It is not the language. It is the faith.

Who told you is doing most of the work

Here is the test I run on myself. If you and I had never spoken, and some website told you to connect your inbox and everything would be set up for you, you would not do it. You would read that sentence as a risk, and you would be right to. Now take the identical sentence out of the mouth of somebody you already know, who runs it on their own business, and who can show you what it did to their week. Same claim. Completely different answer.

So the sentence is not the variable. The source of the sentence is. Which means the job in the room is not a better explanation, and it is certainly not a longer one. It is somebody standing next to the working thing saying I use this, this is my own inbox, have a look.

The question underneath the nervousness

When an owner says they do not trust it, nine times out of ten they are asking one narrow question without saying it out loud. Where does my stuff go, and who else gets to see it. That one is answerable, and the answer is different for different piles of stuff.

So I get people to sort their information into two. The first pile is quotes, suppliers, schedules, chasing invoices, the daily grind of correspondence that every business in the trade has a version of. If some of that passes through a model, nothing of value has left the building. The second pile is the part that actually is the business. The pricing logic, the drawings, the client list, the method somebody spent twenty years working out. That pile gets different treatment.

For the second pile we have built isolated models that sit on the client's own server and talk only to the client's own data. More work, more cost, and for a business whose whole worth is in that pile it is the only version I would put my name to. The mistake is treating both piles as one decision, because then the most sensitive thing you own sets the policy for your supplier emails.

What actually moves a team

• Start on the boring pile, where the worst case is embarrassment rather than damage
• Have one person run it on their own work for a fortnight and say out loud what went wrong
• Show the thing running, not a presentation about the thing running
• Let the people who will use it name what they refuse to put in, and then genuinely keep that out

None of those are arguments. That is deliberate. I have never once watched somebody get talked out of a worry about their own data, and I have watched plenty of them quietly drop the worry after two weeks of nothing bad happening.

Nobody adopts a tool because the explanation was good. They adopt it because somebody they already trust had put their own work through it first and told them what broke.

So if your team will not connect their inbox, they are not being awkward and another demo will not shift them. Use it yourself for a fortnight, in front of them, including the bits where it gets things wrong. It is slower and duller than a demo, and it is the only thing I have seen work.