Does AI Transformation Actually Work? What I'm Learning Running My Own Business Through One

What does an AI transformation actually look like from the inside?

It looks far messier than any case study makes it sound: old systems still half-running, new ones still half-trusted, and real people caught in between. I know this because for the first time in twenty years, the mess is mine rather than a client's.

I run two businesses. Accelerator X is the consultancy Andy and I built together, the one whose name is on this website. Accelerator Solutions is older, roughly twenty-five people, real clients, real payroll, and it is the business currently going through a live AI-led rework of how it hires, how it delivers, how it runs operations, and how it does its finance, more or less all at once, while it keeps trading. Nobody stops answering the phone while we do this.

For most of my career I stood on the outside of transformation like this: brought into a bank, an insurer, a global manufacturer, to help an existing organisation, with its existing habits and its existing politics, get from where it was to where it needed to be. I reviewed the plan. I sat in the steering committee. I warned leadership about the gap between the deck and the shop floor. What I had never done, until this year, was run that gap myself, with my own name against the outcome.

Something broke in week three that I would have flagged in a client's business without a second thought. We wanted to make an offer to a candidate and it went out four days late, because the one step that mattered, a reference call, existed only inside one person's head, and that person was on leave. Nobody had done anything wrong. Nobody had written the step down anywhere an AI tool, or anyone else covering that week, could see it. We found out it mattered when the candidate took another job.

That is not a technology failure. It is the same failure I have watched play out in three different industries over twenty years. It is just smaller and faster, because it is happening inside a twenty-five-person business instead of a twenty-five-thousand-person one.

Why does the same failure keep showing up in every transformation I have run?

Because agreeing something and building a dependable route from that agreement to finished work are two different achievements, and almost no organisation, mine now included, has ever properly built the second one.

In the banks and insurers I worked with, the pattern always had the same shape. Leadership agrees a decision in a steering committee. The minutes get filed, everyone nods, the deck gets archived. Getting from "agreed" to "done" then gets reconstructed from memory: from someone's inbox, from whoever is still in the room eighteen months later when someone finally asks what happened to it. I used to call this the execution gap. These days I think of it as a severed thread, the connection between a commitment and the trail of work that fulfils it, cut at the first handover, unnoticed until the thing that was promised fails to show up.

I found the same severed thread inside my own business in the first month of this rework. Every promise we had made a client at Accelerator Solutions sat in one shared tracker, and calling it a plan was generous. It read more like a running list of good intentions, things we had said sitting right next to things we had actually done, with nothing to tell you which was which unless you happened to ask the one person who remembered. Everything else, the daily running of the business, the prep for Thursday's workshop, whatever had broken overnight, sat in a second tracker that treated a two-minute admin task and the piece of prep that decided whether the week's biggest client session landed as exactly the same kind of line item.

I would have diagnosed that inside a client's business in the first week of any engagement. I had not diagnosed it in my own, because nobody audits the machine they are standing inside.

Does AI transformation actually work?

Yes, in the specific places where you can point at a mechanism actually running inside a real, trading business under real constraints, rather than describe it on a slide. Everywhere else, the honest answer is "not yet, and we are finding out."

My business partner Andy is running the equivalent experiment from the other side of the building. Where I spend my time telling clients what a working operating model ought to look like, Andy is the one who actually builds the systems that run Accelerator X day to day: the agents, the checks, the plumbing meant to turn an agreed piece of work into tasks that get done without a person copying anything from one place to another by hand.

Watching that build up close has been useful, mostly because it forces the same question onto AX that I am now forcing onto Accelerator Solutions: does this actually work, or does it only work in the pitch? AX has set itself a hard, public date, the middle of August this year, for its own version of what I would call the golden thread, an unbroken line from a signed piece of client work to the commitments and tasks that deliver it, to be visibly and functionally live. Not "we believe it is working." A single view that leadership actually opens every day. No more losing Sunday evening to reconstructing what was promised to whom, in a way you can point at rather than simply hope is true.

I want to be clear about what this is not. It is not a tidy AI transformation case study, written up afterwards with the false starts edited out. It is still running as I write this sentence. It could still fail, or half-fail, the way I have watched transformation programmes half-fail for twenty years: not with a dramatic collapse, but with a slow drift back to the spreadsheet and the shared inbox, because nobody trusted the new thing enough to lean on it when it actually mattered.

That is the real test of whether any of this works. Not a demonstration. Not a quarter of good intentions. Whether, a year from now, the thread from commitment to delivery is still unbroken on the Tuesday nobody happened to be watching.

What is actually hard about running real people through this?

The technology is close to the easy part. The hard part is that every process you rebuild belongs to somebody, and that person's working week, sometimes their sense of what their job even is, shifts underneath them while you are doing it.

This is the part I underestimated, and with twenty years behind me I should have known better. Rework how a business hires, delivers, and reports its numbers all at once, and you are not touching a system, you are touching the people who built their competence around the old one. The account manager who was good at remembering which client wanted what. The ops lead who could hold the whole week in her head. Their skill was the workaround. Take the reason for the workaround away and, for a while, you make them feel less essential rather than more, even though the entire point of doing this properly is a business five to ten times as capable with exactly the same people in it, not a smaller one.

I have had versions of this conversation before, always as the consultant standing at the front of the room. Having it as the owner, across the table from someone whose role is genuinely changing shape under them, is a harder conversation to have well. The temptation, moving at the pace this requires, is to skip the explanation and just ship the change. I have watched that shortcut cost other organisations a year of quiet resistance. I am trying, not always successfully, not to take it in mine.

Why should anyone believe a consultancy that has not done this to itself first?

Because a consultancy telling clients to fix their operating model while running its own on spreadsheets and good intentions has no standing to be trusted with anyone else's.

That is the actual reason we are doing this the hard way, in public, against a deadline, rather than waiting until it is finished and tidy enough to present. I have spent twenty years watching organisations commission transformation from people who had never run the thing they were recommending. I would rather not be one of them twice.

So, does AI transformation actually work? Ask me again once the thread has held for a year, through a bad month as well as a good one. What I can tell you now, from inside my own business, with the systems half changed and the people still adjusting, is that the failure I used to diagnose in other people's organisations is exactly the failure that was sitting quietly in mine, and fixing it is slower, harder, and more political than any slide ever let on. That is not a reason to wait. It is the reason to start on yourself, now, before you next say the words "operating model" to a client.