You launched it, it was good, and nothing happened. You still do not know why.
Everyone seems to have an AI initiative on the go. Some are serious, some see a shiny new toy. What they share is a sense of quick wins, the belief that work which used to take a year can now take a fortnight.
The quick speed is real. The problem is what gets compressed.
AI accelerates production. It does not accelerate judgement, and it cannot supply knowledge you do not already hold. It will generate a proposition, a service line, a framework, or a set of client-facing collateral, quickly and competently. What it cannot tell you is whether any of it is right, whether it differs from what your competitor produced from a near-identical prompt, or what will break when a real client puts weight on it.
That is the issue worth highlighting. Not that the output is bad, but that the output looks finished before the real thinking underneath it has been done. The cost arrives later, in delivery, when the service cannot be repeated consistently and nobody internally can explain why a client should choose you over three firms saying the same sentence.
Which is why I keep thinking about a spreadsheet.
Twenty thousand questions
Ctrl+C. Ctrl+V. Ctrl+C. Ctrl+V.
Laptop open on the table, it’s 2007. One window showing an Excel spreadsheet. Another showing an empty content field. Copy the question. Paste it. Copy the first answer option. Paste it. Copy the second. Paste it. Copy the feedback explaining why that answer is right or wrong. Paste it. Save. Move down one row.
Then do it again. Thousands and thousands of times.
Twenty thousand questions, each with four answer options and detailed feedback written against every one. All of it sat in spreadsheets. All of it had to end up inside the e-learning platform I was building. There was no import tool that did what the build required, no team standing by, and no shortcut at any price. There was a spreadsheet, a platform, and one person moving the first into the second by hand, in the evenings, for the better part of nine months. At least I did not have to write them.
Twenty thousand is worth sitting with, because it is the distance between a sentence and a reality.
“We built an e-learning platform for general insurance” takes four seconds to say. It is a line on a website. It is a slide.
Twenty thousand multiple choice questions is what the line really cost.
What the hard build bought
That platform eventually became a market leader from a standing start, and thousands of users across the sector used it every day. It opened a vertical that had not previously existed for the consultancy business, and it played a real part in raising professional standards in UK general insurance, still doing so years after that first version.
The easy attribution is to the idea. Early e-learning, right sector, right moment. But plenty of people had the idea. The base layer was the asset, and it bought three things that could not have been acquired any other way.
Knowledge. Handling twenty thousand questions individually makes it impossible to avoid learning the material, the structure underneath it, and where it is weak. By the end I understood the content architecture of an entire professional landscape at a level no summary would have delivered. That drove product decisions for years afterwards.
Quality. Manual work at volume exposes every inconsistency. Duplicated questions. Spelling mistakes. Feedback contradicting itself between modules. Learning outcomes that had drifted away from the questions supposedly testing them. None of that surfaces in a strategy session.
Defensibility. A competitor could have copied the concept in an afternoon. They could not copy nine months of manual entry, and more to the point, they did not want to. The barrier was not technical. The work was boring and nobody else was prepared to do it.
That last point matters more now than it did in 2007. When everybody can generate the same output at the same speed, the durable advantage is increasingly the part that cannot be generated.
You may recognise this
Founders rarely arrive saying their infrastructure is wrong. They arrive with symptoms.
- You launched a new service line and it produced polite interest, a few conversations and almost no conversions.
- You concluded the market was not ready.
- You describe what you do fluently, but when asked why somebody should choose you over a named competitor, you find yourself describing how you work rather than what makes you different.
- Every proposal takes a different shape, because there is no underlying architecture to draw from, so each one gets rebuilt from scratch under time pressure.
- Delivery quality depends on who is in the room. When it is you, it is excellent. When it is not, it is variable, and you cannot articulate the difference well enough to fix it.
- You have opened three verticals in two years and none of them stuck. Each time the explanation was external.
- You bought a tool expecting it to create a capability rather than accelerate one you already had.
Those are not strategy failures. They are one structural failure appearing in different places. The change was made at the visible layer and the base layer was never built.
Making the change without the mess
Five things will save you most of the pain.
- Separate the layer you are changing from the layer you are standing on. A new vertical is not a new page on the website. It is a proposition, an evidence base, a delivery model, a pricing logic, proof points, and a clear account of who it is not for. If you cannot complete those six, that is the work, not a formality in front of the work.
- Do it by hand before you launch it. Deliver the new service manually a few times, with you in the room, before it is templated or handed over. You cannot productise something you have not felt the weight of.
- Look for the boring barrier. What could a competitor copy in an afternoon? What would they refuse to do at all? Your differentiator lives almost entirely in the second answer, and it is usually unglamorous rather than clever.
- Decide explicitly what you are not. Positioning fails through inclusion far more often than omission. The sharpest founders can name, without discomfort, the clients they turn away.
- Set the time horizon before you start. Difficulty at month four is not new information. Decide up front how long the build gets, so that when the self-doubt arrives you are not making a strategic decision with a tired brain.
Conviction and ease are not the same measurement
I never doubted the platform was worth building. The vision was clear. The execution was mind-numbing.
Founders routinely treat that combination as a contradiction. When work becomes hard, repetitive and slow, the instinct is to read the difficulty as a signal that the idea was wrong or a better, faster route exists.
A clear vision does not buy an easy build, and a slow build is not evidence of a flawed strategy. The founders who lose the most ground are the ones who change direction because the work got heavy rather than because the thinking changed.
Why I build this way
My commercial frameworks start with diagnosis rather than delivery, and build from the bottom up. The 7 Plateaus establish where a business actually is. The 5Rs of commercial infrastructure build what the next stage requires.
The base layer is almost always what has been missed. Rarely through carelessness. It can be invisible from the outside.
Build the engine from the foundations and three things follow. You surface opportunities the business was already sitting on, including services delivered informally for years and never priced. You find more than one route to the goal, and some are cheaper than the plan you arrived with. And you generate far better information about your own business: sharper language, specific proof points, a proposition that can be stated in one sentence and strengthened in ten.
That last one is also what determines whether AI is worth anything to you. Hand a model a generic understanding of your business and it returns generic output, very quickly. Hand it a diagnosed position, a defined delivery model and real evidence, and it becomes a genuine accelerator.
The tool is not the variable. What you hand it is. Build the base layer properly and using AI to speed up delivery becomes a choice rather than a dependency.
Nobody shows you this part
Market leading. Thousands of users. That got quoted. That went on the slides, into the case studies, into every conversation with a prospect. It was clean, it was impressive, and it took no time at all to read.
Underneath it was a kitchen table at half nine at night, two windows side by side, and thousands of rows still below the fold.
The decks showed the platform. The charts showed the curve. The numbers showed the market share.
Nobody showed the Ctrl+C. And the Ctrl+C is where a market leader started from.
