Your AI Is Industrialising Your Marketing Silos

What is the most harmful thing AI did in your business this quarter?

It is not the obvious hallucination. Not the clearly invented statistic. It is something far more subtle. The confident, well argued insight your team acted on, which turned out to be built on the wrong picture of your business.

You are not alone. New research reported by WARC found that 72 per cent of marketers say their business has made a decision based on an AI insight that later proved wrong or misleading. More than a third say it cost them money.

The most harmful thing here is believing the AI is the problem.

Five departments, five customers, five propositions

There is a test we run regularly with leadership teams, and it is worth trying with yours.

Ask five people from five different departments to write down who your target customer is. One sentence each, no conferring.

In a decade of doing this, I have never seen five matching answers. Sales describes their buyer. Product describes their user. Finance tends to deal in demographics. Marketing describes a persona. Everyone is confident, and everyone is usually right, based on their own research and their own data.

That has been true in most companies for as long as we have had departments. It was expensive but survivable while the fragmentation moved at the speed of humans. Research commissioned and paid for twice. Two agencies solving one problem in two different ways, with two invoices. A pricing decision made in one room while the ad contradicting it was built in another. Slow, costly, and rarely visible in a board pack.

Now we are putting AI into every one of those departments.

Each team briefs it with its own documents, its own definitions, its own version of the strategy, and sometimes its own platform. The machine does exactly what it is designed to do. It takes each silo's version of the truth and produces more of it, faster, in beautifully fluent prose.

The 72 per cent is not a story about AI failing. It is a story about AI succeeding at amplifying what it was given.

Conway's Law, expanded

Software engineers have a name for the deeper pattern. Conway's Law observes that any system an organisation designs ends up mirroring that organisation's own communication structure. Build software with four teams and you get software with four seams.

A marketing leader at a global bank described living through it. The software his teams produced was, in his words, a reflection of the org chart, and it was bonkers. This team wanted that, so the system had that. Nobody designed the fragmentation. The structure did it for them.

Your AI outputs are now doing the same thing, in every function, every day, and getting more articulate about it as they go.

Large language models run on language. Not on your intentions, your culture, or what the executive team agreed at the offsite. On the words and definitions that are actually written down and available to them. Every ambiguity in your business language becomes an ambiguity in the output, and the model resolves it the only way it can, by taking the first definition it reaches and stating it with total conviction.

Readiness is a structure question

Gartner's latest survey has 70 per cent of CMOs calling AI leadership a critical goal for 2026, and 70 per cent admitting their processes are not mature enough to scale it.

The industry reads that as a technology gap and reaches for better tools, better training, better governance. I read it differently. Readiness is a structure question before it is a technology question, and the structure in question is language.

The companies getting value from AI are not the ones with the best models. They are the ones where a model can be pointed at the business and find one definition of the customer, one map of the value proposition, one view of the brand personality, the design, the strategy, and everything else.

Give a model clear, consistent, structured language and the output changes profoundly. Give it six departments' worth of shorthand and it will industrialise the confusion, and burn through tokens doing it.

The multiplier is agnostic. It simply multiplies whatever it is deployed within.

The audit is cheap

The good news is that this is one of the few problems where the audit does not take months or an expensive consulting firm.

Run the five-sentence test at your next leadership meeting. Then repeat it for strategy, for the value proposition, for the elevator pitch, even for which film character best personifies the company. Whatever you are most curious to sense check. Put the answers side by side and there is your audit.

The fix depends on how far apart those answers are, and on the size, shape, and nature of your business. But it starts in the same place every time, with a shared universal language. That is why we built the Anatomy of Marketing.

The Anatomy of Marketing is our open methodology, free to access at wiki.anatomyofmarketing.org. Run the five-sentence test this week and see how many customers your business describes.