Your AI has versions of everything. It doesn't have the customer.
Not your CDP’s version. Not your persona deck’s version. Not whatever ‘the customer’ meant in the last strategy meeting. AI is reading all of those simultaneously - and producing something that sounds exactly like insight. It isn’t. It’s an average of contested versions, rendered fluent.
AI doesn’t have a data problem. It has a ‘customer’ problem. The signal its reasoning from was never agreed. It was argued over, synthesised, paraphrased and passed around until what arrived at the model wasn’t what the customer meant at all. AI didn’t create this. It inherited it. Then it scaled it.
The tech world solved a version of this twenty years ago. Every enterprise system contained its own thing called ‘customer’ - the same word, never the same idea. When they tried to make the systems talk to each other, it broke things visibly and had to be fixed. The tech side saw it and solved it, for themselves.
The business side never had that forcing function. A strategy deck doesn’t break when its assumptions about customers contradict each other. A persona doesn’t flag when three teams are using the same word for three different people. Nothing surfaces it. So it hasn’t been fixed.
AI is the forcing function that’s finally arrived on this side of the organisation. Except it doesn’t break visibly. It produces something that looks resolved - fluent, confident, plausible. When things break, people fix them. When things look fine, nobody looks. AI makes this version of the problem impossible to see - for anyone.
What’s been missing is ‘canonical customers’ - not a composite, not a persona, but an expression of what customers actually think they want, held at the source.
I built something for this and called it Stratum.
The technology isn’t new. Video. A graph. What matters is what it’s built to do: persist the expressed moment and keep it open. Not summarise it down. Not overwrite it. Hold it as it is.
It captures what customers express at a moment and preferably when they’ve been invited to be anecdotal. Stratum holds that in a form every team can read from their own angle. A product manager sees what it means for the roadmap. An engineer sees a structured file where every interpretation is marked. The commercial team sees what it means for the relationship over time. Same source. Every register. Nothing overwritten. Nothing averaged.
It’s append-only and the expression is the record. Interpretations fold alongside it without replacing it. So when upstream brings AI to Stratum, they have something real to reason from - not the version that won the senior meeting last Tuesday, not a persona built from half-remembered sessions.
If your AI doesn’t have something like this, it is going to be reasoning from a very messed-up mash-up of ‘customer’. Early days. I’d love to know what you make of it.