Your AI personalisation is only as good as your customer data

Short answer: generative AI can write a thousand personalised messages in the time it used to take to brief one, but it can only personalise on what it knows. If your customer data is fragmented, stale or wrong, AI will scale the mistakes along with the message.

The shiny new toy problem

Most marketing teams I talk to are experimenting with generative AI, and they should be. The ability to create customer-aware content at scale, without a matching increase in production cost, is a genuine shift.

But there's a step that gets skipped in the excitement. A large language model writing an email to a customer has no idea who that customer is unless your systems tell it. Personalisation has always been a data problem first and a content problem second. AI hasn't changed that. It has raised the stakes.

What good customer data infrastructure looks like

Unifying customer data is hard because the data is large, spread across systems that weren't designed to talk to each other, collected inconsistently, and constantly changing. A useful customer data foundation has three layers.

Build. Getting data in, resolving identities so one person is one profile, modelling it into a shape your tools can use, and building the workflows that move it to where it's needed.

Maintain. Data quality checks, monitoring, consent and compliance management, error handling and governance. This is the part most budgets underestimate, because data doesn't stay clean on its own.

Change. Every new data source or destination is a risk to what's already working. You need a controlled way to test and release changes, not a live edit at 4pm on a Friday.

Where AI fits

The real opportunity is in connecting the two: a customer data platform or warehouse holding accurate, consented profiles, and a language model using those profiles to tailor content, offers and service in the moment. Done properly, that combination gives brands personalisation at scale while keeping control of their data, privacy and security.

Done badly, it means confidently addressing the wrong person, recommending the product they returned last week, or using data they never agreed to share.

Three questions to ask before scaling AI personalisation

1. Would you trust this profile if a human were using it? If your team wouldn't send a personalised email based on it, an AI shouldn't either.
2. Do you know where each data point came from and whether you have consent to use it? If not, fix that first.
3. Can you measure the lift? Hold back a control group so you know whether personalisation is actually changing behaviour, not just producing more content.

The brands that win with AI personalisation won't be the ones with the cleverest prompts. They'll be the ones with the cleanest data.

If you're working through this, I'm happy to compare notes. Get in touch.