In small markets, AI personalises your customers as someone else

Global CMOs are rolling out AI to make marketing more personal.

In small markets, it often does the opposite.

Personalisation runs on data. New Zealand has just over five million people. Australia has about 27 million. Your segments here are thin, your history is short, and what counts as an edge case globally is someone's entire customer base here.

When a tool doesn't have enough local signal, it doesn't stop. It fills the gaps with whatever it has most of. Usually, that's somebody else's market.

Your customers don't get personalised. They get personalised as someone else.

What it looks like

It's rarely dramatic. It's a string of small errors.

A retargeting ad in the wrong currency.

A help-centre answer that assumes US regulations apply.

A loyalty email promoting stores that don't exist here.

A summer sale in June.

A chatbot that can't answer a question any local would ask.

No single one is a crisis. Together, they tell customers the brand isn't really talking to them.

Why the dashboard won't catch it

In a big market, a few misfires disappear into the averages. In a small one, they are the average.

And the damage shows up first where dashboards are weakest: in sentiment, in referrals, in the tone of the replies you get. Your local team sees the drop before the numbers do. Most global teams never ask them.

It felt off

I saw this with a customer email journey rolled out across Australia and New Zealand. The logic was sound on paper: use customer behaviour to personalise the content, let the platform decide which messages were most relevant, and optimise the journey based on engagement.

Within a few weeks, the New Zealand unsubscribe rate had jumped by roughly 40%.

The dashboard showed the change. It didn't explain it.

The first instinct was to treat it as a frequency problem. Maybe we were sending too often. Maybe the audience had changed. Maybe the segments needed tightening.

Then someone on the local team read through the emails and said, essentially, "This just doesn't sound like us." That was the diagnosis.

The personalisation was technically working. It was just personalising from the wrong assumptions. Some of the recommendations, examples and language made sense for the larger markets the system had learned from, but felt strange or irrelevant to New Zealand customers. What looked like relevance in the dashboard didn't always feel like relevance in the inbox.

We pulled the journey back, reviewed the content locally and rebuilt the rules around what we actually knew about the New Zealand audience, not what the platform assumed about them. Within six weeks, the unsubscribe rate had returned to its previous level.

AI could personalise. But without local context, it became very efficient at being right about the wrong person.

Where AI does work here

Start internal. Research, analysis, reporting and operations are places where a wrong default costs you time, not trust.

Give it local context before you let it talk to customers: spelling, currency, regulations, seasons, examples, tone. If it can't pass as local in a test, it isn't ready to be local in production.

Give your local team a veto. Not a review step that gets skipped when deadlines tighten, but the right to say "not here" and be backed.

Big market, small market

In a big market, AI's mistakes are noise.

In a small one, they're the message.