AI makes it cheap to produce many versions of a message. That makes personalization possible at a scale that used to need large teams — and makes it easier to get wrong at scale.
Levels of personalization
- Segment: different messages for a few audience groups (industry, life stage, region).
- Context: messages that change with the moment — season, weather, location, product viewed.
- Individual: content chosen or generated for one person, based on their behaviour and preferences.
Most value usually comes from getting segments and context right; individual-level AI personalization needs strong data, consent and testing.
Why relevance matters
Buyers punish irrelevant outreach. A Gartner survey of 632 B2B buyers (August–September 2024) found 73% actively avoid suppliers who send irrelevant outreach, and 61% prefer a buying experience without a sales rep overall. Personalization that is wrong, or that reveals more than the customer expects you to know, damages trust.
A real example at scale
In India, Cadbury Celebrations' "Not just a Cadbury ad" campaign (Ogilvy and Wavemaker) used machine learning to recreate actor Shah Rukh Khan's face and voice so that ads could name local shops. WPP reports 130,000 personalised ads featuring about 2,000 small businesses, with 94 million views. It worked because the star took part, the purpose (helping local shops) was clear, and the brand controlled what was said.
Rules of thumb
- Use only data the customer would expect you to use, with consent.
- Write relevance rules: which signals you use, and which you never use (for example health or financial hardship).
- Keep a control group to prove the personalised version adds value.
- Run every generated variant through brand and claims checks.