A team spends a quarter building personalised messaging for six audience segments. The segments came from a workshop rather than from data.
The result is six variations of a message that differ in adjective. Conversion moves by an amount indistinguishable from noise, and nobody can tell whether any individual segment improved.
In the same quarter, nobody added the price range to the pricing page. That single omission was costing more than every segment combined.
1. Fix relevance before attempting personalisation
A page that answers the visitor's actual question serves every segment well. Personalising a page that does not answer it produces six versions of the same failure.
2. Use the context you already have
The page they are on, the campaign they arrived from, the time of day, the device and whether they have visited before. All reliable, all free and all more useful than inferred demographics.
3. Personalise by intent, not by identity
What somebody is trying to do matters more than who they are. A first time visitor comparing options and a returning customer with a problem need different things regardless of industry or job title.
4. Start with the returning visitor
Somebody who has been before is the easiest and most valuable case. Acknowledging that they have returned, and continuing where they left off, is worth more than any segment model.
5. Use the page as the segment
A visitor on a specific product or service page has told you what they care about. Tailoring to the page is simple, accurate and requires no data project.
6. Use the campaign as context
What was promised in the advertisement should shape what is emphasised on arrival. This is the cheapest and most effective personalisation available.
7. Do not personalise things that should just be correct
Coverage, price and availability are facts rather than messages. Making them visible to everybody outperforms showing a tailored version to some.
8. Be careful with anything that feels observed
Referencing information a visitor did not knowingly provide is unsettling and reduces trust, which costs more than the relevance gains.
9. Keep the sensitive contexts entirely neutral
In health, legal and financial settings, tailoring can be intrusive or inappropriate, which is the discipline in keeping clinical questions out of dental website chat.
10. Test against a proper baseline
Captured before anything changed. Personalisation projects frequently cannot be evaluated because nobody recorded what the unpersonalised version did.
11. Measure per segment, not in aggregate
A blended improvement hides the segments that got worse, which is the measurement discipline in conversion rate questions marketers ask about chat.
12. Beware of segments invented in a workshop
A segmentation drawn on a whiteboard describes how a team thinks about its market rather than how visitors behave. Segments worth building for come out of what people actually do on the site, and they rarely resemble the ones anybody named in a meeting.
12. Keep the attribution honest
Personalised journeys complicate credit assignment considerably, which is the problem set out in attribution questions when chat is in the funnel.
13. Let the conversation do the tailoring
Asking somebody what they need and responding to the answer is personalisation that requires no model, no data project and no assumptions, and it works from the first visitor.
14. Count the cost of maintaining it
Every variant is a thing that must be updated when a price, a policy or an offer changes, and six versions of a page means six places for something to go stale. A great deal of personalisation quietly becomes a set of pages nobody has checked in a year.
14. Use what they ask to improve the page for everybody
The questions people put to an assistant are the best copy research available, which is exactly what chat questions improve website copy describes. See features, start at try, and the sector view sits under marketing agencies.