Tuesday 26 May 2009

Time for segmentation?

puzzle alarm

There was an interesting article on Wired.com the other week discussing some new research about email send times and how these can be used to derive useful information about individuals.

The article related to a paper by Yahoo Research and Northwestern University entitled “Characterizing Individual Communication Patterns”. They state within the introduction to the paper that:-

Whilst demographic details like gender, age, race, education and income are generally used to segment people, they can be expensive and time consuming to gather and more importantly, are often poor predictors of outcome.

Instead they argue that behavioural data – in the case of this paper email usage – can be a much better way of characterising individuals.

The paper goes on to describe two distinct groups they identified from email usage patterns - particularly time of day - and how these could be used to help in the fight against spam.

What interested me however was the thought of segmenting customers by usage, specifically date and time.

We would normally utilise behavioural data to segment customers, looking at pages browsed, items viewed, goods purchased, stores frequented – anything that indicates you did something. The issue is we typically look at the “something” – not the when or the where.

Web analysis has for a long time utilised time of day (site usage) to help segment people as for many sites - especially those without any formal member accounts - this can be some of the only data available.

A case study for Denmark’s most popular Internet portal Jubii showed that creating profiles based on when customers used the site helped to target online advertising and increase click-through rates by 30-50%.

Within loyalty programmes where we typically have an abundance of both demographic and transactional information – we sometimes discard seemingly less important information like time when creating customer segments.

What this research shows however is that time on its own can be a power profiler of customer behaviour.

It’s kind of obvious at a simple level that time of day would provide an element of segmentation. If comparing retail spend transactions - someone regularly shopping in DIY during the middle of the week is likely to be very different to someone purchasing DIY only at weekends. Both customers show an interest in home improvement but one is more likely to be retired and the other working full time. Of course there may be many other reasons, but this simple example shows how just adding the dimension of time can begin to elicit a wealth of undeclared information.

We have recently experienced this ourselves when doing some research and analysis on a frequent flyer programme. Whilst I can’t divulge the details of the research, what it did show was that rather than what you purchase, it was when and where you made a purchase that was highly predictive of the likelihood of taking out an additional service.

Running a loyalty programme and not using the information it provides, is akin to simply running a deferred discount – and you’d probably be better off just doing that.

However those that do use the data are sometimes in danger of gathering too much information in the hope that it will provide greater insight and I think what this research shows is that it can be a case of less is more.

Certainly early in a relationship, behavioural data will be sparse and declared information limited – using everything you have at this stage including information like time could provide a better way to predict likely future behaviour and fast track these customers to programme benefits earlier.