For many years, technology vendors have promised companies systems that provide the “one true view“ of their customers. CRM vendor PeopleSoft had the 360° View (somehow lost during the acquisition by Oracle); other CRM vendors provided insights into past customer behaviour and analytics vendors touted clever ways of predicting future behaviours based on visualising past activities through graphical and interactive dashboards.
The main problem with such systems lie in that they are pretty dependent on having enough past information to work against, and in analysing large data sets to provide the required visualisations – which can require large compute farms and data warehouses. That the future predictions take time to come through can also be a problem – the aim is to capture customer activity in real time and make the most of them.
Some approaches managed to get close to giving real-time value through using pattern matching – if a given customer is doing this, then based on past behaviour, we should point them in this direction. Makes sense, but requires deep analytics of past data (again) and the formalisation of the rules that will need to be in place.
Quocirca recently spoke with Featurespace, a Cambridge Ring company started in 2005. The company is currently touting itself as a customer retention and fraud identification and management company – but there seems to be a lot more underneath the hood.
Featurespace uses the real time data streams for its main feeds. It is self-learning and can work against minimal historical data. Through using advanced algorithms for analysing on-line (or other – see later) behaviour, fraudulent activity can be identified at a very early stage, and actions taken to curtail it. Yes, this has value to a business, but will only tend to be seen as massively valuable by the Chief Risk Officer (or equivalent). Customer churn is an accepted occurrence in most markets, and as long as a company sees its churn as being no worse than the industry average, they are likely to stick with what they have.
The trick for Featurespace is to take what it has and create messages that have better value to businesses. For example, behavioural analysis not only identifies bad behaviour, but also good behaviour. In real time, customers can be encouraged in their good behaviour, spending more in the process and ensuring that shopping carts are completed and the customer-to-cash process is fully optimised.
Also, bad customers can be easily identified – the bane of markets such as telecoms, where the top 20% of customers make 80% of profits, and the bottom 20% make 80% of the losses. Behavioural analysis can identify whether there is any hope in turning the customer through to profitability – if not, then bidding them a fond “farewell“ (maybe even offering them a £5 voucher to go to the competition) can improve profitability – and lower churn, as many of these bottom 20% are the ones that hop from deal to deal.
Such cluster analysis can lead to identifying interesting opportunities that many analytic approaches miss – and if supplemented with other data, such as the (somewhat outdated, but still widely used) ACORN scoring, can further be used to optimise offers at an ad-hoc immediate level and a strategic future product or services level.
Featurespace can help in the on-line retail space in optimising customer behaviour, but it is also showing how it can operate outside of the “standard“ markets. For example, it can analyse video streams. Imagine at an airport: your average traveller is doing all the “normal“ things – gawping at shops as if they have never seen them before; coming to a halt at the bottom of escalators and causing others to fall over behind them.
Consider someone who is not a normal traveller – a terrorist, say. No matter what they do, their state of mind will not make it possible for them to look as relaxed or normal as the average passenger. Tracking all behaviours enables differences to be picked up very rapidly – and it doesn‘t have to be hidden in how it is used. No matter how aware the person is of the system, they cannot work around it: their behaviour patterns will just look more false the more they try to be normal.
Featurespace has to change its messaging, and the new(ish) CEO, Martina King, knows this and is going to be making a big push around Featurespace for behavioural analytics.
There are competitors out there – the big one that springs to mind is IBM with the work that Jeff Jonas has being doing for some years. However, there is more than enough room for other players, and Featurespace looks like it could well be one to watch.