Customer Tracking with BI

Tracking good (and bad) customers is easy with BI

BI is about extracting information from data. We have various techniques at our disposal – aggregation, OLAP, data mining and so on. Often we can take the raw data, run one process upon it and lo, information appears. But there are times when we may benefit from taking raw data and running multiple, overlapping BI processes on it in order to get what we want. In one sense we have to be careful about doing this – we could easily drift absent-mindedly into the territory of averaging averages. On the other hand, if we are careful the rewards, in terms of gaining a greater understanding, can be huge.

Take RFI for instance. That’s a Request For Information isn’t it? Well, yes and no. Yes, it does stand for that but in the business intelligence world it also stands for Recency, Frequency, Intensity. (By the way I've also heard it said that the R stands for 'recentivity' but can't bring myself to use such an inelegant word - 'recency' is bad enough.) RFI is a type of analysis which aims to produce a thumbnail sketch of customer behaviour and I’ll walk you through how it is used.

The important point here is that RFI is simply an example of the more complex analysis that we can perform in BI. You may have absolutely no interest whatsoever in customers and how they behave, but you could still have need of the kind of analytical process that RFI exemplifies. In the main RFI analysis is applied to sales data stored in a warehouse, for sales made in stores, over the web or from call centres, or any permutation thereof. Any business can benefit from knowing what its customers are up to. This is, of course, the basic idea behind that other abbreviation, CRM (Customer Relationship Management). The requirement to identify the big spenders is commonplace, but it can also be worthwhile identifying those who soak up staff time (in processing returned goods or in chasing outstanding payments) but don't buy much, or even anything, from your company.

RFI analysis is a method of identifying clusters of customers exhibiting the same behaviour and then tracking how that behaviour changes over time. The first step is to define what is meant, within your company, by Recency, Frequency and Intensity? Here are some examples: · Recency is the time that has elapsed since there was any contact between our company and a customer. Contact can be instigated by either party, for instance it could be the customer's visit to our web site or a monthly call to a customer from a member of the sales team. So, for your company you might identify five flavours of recency:
· A – customers with whom contact has been made in the last week
· B – contact within the last month
· C – contact within the last 6 months
· D - within the last year
· E – contact at some stage since life evolved on planet Earth 

Frequency is the number of occasions there has been contact with the customer over a specific period, a year, a month, whatever.
· A – 12 times in the last year
· B – 5 times in the last year
· C - once in the last year
· D - zero times in the last year

Are these values appropriate for your company? I have no idea. But RFI is not about adopting a fixed set of parameters, it is about deciding on the parameters that your company finds important. 

Intensity is the least self-explanatory: it's an indication of how worthwhile the contact with a customer has been. This will often be the money a customer has made the company, but can encompass other aspects of business such as, for instance, the number of searches undertaken from a web site.

I have kept these definitions simple but there is no reason why you cannot have several factors contributing to each of the three main categories. Once we have values for these three indicators for a customer, we can plot each customer as a single point in three-dimensional space. Then we can use data mining techniques to identify groups of customers displaying similar behaviour. Each group can be described in terms of its main characteristics and we might arrive at categories such as:
1. Frequent customer
2. Occasional customer
3. Non-spending customer
4. New customer (with the potential to change category over time)
5. Previous frequent customer, now inactive
6. Frequent, non-spending customer

These are somewhat simplistic categorisations: by data mining we could also include aspects of behaviour, for instance whether a customer is a habitual returner of goods or habitually slow to pay or even a 'difficult customer'.

The number of clusters/behaviour types depends on how the data mining is performed. The value of these thumbnail sketches of customer behaviour becomes greater with the passage of time. RFI values could be calculated every month and the cluster within which a customer falls re-calculated.

Over, say, a six-month period a customer might develop a history like this:
4, 4, 2, 2, 1, 1

This could indicate a new customer who starts by checking out your company, moves into a period of experimental purchasing before becoming an enthusiastic buyer. Alternatively, a pattern like this:
1, 2, 2, 3, 3, 5

could indicate a growing disenchantment with your company. In the first scenario, special offers of some kind might keep the customer in the most profitable category, and in the second, investigation into why dissatisfaction has set in could turn the situation around and restore the customer to the most profitable category.

The bottom line is that RFI type analysis can often allow us to summarise very complex data into simple information. The trick is to do this without simplifying too much.