Innovation: Netting stray clients

RADIO RENTALS is using neural networks to predict which customers are likely to terminate television rental agreements. The technique is 30 per cent more accurate than existing methods in spotting customers who are about to defect - customers who might then be tempted to stay with a targeted promotion.

Neural networks are programmes that learn by experience. The system is trained to identify which customers are most likely to terminate an agreement by studying a sample of 5,000 who have maintained rental agreements in the past six months and 5,000 who have not, all selected randomly to be as representative as possible of the whole customer database. By analysing this data, the neural network 'learns' how to score parameters such as payment and rental history to highlight wavering customers.

The system is paying even greater dividends in detecting customers who are likely to upgrade their equipment. This is improving the effectiveness of Radio Rentals direct marketing, because specific campaigns and promotions can be offered to customers who are most likely to respond. It has also enabled Radio Rentals, part of Thorn UK, to make cost savings on direct mail by reducing time, manpower and materials.

The neural network system, developed by Central Research Laboratories, also a Thorn company, can be tailored to fit any marketing database. It can also be used to predict other types of behaviour, such as propensity to buy. Jerry Severwright, project manager for the development at CRL, said the system would allow companies to extract extra value from their existing customer data.

'It is accepted wisdom that it is cheaper to keep customers than to find new ones, and using neural network techniques will improve a company's ability to do this.'

CRL has been researching neural networks since 1986, but did not realise until 1991 that they could have important applications in direct marketing. Since then CRL has worked with its sister company to implement the technique, and is now seeking outside customers.

In one trial based on historical data for a finance house, which sells personal loans by direct mail, the neural network system achieved a 50 per cent improvement in the response rate achieved by the existing linear regression method. Mr Severwright said using neural networks to analyse customer data could also be a cheap alternative to market research.

'It is very expensive to conduct interviews, and surveys are usually limited to a few hundred people. Companies could instead analyse an extract of 10,000 customer records with a neural network system to look for behavioural patterns.'

Another application would be for companies such as banks, which use direct mail to sell a range of products to existing customers. Because they regularly post statements, the cost of inserting an extra piece of paper promoting personal loans, mortgages or whatever is marginal. However, it creates a bad impression to try and sell a mortgage to someone who has just taken one out. Using neural networks to analyse how a particular customer uses the bank will make it possible to promote only the products he or she is likely to be interested in.

'Applying neural networks to customer behaviour is a fairly small part of the sales and marketing process. You still have to work out how to approach the customer. But the sort of inferences neural networks allow you to form makes direct marketing much more powerful,' Mr Severwright said.