The customer churn rate for our mobile telecoms business has been rising for the past few months. What can I do to reverse this?
Simon James replies: First, never trust anyone who tells you that you can build one predictive model to solve all your woes. Different groups of people churn at different rates for different reasons. It is important to identify natural moments of truth where churn is most likely to occur.
For example, the first three months generally have an abnormally high churn rate. These people have yet to build any brand equity in your company and require a specific strategy. This should revolve around welcoming a new customer to your company, reinforcing the positive decision that person has made and educating them on all the functionality of their phone.
Another example of a moment of truth is at the end of your first 12-month contract. This is a natural watershed in a customer's relationship with your brand. Many mobile companies attempt to re-sign their customers with handset upgrades or realigned price tariffs. However, the benefit achieved in churn reduction is often outweighed by revenue dilution. If this is the case, be very careful about who you are attempting to save.
Once these specific instances of vulnerability have been identified, customers should be segmented into like-thinking and like-acting groups, and a churn model built for each one.
The weakness with most predictive churn models is that they are poor at predicting when churn will occur. This needs to be addressed by building time-sensitive variables that describe events that have occurred recently.
Someone complaining, or telling you they are going to leave, is the most predictive variable you'll ever have.
I work for a financial services company. How can I use lifetime value modelling across my product range, including life insurance, pensions and home and car insurance to predict the most profitable customers?
Scott Logie replies: First, try not to predict too far into the future.
I don't have much faith in lifetime value figures much beyond 18 months ahead, as there are just too many unknowns.
Then make sure that you have the information available to carry out this work - such as income transactions and customer communication frequency and cost. Do you know which communications each customer has responded to and the transaction value?
Keep things simple. Do you need an estimate of lifetime value, or do you need a figure precise to the nearest penny? How are you going to change your strategy based on the results? You probably won't have completely accurate information available across all your products and all your customers, so a broad-brush approach may be the most appropriate.
Think about the view of the customer. If someone has 拢50,000 in a savings account with a low interest rate, consider this customer to have invested quite heavily in your company and that they should be recognised. In addition, look at the cross-relationships. Is lifetime value determined by income alone? I don't believe it is; loyalty should also be taken into account.
We hold channel preference information (mail, e-mail, SMS, telephone) on our customer database, but each is treated separately. This causes conflicting messages and no coordination of content or timing. Mark Patron replies: To achieve better channel integration you first need to build a single customer view, so you can understand how each customer wants to be communicated with. It needs to pull in data from right across the organisation to ensure you understand all the interactions.
You should then analyse the value of each customer. Some channels are more expensive than others and you need to ensure that you are using the most appropriate channel. Banks are a good example of this. Their most valuable customers use the cheapest channel, the internet, and their least valuable customers use the most expensive channel, the bank teller.
The next step is to establish a coordinated communication strategy. How many times will each segment be contacted? Ensure that different channels don't contradict each other. Finally, record all communications with each customer and continually profile and analyse the campaigns to understand what is working and refine the communication strategy accordingly.
The decision as to which mailsort discount scheme to use has become less straight-forward with the advent of Mailsort 120. How do I know if I should be using the Mailsort 120 or Mailsort 1400 routines, in terms of cost and efficiency?
David Laybourne replies: Postage is one of the most costly elements of your campaign and the right level of sortation can reduce those costs by over 40 per cent. A bureau will be able to advise you or undertake test mailsorts at each level to find the best discount structure. It will need some key information from you to provide this advice, such as envelope size, pack weight and typefaces.
The bureau cost to sort data to the 120 or 1400 scheme is the same. However, the Royal Mail discounts are different.
As a general rule, the smaller your mailing quantity (say, less than 20,000) the more likely it is that Mailsort 120 will offer a better discount.
If your low volume mailing is going nationwide, then this rule applies even more, as the best discounts in Mailsort 1400 come with large quantities of mail being routed to each post town.
To have our experts solve your data problems, send your questions to: noelle.mcelhatton@haynet.com
DATA DOCTORS
Simon James heads Zalpha's analytics team. Before this James was at Carlson Marketing where he was head of data planning.
Scott Logie is managing director of Occam, a database solutions provider. Prior to joining Occam, Logie was head of analysis at Bank of Scotland.
Mark Patron is a non-exec director of data managers thinkdata. Previous roles include one as managing director at Claritas.
David Laybourne is technical director at DPS Direct Mail and has over 20 years' production experience in bureaux and DM agencies.