Less is more: TeliaSonera targets for value with two-stage modeling

Direct marketers traditionally regard response rates as a measure of a successful campaign. However, some customers who are hard to activate can deliver high revenues. Should they be ignored or pursued? TeliaSonera used SAS Telecommunications Intelligence Solutions to target customers for value and reach them at low cost.

With 22 million customers, TeliaSonera is the leading telecommunications services provider in the Nordic and Baltic region. It also has a strong presence in Russia, Turkey and Eurasia. TeliaSonera offers customers services within mobile communications, Internet, data communications and fixed telephony. In Finland, where it operates under the Sonera brand, TeliaSonera is the market leader in mobile telecommunications and had total sales of SEK 81,772 million (€8.9 billion; US$10.9 billion) in 2003.

Like other telecommunications companies operating in a largely saturated market, TeliaSonera is always looking for ways to boost average revenue per user (ARPU) by selling new services to its customers. To achieve its goal of maximizing profit, the company has to engage in a lot of well-targeted marketing activities.

Direct mail is one of the most important channels to achieve this, but it is very difficult to predict whether a direct mail campaign will be profitable. Traditionally, the success of a campaign is measured on response rates; but on low-margin services, a good response rate does not necessarily mean the campaign will deliver a return on investment. Marketing analysts at TeliaSonera in Finland demonstrated that if you leverage customer intelligence correctly and target the customers who offer greatest revenue potential, you might secure greater profits while actually reducing response rates.

One of the booming areas of mobile usage in recent years has been SMS messaging: TeliaSonera's 2.4 million customers (more than half the adult population in Finland) sent 815 million messages in 2003. A potentially more profitable, but as yet underused service is MMS (multimedia messaging service). MMS is a store and forward messaging service that allows mobile subscribers to exchange multimedia messages (combinations of text, picture, audio/voice and short video clips) with other mobile subscribers. In the first quarter of 2004, TeliaSonera customers sent just 645,000 MMS messages.

"Our aim was to activate MMS usage profitably through a well-targeted direct mail campaign," says Riku Mäkeläinen, Development Manager, Consumer Customers at TeliaSonera. TeliaSonera decided to use two products from the SAS Telecommunications Intelligence Solutions suite: SAS Cross Sell and Up Sell for Telecommunications for cross-selling data analysis and SAS Marketing Automation to execute the campaign. Together they would be critical to ensuring profitability, first by identifying the right customers, and second by keeping campaign costs to the minimum.

"Traditional one-stage modeling identifies people who are likely to buy a product or service and are therefore easy to activate," says Mäkeläinen. "Unfortunately, it tends to overlook those customers who are hard to activate, but who would yield much higher revenues. It occurred to us that it is worth the extra investment to activate these potentially high-value customers."

TeliaSonera started by carrying out some preliminary analyses on a set of active MMS users from the specially selected target segment. One-time and random users were excluded from analysis. Mäkeläinen identified that on average an activated user brings in only five times the campaign cost per customer.

Data preparation involved making a stratified sample containing 25 percent target events and 75 percent non-target events, and partitioning the data between 60 percent training and 40 percent validation. Outliers were filtered and variables standardized to create a data set with a distribution of variables that represented the behavior of the mass market. Mäkeläinen then tried a variety of SAS data mining techniques, finding that logistic regression was the method that gave the best results. This first-stage modeling gave an impressive lift value, showing that just with traditional methods it was possible to conduct a profitable campaign - giving a positive profit. But Mäkeläinen was confident that even greater advantage could be gained from using SAS.

So he applied the Two Stage Model Node in SAS Enterprise Miner. "It's very easy to use. You just make a few selections and the tool runs all by itself," says Mäkeläinen. Stage one used decision-tree modeling to determine the activation probability, while stage two used logistic regression to assess the revenue that would result from usage. After assessing the models, customers were selected if their activation probability predicted revenue exceeding the cost of contacting them. On this occasion, the indicated profitability of the campaign (extrapolated from the sample) rose to approximately double, a 118 percent increase over traditional one-stage modeling. Still, Mäkeläinen believed it was possible to improve even on this figure, by using the advanced features of SAS Enterprise Miner rather than the automated tool. "SAS Enterprise Miner offers power and flexibility that an experienced data miner can leverage to get optimum results," he explains.

Based on his business knowledge of the data, Mäkeläinen tweaked the model manually -- for example, by using the R-square selection method to pre-select variables for modeling and to override some rejections. Upon assessment, the profitability figure went up again - an increase of 7.24 percent over the automated Two Stage Model Node approach and an increase of 134 percent over the traditional one-stage regression.

"The Two Stage Model Node in SAS Enterprise Miner is easy and fast to use, even for a non-specialist, and it showed we could make a considerable increase in profits. Manual two-stage modeling can boost profits even further and gives the data miner greater control and flexibility over the models, but is rather more demanding on the analyst's time. The question as to whether profits justify the extra time required will depend on the business case for the project, of course. For future and larger activation campaigns, it could bring in a lot of additional revenue.

"But the point is that in SAS Enterprise Miner, you have both options," says Mäkeläinen.

Having done the analysis, how does one apply it to marketing campaign execution? Clearly, those customers who are more difficult to activate, but have higher revenue potential, require a rather different approach. Teliab used SAS Marketing Automation to split the message between campaign groups in the target audience. Based on the scores assigned in the Two Stage Model, customers were extracted from the data warehouse, and contact policies were applied.

"We do not want to bombard customers with a lot of messages at the same time. SAS Marketing Automation makes marketing evaluation easier and minimizes the overlaps in marketing programs. It also cuts the cost of campaigns, boosting profits. Last but not least, it helps us to maximize customer satisfaction," says Sakari Forslund, Development Manager, Consumer Customers.

"We don't waste time manually setting up groups. They are created automatically, which helps keep costs down," says Forslund. SAS Marketing Automation schedules mailings optimally, for example to send a slightly different message to customers who do not respond to the first. "A one-shot campaign is not enough. Repeating improves performance, if it is done sensibly," adds Forslund.

TeliaSonera has been successfully using SAS Enterprise Miner for some time, and Mäkeläinen now sees advantages in combining it with campaign execution tools from the same supplier. "It eliminates integration costs, and it is convenient to deal with one local company. Added to which, we know SAS is going to be there to provide support whenever we need it," he says.

Copyright © SAS Institute Inc. All Rights Reserved.