Predictive Product Keeps Debtors’ Prison Empty

  • Written By: D. Geller
  • Published On: July 26 2000



Predictive Product Keeps Debtors' Prison Empty
D. Geller - July 26, 2000

Event Summary

SLP InfoWare specializes in CRM (customer relation management) systems that predict customer behavior. The company's flagship product is Churn/CPS, an application for determining which customers are likely to switch loyalty from merchants such as cell phone providers and ISPs (see TEC article, "Do We Already Know Whether You're Going To Read This Article?"). The "CPS" in the name stands for "Customer Profiling System," the core engine behind all of the applications. The latest product, BadDebt/CPS, attempts to identify customers who are likely to fail to pay bills.

BadDebt/CPS joins applications for making up-sell and cross-sell recommendations, predicting repeat buyers and exiting buyers, and for customized behavior analysis, in addition to churn prediction. The company boasts a scalable architecture that makes it easy to integrate its suite with other front and back office applications.

A user of this or any of SLP InfoWare's products will begin using their modeling environment to accumulate the data to be tracked and define the behaviors to be predicted - in this case, which customers end up subject to debt collection. The data that you select are split into two sets; one is used to create the model, and the other is used to validate the model. Creating and validating may require repetition until the performance on the validation sample is satisfactory. The resulting model is made available for use in making real-time recommendations or in creating campaigns. This distinguishes SLP InfoWare's products from those that provide only analytic tools. The predictive knowledge is available to be used at all customer touch points to drive one-to-one targeted real-time campaigns.

Market Impact

Applications like this are becoming increasingly common among CRM providers. Established CRM companies are following the lead of startups and behavioral analysis specialists in adding data mining and predictive modeling. Although some of these products may reflect genuinely new research, by and large the complex algorithms are well understood and available to anyone, although there's as much art as science in adapting and tuning the theory to a particular application. SLP InfoWare has established a presence in this area and should be able to cross-sell this new application into its existing customer base, as well as selling it to new markets.

The company has its roots in European telephony and Internet providers, these representing the types of data that were used to create and validate the product. The current business plan calls for adding a variety of predictive modules and leveraging the company's current client base and expertise to become a lead player in the emerging mobile computing market.

The more general e-commerce space, which already has a number of eCRM players, is of interest but not a top priority at this time. However, the company has such customers as Chteau Online, an online wine merchant. The company demonstrated its ability to build product association rules and to predict repeat buying behavior. By identifying characteristics of repeat buyers and targeting e-mail promotions to them the system increased repeat buying by 60%.

SLP InfoWare has a good head start, especially in its current and target verticals, but can expect increasing competition going forward. While SLP may be willing to take a hands off attitude toward the application of e-CRM to e-commerce, companies like NetGenesis and E.piphany currently in that market are not likely to forgo the burgeoning mobile computing arena.

User Recommendations

SLP InfoWare is a company that offers a wide range of predictive CRM products, and seems likely to be bringing out more in the future. This makes them particularly attractive to companies that will want or need more than one such type of analytic application. Since SLP InfoWare's traditional customer base is in industries like telecommunications and finance, other types of business will want to probe harder for relevant experience.

Almost any predictive CRM product is likely to give you results that make a demonstrable difference to the bottom line. We almost take it for granted that the majority of the products can pass your ROI test. Most of the companies in this line of work know that, and capitalize on it by offering one or another kind of "try before you buy" program. It is certainly wise to take advantage of such programs, but doing so usually turns out to involve enough of the integration work that the product signing is almost inevitable, because of the work involved in the setup.

To get a better handle of which of the companies on your short list for any CRM product are likely to produce the best results, we recommend creating a test that can be given to a number of vendors simultaneously, similar to the modeling step described above. To do this, establish during the initial round of interviews what data they would want to work from to do a demonstration. The fields will vary based on the type of application, but the ideal here would be a statement on the order of "with n months of past data we can predict churn over the subsequent 2 months."

The number of past months n should be on the order of three to six. You'll also want to have each vendor name its level of accuracy in identifying targets and its expected false positive rate. A false positive is an identification of a customer as a member of the class (likely bad debts) that turns out to be incorrect. Your ROI calculations should include the cost of false positives, which might be measured in lost business (turning away good customers) or unnecessary promotional expense (offering special deals to retain customers who would stay without them).

From the responses provided by the candidates that are most interesting, prepare a data set in a format that all of them can use, such as a comma separated text file. Go back far enough in time so that the most recent two months of data are not included; you will want to use these to verify the predictions. With such a common data set (even if it does not represent the full richness of your data) you will be able to develop a comparison of how well each product does on real data; while getting the data into a form that will make it easy for the vendors to agree to the test will take up some of your staff time, the end result will be most useful in quantifying the expected payoff from each product under consideration.

 
comments powered by Disqus