If you have more than 10,000 customers or prospects in your B2B universe, you might be able to boost performance substantially through predictive modeling.
Posted Jul 26, 2004
Your sales force may be doing well or it may be struggling, but if you have more than 10,000 customers or prospects in your B2B universe, you might be able to boost performance substantially through predictive modeling.
Predictive modeling uses outside information like economic and firmographic data combined with information you already have to predict account behavior. With predictions of total spending, attrition, next purchase, acquisition rate, and lead conversion, a company can wisely invest sales efforts and marketing dollars.
Where to start? The biggest bang for a first-time analytical investment can come from a spending model. Not only can you target better, but the scores can also be used in other valuable ways. The model we recommend has two stages. The first stage predicts probability of purchase over a certain time period (e.g., six months). The second predicts the expected dollar spending over that same time period. Combine the two or use them separately based on your business need.
Examples of applications:
1. Targeting: Invest your resources against accounts most likely to buy and spend the most.
2. Better sales forecasting: Use the spending estimate as a sanity check against field numbers to improve accuracy.
3. Reallocate marketing investments: Add predicted dollars to your value segmentation. Target your high-current/high-future value accounts with retention messages. Limit your investment in low-potential accounts.
4. Fairly allocate accounts among reps/territories: Evaluate each rep's portfolio based on predicted value. Consider adjusting quotas or moving accounts to ensure coverage on high potential accounts. Allocate orphaned accounts more equitably.
5. Set quotas and adjust comp plans: If you have predicted purchase rates and spending, you can set a reasonable, data-driven quota.
a large enough account base--10,000+ accounts with 500+ purchasers in your target time frame
a range of spending amounts--or you cannot predict spending
three to five years of purchase data
data about sales, billing and other contacts at each company location
an outside source of demographic or firmographic data if you want to predict spending for prospects
To build the model you need the right staff. For example, if you don't have the right analysts on staff, your CRM database vendor or services firm might have an analytic group. The skill set is specialized, and an off-the-shelf tool combined with inexperienced modelers will give you bad results. Statisticians generally have an MS or Ph.D. in statistics or economics. Expect to work with them beyond model delivery so your organization is trained to understand and use your predictive toolkit properly.
You are now ready to deploy and measure. Your IT staff will load the data into your sales system and/or marketing database. Model scores or spending segment assignments should be added to existing reports to track how well you are doing. Rescoring will keep the predictions up-to-date. Many companies update the predictions monthly or quarterly. Your statisticians should track the performance of all of the predictors and determine when the equations must be tweaked or the models rebuilt.
About the Author
Lynne Mysliwiec is vice president of analytics at relationship marketing firm Epsilon. Mysliwiec has more than 10 years of experience in analytical research and consulting to support CRM. She develops marketing programs that improve customer retention, maximize customer revenue and profitability, acquire high-quality customers, and improve sales force performance in the B2B marketing area.