5 Ways to Cut Costs with Predictive Analytics5 Ways to Cut Costs with Predictive Analytics

I was already in Washington D.C. for the Teradata Partners user conference this week, so I figured I'd stop in at the Predictive Analytics World event in nearby Arlington, Virginia. Tuesday morning's keynote by event chairman and founder Eric Siegel, pictured below, offered a nice primer on "Five Ways to Lower Costs with Predictive Analytics."

Siegel's presentation offered a primer on five popular forms of predictive analytics: response modeling, response uplift modeling, churn modeling, churn uplift modeling and risk modeling. In the process of describing each approach for segmenting customers and improving marketing performance, Siegel offered the following tips: 1. Don't spend money on those who are not as likely to respond. The response modeling technique helps you figure out which customers are most likely to respond to marketing offers, based on past history and customer segmentation, and those who are lost causes. For example, if 40 percent of customers are determined to be likely responders and you mail promotions only to them, you'll increase lift while saving 75 percent on marketing costs. 2. Don't waste money on those who would have purchased anyway. Response uplift marketing (a.k.a., net lift modeling, incremental modeling, impact modeling and differential response modeling) lets you segment out customers who would be less inclined to buy if you send them a marketing offer (do-not-disturb customers), those who won't buy whether you send them an offer or not (lost causes), those who will buy whether you send them an offer or not (sure things) and those who might buy if you send them an offer (persuadables). The point here is to forget about the sure things (who will buy anyway) and save your marketing bucks for the persuadables. 3. Don't waste money retaining customers who will stay anyway. Churn modeling lets you spot customers who are likely to bolt and those who are likely to stay. Since retention offers are generally expensive, entailing discounts for renewals, premiums and so on, it's crucial not to be indiscriminate with these types of campaigns. 4. Don't trigger departures by those who would have otherwise remained as customers. Churn uplift modeling helps you spot "sleeping dogs." Think of health clubs sending out promotions to renew annual memberships or cell phone companies sending out reminders that two-year contracts are about to expire. You may just be reminding them to end their membership or to shop for a new provider. Without a reminder, these customers might have taken no action. 5. Don't let thieves get away with it (or spend less screening them). Risk analysis helps insurance companies spot risky customers before signing policies (can you say, "preexisting conditions"). The technique can also spot sales prospects that aren't likely to convert or customer prospects who are less likely to be profitable. In the case of fraud, risk analysis helps you waste less time on transactions or applicants that are likely to be denied. Or you can achieve current risk-detection success levels using more predictive analytics and fewer analysts.

Going beyond these five tips, Siegel listed additional ways in which predictive analytics could cut cost, including forecasting so you don't acquire or produce too many goods, avoiding hiring unreliable employees, acquiring better (more profitable) customers, improving cross selling, improving clickthroughs and so on. He concluded with the key advice to always hold aside a control group so you can:

I was already in Washington DC for this week's Teradata event, so I figured I'd stop in at Predictive Analytics World in nearby Arlington, Virginia. Tuesday morning's keynote by event chairman and founder Eric Siegel offered a nice primer on "Five Ways to Lower Costs with Predictive Analytics." Here's the list...