The goal of predictive modeling is to either increase response rates to a given mailing quantity or reduce the mailing quantity without suffering a drop in response rates. This means you can control and use models to drive up your response rates and/or reduce your expenses more than you can with a random sample. Working with an experienced statistician and analyst, our model design, creation and development are free of charge with the understanding that an essential element in this process is a reasonable sample of your past two direct marketing campaigns.

Once available, your customized model is applied across our entire database of 160 million names, all of which are purchase-based, transactional records identifying the type of transaction, product or service bought as well as recency, frequency, and monetary value. Add psychographic and demographic characteristics where needed and the result is a highly qualified list of names that reflects the profile of your best customers.

Lift and gain tables are useful tools for measuring the value of a predictive model versus random sampling.

To better understand lift and gain, consider the example of a company that wants to do a direct mail marketing campaign. The company has a database of 100,000 potential customers, and they calculate that each mailed advertisement will cost $1.00. Prior experience has shown that the average response rate is 10%. So if they send the advertisement to all of the randomly selected prospects, they will incur an expense of $100,000, and they will likely receive approximately 10,000 units sold at $20 per unit.

Hoping to improve their return on investment (ROI), the company builds a predictive model using data from previous campaigns and various demographic variables. The predictive model is used to prioritize the prospects so that they can be sorted by scored rank in decreasing order of expected sales.

Using the cumulative percentage of mailings and the percentage of total responders from the Lift/Gain table below, the marketing director of the company prepares the following table adding expected sales results, cumulative gain, and lift columns:

The table divides the total prospect set into 10 “buckets” by cumulative percentage of mailings, with the best 10% of the prospects in the first bucket, the second-best 10% in the second bucket, and so forth. The table has five columns:

1. Cumulative % of Mailings - This is the cumulative percentage of ads mailed starting with the best prospects and advancing to the least qualified prospects as a result of the scoring within the predictive model.

2. % of Total Responders - This is the cumulative percentage of the total sales expected from ads sent to prospects in the buckets up to and including the one with the percentage being reviewed. For example, we expect to receive 50% of total sales (10,000 expected) from ads sent to the prospects in the two highest-priority buckets.

3. Expected sales - This is the total number of sales that can be expected from the cumulative number of ads mailed. If no model was used, the expected sales would always be 10% of the ads mailed. With the model, we see that expected sales are considerably better for the best prospects. The cumulative expected sales for a bucket are calculated by multiplying the total expected sales (10,000) by the cumulative percent of total responders figure.

4. Cumulative Gain Ratio - This is the ratio of the projected sales using the model to prioritize the prospects divided by the expected sales if a random mailing was done. For example, we expect to receive sales of 7,200 with 40,000 ads mailed using the model versus sales of 4,000 with 40,000 ads mailed randomly resulting in a cumulative gain of 1.80%.

5. Lift Ratio - This is the ratio of the change in expected sales for the modeled prospects in a bucket divided by the expected sales for the random prospects in the same bucket. For example, we have a change in expected sales of 700 when we go from expected sales of 6,500 to 7,200 so we divide by the expected sales of 10,000 for the random prospects resulting in a lift of .70%.

What we learn from the table is that by targeting the campaign at the best 10% of the prospects, we can expect sales of 3,000 which constitute 30% of the total expected sales. By targeting the best 50,000 prospects, we can expect 8,000 sales which constitute 80% of the total. The mailings done to the 10,000 prospects in the last (worst) bin are likely to yield only 200 sales for a return of only 2%.

Lift and gain tables and their graphics are ideal for executive presentations because they are visually appealing and readily interpretable by senior managers.

  Healy List Marketing is your source for purchase-based, multi-sourced consumer lists and databases

Healy List Marketing • 153 Andover Street • Suite 108A • Danvers, MA 01923-1450
Phone: 800-281-8956 • Fax: 978-336-0463 • info@healylistmarketing.com