This is how we see modeling.
For any push marketing client acquisition campaign, there is a message (copy, creative), a distribution channel (postal, email, text, or phone) and a target audience (the list). All three impact success. When the message and distribution channels are constant, using a model to determine the target audience will improve your results.
Think positive. High level, successful modeling is cloning. Tell us who your best clients are, and we’ll find others just like them, broken down into 10 groups of equal size (deciles), ranked from most likely to least likely to respond to your offer, within your geographical target area. If the model is built successfully, marketing to deciles 1-2 will significantly outperform the most targeted lists, and deciles 3-7 will show a lift over compiled data. The graph below compares modeled data to targeted data to compiled data, and identifies potential audience size according to the minimum acceptable conversion rate of your offer. If a conversion rate above 2.75% is desired (in graph below), only data from deciles 1 and 2 should be selected.
Now think negative. Models are not only used to build prospect lists, they can be used to remove unlikely responders. Saturation mailers should use a “least likely to respond” model to remove potential non-responders. The goal is to make your marketing effort more profitable by reducing the size of your postal, email, text or phone campaign, while maintaining the same number of conversions.
Here’s how we do it. We use a modeling technique called logistic regression. This is how it works:
1. Our national, compiled database contains 212 million consumers. Each consumer is defined by up to 700 different selects (gender, income, marital status, level of education, ethnicity, presence of children, age of children, pet owner, etc). We match your client list to our database, and append all of our selects to your list.
2. Each record on your client list (with our selects newly added) is then compared to our national compiled database, and a score is given to each record, according to how each element differs from the norm (the index). For example, if 80% of the records on your client list are female, but the population (in our database for the same geographic area) is 50% female, that would result in a high score over our baseline index. And if 1% of your clients are pet owners, but the general population (again, compared to our database for the same geographic area) is 5%, that would result in a low score under our baseline index.
3. Our program selects the 20 highest or lowest indexes from our baseline index to exactly profile clients on your list (ex: the 20 highest and lowest indexes determined that your best clients are age from 27-29, female, Jewish, divorced, 2 children, read “Dog Fancy” magazine, income $25,000-$29,999, smoker, etc). You tell us your desired geography (by zip code, city, state, or drive time to an exact address), and our software outputs a report, detailing by decile, from highest to lowest, the percent match to clients on your list, including available record quantities for the marketing channel you select.
Why is this different than other modeling applications? Off the shelf models (segmentation models) are built using a “birds of a feather, flock together” theory. It’s true, but not granular. Every household in the country is put into one of 40-80 “buckets”, depending on the known characteristics of that household. We at Exact Data know that you don’t always look or react like your neighbors, so we don’t predefine your future potential clients into set categories.