Amex and Acxiom create a prediction engine based on purchase data
This is the latest effort to utilize extensive, identified customer records to predict future purchase behavior.
American Express (Amex) tracks about a trillion dollars in annual spending. Today, it is announcing a new effort with data service Acxiom — called Predictive Intent Segments — to use that spending data to predict which individuals are likely to buy certain products.
Let’s say you’re a bike manufacturer that wants “to find households in America most likely to buy a bike,” Amex Advance VP and GM Marc Ginsberg suggested to me.
First, Amex will find personally identified individuals with American Express cards who have actually bought a bike recently.
Then, using the personal identification, it appends the record with additional data that Acxiom has in its database about that particular buyer. Acxiom might have, for instance, recent purchases by the individual using other forms of payment at commerce sites, physical athletic stores and baseball stadiums, and it matches that additional data through the person’s name, address and other personal info.
Although these are personally identified matchups — meaning the name, street address and so on is in the record — Ginsberg said it’s done in a “black box” by a third-party service, so neither Amex nor Acxiom can see the identities.
Once the bike purchasers are further detailed with additional attributes, a predictive model is created via machine learning, which finds patterns among the bike buyers. Ginsberg pointed to age, spending patterns, an active lifestyle, gender and marital status as being particularly predictive variables, such as young parents with discretionary income who live in the suburbs.
The model creates a score that predicts an individual’s propensity to buy a bike, based on anonymized attributes.
Prediction from purchases
Then the model is used to score other personally identifiable individuals in Acxiom’s database, or in a brand’s collection of its customers. The bike manufacturer, for instance, might want to know which identified customers of a sneaker maker are most likely to buy a bike.
Acxiom has matched those personally identified profiles to their cookie and mobile device ID, so an ad for a new bike can be targeted at them, or they might receive an email from the cooperating sneaker maker that includes info on new bike sales. This is essentially the kind of targeting that Acxiom supports day in and day out, except now it is guided by a predictive score based on actual purchases with an Amex card.
This Amex-Acxiom joint effort is only the latest to employ extensive PII data for generating a propensity score to predict if someone is in-market for a given product.
Earlier this week, for instance, Infutor launched an Auto In-Market propensity marketplace. It utilizes the Illinois-based company’s massive repository of several hundred million identified profiles of US consumers to similarly create a score based on previous buyers’ attributes, in order to determine which potential new customers would be most receptive to marketing about a new or used car or related products. Infutor plans on creating other propensity marketplaces for other product categories.
Rick Erwin, president and GM of Acxiom’s Audience Solutions, told me that “a lot of [targeted] segments are list-selected,” such as Soccer Moms. In other words, those targeted segments are selected by geography, or demographics, or inferences made from behavior, like visits to sports sites. Here, he said, the target is scored based on actual transactions in the chosen product category.
As the joint effort is new, there are not yet enough results to determine how well the scoring works. But Ginsberg pointed out that this kind of prediction — looking for signals from past behavior to indicate future propensity — has been used successfully by Amex to predict fraud.