In Isaac Asimov’s Foundation series of science fiction novels, “psychohistory” employs massive amounts of historical data to predict the future behavior of large groups of people.
This, essentially, is predictive analytics, which we tackle in this installment of our MarTech Landscape series.
“Think Of A Credit Screen”
Instead of predictions about Asimovian empires, though, our generation of predictive analytics is content with more modest goals, like forecasting whether it’s likely you’ll become a customer.
“Think of a credit screen,” predictive scoring service Versium’s CEO Chris Matty suggested to me.
Analysis of several data points in your history — such as your salary and your payment history — is used to answer one question: “Will this person pay back the loan?” That kind of prediction has been around, in one form or another, since the invention of loans.
And every seller has played a hunch about whether this person or that one is a likely buyer. But marketers and others now have the resources to make much more accurate guesses. Computer-based predictions, utilizing mounds of big data, are sprouting up in many industries besides financial services, including health care, fraud detection and political campaigns.
In marketing, predictive analytics is becoming so commonplace that — one can predict — there may come a time in the not-too-distant future when its use is assumed in most systems. Already, it would almost be easier to list the marketing vendors who do not say they use some form of predictive technology.
There are predictive lead scorers like Versium, Everstring, Mintigo and Infer. Boomtrain offers predictive targeting for email, Tapjoy does predictive advertising and Lytics provides predictive content. Some level of prediction is built into a variety of marketing platforms, including Act-On and Salesforce.
In fact, any area where it’s valuable to assess who will most likely respond, or what they will most likely respond to, is fair game for predictive analysis.
Why Now?
The two big drivers of this boom, of course, are big data and what you might call intuitive computer processing.
It’s difficult to overestimate not only how much data has been acquired on every individual in a modern society, but how fast it is increasing.
Of course, there are your credit and checking transactions, your online browsing and downloading history, your visits to physical retailers (through location tracking of your phone), your health regimen (through your online fitness tracker) and the demographics of where you live, who you are and what you do.
And there will soon be histories of your car, your appliances, your visits to webpage-transmitting vending machines or movie posters and much more.
The fact that much of this data is either anonymous or anonymized doesn’t mean it can’t be used to pin you down. “You” may be a tag instead of a name, but you are still targetable, and your tagged self can often be matched with your real identity. In any case, if data tracking left streams of color behind, the modern world would be awash in larger and larger rivers of rainbows.
In other words, we are generating more data than even Asimov might have imagined.
The second big driver is the branch of artificial intelligence called machine learning.
In essence, it’s a computer system creating models based on patterns recognized in big data. Unlike a human loan officer, a computer can obviously track, analyze, abstract and compare thousands — even millions — of data points about what you’ve done and who you are.
Matching What You Want
Email campaigns, for instance, operate differently when they are segmented and rule-based, compared to when predictive analytics is used to forecast responses.
In the former, all men aged 18 to 34 might get the same email. In the latter, machine learning is used to find patterns among the many data points of people who have responded well to this kind of email, and then individual email recipients are scored on the basis of how well they fit the pattern.
Like Moneyball, where data analysis was used to find the hidden traits that made a good baseball player and then to find other lookalikes, a common way to create a predictive model is matching to what you want.
Take how San Francisco-based Insightly uses predictive tools. To determine which people using trials of its relationship management and project management software are likely to become paying customers, it uses commercial predictive analytics tools like Preact to compare their behavior to those who have already become paying customers.
CEO Anthony Smith told me that, while the tool isn’t 100 percent accurate, it is still about 85 percent.
Avention focuses on finding the best company targets for B2B sellers, with individuals included to the extent that they’re part of the company picture.
CMO Vicki Godfrey said her company uses data “to predict which companies would be similar to the ones you’re successful with.” This creates an Ideal Profile Score, which is based on how close a potential target matches the Ideal Profile.
She made a distinction between behavioral matching and profile matching. Behavioral matching is a form of prediction, such as a sports site that predicts you’ll like this page about golf because you’ve previously visited pages about golf.
On the other hand, Godfrey said, a company profile used for prediction is a more holistic view that includes the firm’s reported revenue, its recent office move, its hiring of new people and other company-specific data, matched with external data about companies in its industry, long-term trends and so on.
“Better Than Human Intuition”
Versium’s Matty noted that you can also “create a negative performing class,” where you analyze the data attributes of people who don’t buy, in order to create a profile of the kind of person to avoid.
Matty is one of those people who subscribes to the Asimovian ideal. “The more data you get, the more accurate it will be,” he contended.
But don’t fret if this sounds like everything is already preordained, if only there were enough information. Even a data-believer like Matty acknowledges that “you can’t model when the ball fumbles.” The world is still full of chance events, like weather, random encounters and spontaneous, uncharacteristic, inexplicable human decisions.
The key here is to understand that many predictive systems derive their value from giving you, the marketer, an edge in assessing large numbers of possible customers. Any one user might be subject to chance events, but, even if you have a small likelihood that 5,000 people in your customer list of hundreds of thousands will be more likely to buy, you’re ahead.
“It’s never 100 percent,” Matty said, “but it’s better than a coin flip, and better than human intuition.”
Komentarze