Malcolm Houtz
SVP, Data Science
When to Use a Binary vs. Multi-behavioral Model for Predicting Marketing Performance
Choosing which type of model to use for your marketing campaign is a lot like choosing which superhero is better equipped to save the day: Captain America or Spiderman. The answer? It depends. In scenarios requiring feats of strength, either would get the job done. But if the dilemma demanded speed, agility, spider-like senses, and flight-like abilities — you know who our guy would be.
Binary and multi-behavioral models, in a way, are similar. Both have tremendous predictive power as a proxy for marketing performance — but depending on the scenario, the multi-behavioral solution may be a better candidate. Let’s unpack this further to understand their key differences.
What is a binary model?
Binary models predict the probability of a single action (e.g. how likely the consumer is to respond to an offer). The answer is always on a scale of No to Yes, 0 to 1. Binary models are proven performers and the gold-standard for predicting response, product propensity and price sensitivity in marketing.
What is a multi-behavioral model?
Where binary models focus on the probability of a single action, multi-behavioral models simultaneously predict the probability of many actions — and incorporate profit values for each possible outcome to predict the estimated profit value of an audience. While not commonly used, multi-behavioral models can be incredibly powerful for subscription-based campaigns that only turn a profit if the prospect pays for multiple shipments.
Head to Head Comparison:Binary Model vs. Multi-Behavioral Model |
Binary Model (Regression) |
Multi-Behavioral Model (Generalized Logistic Regression) |
|
|---|---|---|
What does it predict? |
The probability of a consumer performing a single action. (e.g. Will they pay?) |
The probabilities of the consumer performing each possible action. (e.g. Will they respond? Will they respond and pay? Will they return? Will they result in bad debt?) |
What is its superpower? |
Predicting actions (e.g. Payment vs. Non-payment) |
Predicting profit value (e.g. Response + Payment + 2 Shipments + 1 Return = $11.50) |
What offer type does it work well for? |
Single order or subscription offers with low bad-debt/return costs |
Free trial, introductory, or subscription commerce offers with high bad-debt/return costs |
How well does it find good performers? |
Strong |
Excellent |
What data is needed for development? |
Promotion File + Response File |
Promotion File, Response File (with well-defined mutually exclusive behaviors) + Cost for Each Action |
What’s the development time? |
Average |
Slower |
When considering the two options, talk to your data scientist about the best approach. The multi-behavioral model is a workhorse, but also labor intensive. You’ll need to provide more data and determine the cost for each action, which isn’t always an easy calculation. Depending on the nature of the offer and range of associated costs, a binary model may be just as predictive.
No matter which superpower you need — the ability to predict an action or profit value — make sure you leverage a large source of comprehensive transactional data to maximize the performance of your model and, ultimately, your campaign.
For more information, check out our FAQ guide to multi-behavioral models — or throw a lifeline to your favorite Alliant account executive or statistician. Because at Alliant, our data scientists are superheroes, too.
Submit a Comment