Matt Gershoff, CEO
Conductrics recasts digital marketing optimization as a reinforcement learning problem – a type of machine learning for adaptive predictive targeting. However, regardless of how accurate the machine learning was, many clients were uncertain about ceding control, even of low risk aspects of the user experience, to automated systems if they couldn’t understand how the machine learning was making decisions. This led the company to reevaluate what is considered a good machine learning solution, especially for customer facing applications, and to redesign its machine learning engine to solve for accuracy and be human interpretable.
Why does human understanding matter?
- Trust: People often need to feel like they understand something before they can trust it.
- Insights: Not only do people need to trust these systems, but they also want to be able to glean insights from them.
- Accountability: Will this decision from Conductrics’ machine learning be consistent with its stated data policies?
- Explanation: Can one communicate and explain the prediction, or decisions, to customers?
The solution is based on a mapping of regression-based function approximation into sparse decision trees. Sparse decision trees have many appealing properties for the marketing use cases:
- Human Readable: Easy to tell what experience or output the system will take.
- Discrete Rule Set: The rules cover all possible use cases, which enables review/legal to predetermine all outcomes for all inputs.
- Loggable: Since the decision policy is represented as rules, each decision can be logged with the exact policy/rule that was used.
Conductrics has had great success with clients such as the Finnish national lottery, Veikkaus. “With Conductrics, we delivered a richer, more personalized gaming experience, resulting in significant growth in player participation and 25 million euros in additional revenues” (Eetu Paloheimo, Head of Ecommerce at Veikkaus Oy).