Faraday Core — our AI engine
With historical examples, rich data, and key business outcomes in hand, Core continuously optimizes your customer lifecycle. Important opportunities are promoted, and likely dead-ends are deprioritized automatically.
Machine learning that works
- Faraday Core employs the most accurate, flexible ML techniques to optimize your outcomes. For a quick tour of the Random Decision Forest (RDF) algorithm, use the controls above.
- RDF starts with your known historical examples, both good and bad. For example, if we're optimizing lead conversion, we collect your leads that converted and your leads that didn't.
- Because these leads are synced to the Faraday Identity Graph, they include hundreds of rich attributes. To start, the algorithm picks one attribute at random.
- Here, household income was chosen. Now, the algorithm chooses a random "split" in the range of values. In this case, $53,439 was chosen.
- Next, the algorithm checks to see if this random selection has led to a better sorting of the converted and non-converted leads. It doesn't have to be perfect, just better than before.
- In this case, yes! The resulting subgroups of leads are better sorted than before. If that weren't the case, the algorithm would toss out this particular random selection and try again.
- Now we just continue to build our "tree" of random attributes and splits. Here, education level (college or below versus graduate degree) and football interest (yes versus no) were selected.
- In practice, the algorithm continues in this way until it reaches a decent stopping point. As you can see, each of the "leaf" circles on the tree are all mostly either good or bad. This is a successful tree!
- Now, when we receive a new lead, we can simply follow it through the tree to find the predicted outcome. For example, a lead with high income but no interest in football is unlikely to convert.
- On the other hand, a new lead with lower income that never went to grad school probably will!
The best models, automatically
Core always maintains a "best model" for each configured business outcome. When new data emerges, new candidate models are continuously built, cross-validated, and evaluated to see if a "new best" is available.