When DataRobot eats more data then can be provided

The python API was super important to fully automate the model creation.

The capacity to use the graphical interface but then also use the python API the grab the ROC curve and really tweak the results to our needs. We had a business bias where false positives had much less of an impact than false negatives. By using the ROC curve data we could adapt the thresholds to our needs.

The capacity to very quickly build models and even use the learning done on a similar model to speed up the next model. A lot of our binary classification models were using similar data points. So we would build the first few models with the normal experiment configuration. This would pinpoint the best algorithms and we would then limit the next models to these best candidates. This greatly sped up our process.

The help of your customer-facing data scientists was incredible. Each time they exceeded our expectations. And it was really tailored to our needs. Really this was amazing.

Sadly the project was put on pause. The problem was really not with the data robots but with our inability to read the data from production in an efficient manner.