Item: DataRobot
Rating: 9
Author: Verified User

Overall Satisfaction with DataRobot

Use Cases and Deployment Scope

We build predictive models with the core supervised learning product. These include attribution models, churn/retention models, segmentation models, and others. Basically, anything that can be accomplished by taking a supervised, labeled set of training data and turning it into a predictive model, we use DataRobot. We have also dabbled with unsupervised learning and time series modeling but have not purchased those packages.

Pros and Cons

Pros

Iterative model development
Fast training of a very large number of models
Easy deployment to their cloud solution, or export as an approximate model
Visualization and explanation of important model components

Cons

We should be able to download data sets from our own projects--after all, we uploaded them originally (and they were not stored locally; they were created specifically for a DataRobot project).
The sales team is very aggressive at pushing features that we would never use, such as data hygiene (clunky integration of Paxata), ML Ops (just don't need it), and AI services (we're a mature company; we don't need help coming up with use cases).
Pricing changes every year--not just the amount but what you actually get, so we need to nitpick the contract each year because DataRobot has inevitably eliminated something we need.

Return on Investment

Major increase in productivity because we're no longer building models "by hand"
Peace of mind once models are built that they won't break
Ability to rapidly test out ideas that may or may not have machine learning solutions

Alternatives Considered

We consistently return to DataRobot for its ease of use and ability to get the job done without major hurdles. Thus far, we just haven't found that in other products.
H2O.ai (Driverless AI): several test models did not complete, and H2O.ai team could not explain why.
Sagemaker: Much easier if you're already experienced with the AWS ecosystem; otherwise, good luck, you'll need it.
Kortical: Not quite ready for prime time, though we liked the direction they were going
KNIME: great for data analytics, not so deep on the modeling side

Key Insights

Do you think DataRobot delivers good value for the price?

Yes

Are you happy with DataRobot's feature set?

Yes

Did DataRobot live up to sales and marketing promises?

Yes

Did implementation of DataRobot go as expected?

Yes

Would you buy DataRobot again?

Yes

Other Software Used

Python IDLE, Apache Spark, Microsoft Excel

Likelihood to Recommend

It's appropriate for speeding up the work of your experienced data scientists. If they spend more than 15% of their time building and tweaking models, DataRobot will cut that down significantly. Caveat emptor: while the DataRobot marketing materials promise to turn any analyst into a data scientist, this is far from the truth. If your potential users do not already understand how machine learning models work, and have not built some models on their own, then they will make mistakes that DataRobot will not correct because it assumes you know what you're doing. Interpreting the results and iterating on models is easy for a trained data scientist but would be baffling for a typical financial analyst.

Comments

Please log in to join the conversation

DataRobot will free your data scientists from the boring part of their job and allow them to focus on the human part