Automated Machine Learning Takes Root

Developing and deploying predictive models is no longer the preserve of data scientists as a growing number of third-party vendors have lowered AI adoption’s barriers-to-entry for the financial services industry.

The new automated-machine learning platforms “automate the end-to-end life cycle of developing and deploying predictive models – from data prep through feature engineering, model training, validation, and ModelOps,” noted the authors of The Forrester New Wave: Automation-Focused Machine Learning Solutions, Q2 2019 research report.

The technology-analysis firm divided the emerging market between vendors that cater to data scientists and business users, such as Datarobot,, dotData, and edgeverve, and those targeting business users, such as Aible, Bell Integrator, Big squid, DMway analytics, and squark.

Wall Street runs the gamut when it comes to adopting machine learning, according to Rob Hegarty, general manager Rob Hegarty, general manager of financial models and fintech at DataRobot.

“We’ve interacted with folks who are well down the path of AI and machine learning at truly AI-enabled firms,” he told IntelAlley.

“Also, some large firms are just coming into their own as well as some mid-tier firms that have not implemented data science or have data scientists on board.”

Platforms like the one from DataRobot that support supervised and unsupervised machine learning automate model development and selection, which permits data scientists to expand their focus beyond their initial core focus.

“Some firms know that advantage data science would have in other parts of their organization, but they do not want to put data science into those less differentiating parts,” said Hegarty. “The will not have someone who is producing predictive stock models create models that predict trade failures in the back office. The return is not just as high.”

“Across every line of business and industry, it’s less about how to build models, but how to get value from them,” agreed Jonathan Dahlberg, a customer-facing data scientist at DataRobot. “Machine learning and AI have moved from being something buzzwords and things you see on Jeopardy to something that can help operations, marketing, and risk teams as well as provide insight to what is happening.”

Ease-of-use is an essential feature among the various platforms that are intended for a spectrum of machine learning expertise. Users of DataRobot’s browser-based interface can drag-and-drop their data sets into the vendor’s internally or externally hosted platform via CSV and XLS files, generate JSON calls for Hadoop data sets, or use other options according to Hegarty.

Once the data sets are in the platform, it will clean the data before launching between 50 and 120 pre-made models from the vendor’s library of more than 400 predictive models to determine which model provides the best performance.

The platform automates a lot of the tasks, understanding how the models work and makes sure that you can get them into operation, added Dahlberg. That is the piece that is not necessarily ignored, but is often passed off by data scientists and can be an incremental gap.