One example, we are able to automate life insurance. We have to underwrite policies. When somebody applies for a policy, we take their blood, then assign them a risk: substandard, standard, preferred, etc. Depending on this, we price our products. Usually the process is that you take the blood, then it goes to a lab and we get the lab results back, then an underwriter takes a look at the lab results. This is usually done in a two week time frame to get a rating. We were able to build models to automate all of this, and now, it happens in real-time. Somebody can apply online and get issued a policy right away.

It has enabled our work force to be more efficient.

We previously needed a four-machine Spark cluster to be able to train an ML model using tens of millions of transactions, and hours of time during the modeling phase. Currently, same training can now be done on an old MacBook pro with 8GB RAM within few minutes.

Still on it. The idea is to save the cost of internal development but keeping enough flexibility to choose ML techniques and performance indicators.

We are using it for prototype projects. We have not deployed it.