The other side of spectrum is machine learning research. That’s really only done at the largest technology companies – Google, Facebook, Microsoft, and so on – and maybe a few not-for-profit research organizations. These companies are focused on machine learning for developing new algorithms or improving algorithm architecture to deal with a particular type of problem.
We don’t do either of those at Crowe. And we’re not delivering one-off insights. Our focus is in the middle area that could be called “product machine learning.” We’re building machine learning into software products. Going back to the example of anomaly detection, we create an application programming interface, or API, that’s really the “brain.” It can live in a production environment or in the cloud, and that could be integrated into other software applications that now have the power to use this brain.
So there’s a much larger engineering focus, but it also scales much better because now we aren’t relying on the human time of data scientists to create those one-off insights, but rather building an entire product that can be used repeatedly.
The first is machine learning consulting. This is what’s primarily taught in academia. It’s about getting some data, creating a machine learning model, and delivering insights once or maybe a couple of times. It could involve getting financial data and making a prediction of which loans are going to default based off of historical loans that have defaulted. But the analysis is done once and the insight is delivered to the bank.
This is what we see with just about all of our traditional competitors to Crowe in this space. You might hear other firms say they have hundreds or thousands of data scientists. But what I just described is what they’re all actually doing.