Florian Quarre: The final use case is automated decision making. We’ve gone through defining the data meaning and dictionaries. We’ve gone through digitizing data.
The last stage is executing on the data and start making decisions. One of the use cases that we help enable is making decisions on whether a claim should be paid or rejected.
The algorithm detects a pattern that brings together treatment and condition in ways that say, “This is not the right context.” It’s a high-level of churn that leads to a lot of administrative waste. Streamlining the exchange and the processing those claims leads to not only qualitative treatment but also a reduction on administrative burden that could be allocated in different ways.
Those are three different types of use cases that we typically partner with our clients.
Sramana Mitra: What about quantifying the impact of this kind of use cases? Have you done ROI analysis?
Florian Quarre: There are a few metrics that come to mind. One of the first is error rate reduction. We typically see between 30% and 65% reduction on errors that are generated on processing that is usually manual. Throughput is another aspect.
Once you’re able to delegate or augment a workforce with an AI system, we’ve seen throughput that is 10x to 20x. That is a tremendous acceleration to not only solidify your process but also allow your business to make decisions faster and sooner.
In terms of monetary benefits, we see between a quarter of a million and $10 million annualized over three years. They were removing roadblocks and creating added value along the way that changed the way our plans are operating in those specific fields.
Sramana Mitra: What do you see as open problems in the domain that you are close to? If you were starting a company today, what kind of a company would you start?
Florian Quarre: We are going through a rewrite of our company. The company was established about three of four years ago. We started by doing a lot of custom application development within the field of AI and helping our clients adopt AI techniques.
One of the main issues that we’re now trying to solve is the ability to transport what you build with AI techniques in development into production. There is a wide gap in technical wiring that allows you to build models that are validated by limited datasets and transpose it to a production-ready, scaled-up environment that can be bombarded by millions and millions of transactions with a drastically reduced SLA.
What we’re working on is creating the bridge between developers and the production environment so that there’s continuity. That’s one of the main things that we address as one of the white spaces.
Sramana Mitra: Thank you for your time.