Thought Leaders in Artificial Intelligence: Daisy Intelligence CEO Gary Saarenvirta (Part 2)

Sramana Mitra: It’s consumer-facing merchandising. You’re not driving the ordering of inventory?

Gary Saarenvirta: We do decide that also. We set the prices and we also do inventory forecasting and allocation. Those are the three core inputs – having the right product, right prices, and having the right inventory.

Sramana Mitra: What kind of quantitative impact do your customers get by bringing you into their workflow?

Gary Saarenvirta: We double the company profit. In grocery retail, we more than double the company profit. We generate 5% to 10% total sales increase for companies that are $200 million in revenue. For a client that’s doing $30 billion in revenue, we grew their revenue by more than $1.5 million a year. In a 1% net-margin industry, we double the total company net income. The financial impact is massive.

Sramana Mitra: Is grocery retail your primary segment within retail? 

Gary Saarenvirta: We work with high-frequency retail – grocery, drugstore, hypermarket, and home and garden. It can be applied to every retail segment but as a small company, we focus our efforts on high-frequency retail. The mid-market is struggling, so we thought that was a good place to start.

Sramana Mitra: Let’s do a use case from insurance next.

Gary Saarenvirta: It’s fraud detection like group health benefits. When you file a claim, we identify what are the claims that shouldn’t be paid that are fraudulent. We look at the people who are committing fraud and find organized activity.

The system looks at claims and says, “Don’t pay this claim.” You can auto-adjudicate. Our goal there is to autonomously adjudicate with no human in the loop. The users of our systems are the special investigators who investigate fraud alerts. We see the human being doing the most sophisticated investigations and the AI doing the autonomous decision-making and auto-adjudicating the fraud.

The goal is to do this in a prepayment mode where before you pay the claims out, the system would have determined whether you should pay it or not.

Sramana Mitra: I’m going to go one level down. Talk a bit more about the technology. Algorithmically, what kind of AI are you using to accomplish what you’re accomplishing. Explain in whatever way is comfortable for you.

Gary Saarenvirta: My background is Aerospace Engineering. What NASA has been doing since the 1950’s, that’s the branch of AI which is optimal control or reinforcement learning.

Most AI is really statistical analysis. If you analyze only historical data, that’s super-biased learning. That means you can never do what you haven’t seen before. You have to have events and model those events. There’s no decision-making process wrapped up in predictive analytics. You have to have a decision framework on top of that.

Because you can learn only from data, to learn something new, you’d have to do experiments in the marketplace and collect that data. The learning is mathematical. Even deep learning is a sophisticated form of regression.

What we do is what’s called reinforcement learning. It’s like a self-driving car. You don’t drive with just a model. You start with no historical data. Let’s say you have a car on a race track. The autonomous system drives the car and decides the optimal sequence of steering, brake, and gas pedal that gets me the fastest lap time.

It’s not greedy. There’s delayed reward. You can slow down on one corner to get the fastest lap. You use historical data to inform your simulation. You do random trial and error to find out the optimal sequence of steer, brake, and gas.

Since it’s a computer simulation, you can do 100 million hours of driving and one hour of computing time. You can learn faster than the pace of time. You can learn new ways to drive that human beings haven’t learned. You can drive more than all of humanity ever has. This is what game-playing bots like Go do.

This segment is part 2 in the series : Thought Leaders in Artificial Intelligence: Daisy Intelligence CEO Gary Saarenvirta
1 2 3