Talking Inventory Drones With Pensa's Richard Schwartz
How computer vision can fly to the rescue for out-of-stocks. The founder and CEO of Pensa Systems discusses how "flying whiffle balls" can defeat "electromechanical beasts" in the race to eliminate out-of-stocks, and how winemaking is like tech.
![Pensa Systems, Schwartz Pensa Systems, Schwartz](https://eu-images.contentstack.com/v3/assets/blt58a1f8f560a1ab0e/blt8450b0d034f59cab/669eedaf1e87a90f23b05472/PensaSchwartz-web_1582648412.png?width=1280&auto=webp&quality=95&format=jpg&disable=upscale)
Richard Schwartz is the president and CEO of Pensa Systems, the Austin, Texas-based technology company that has introduced a drone-based inventory robot for retailers and brands.
Jon Springer: Welcome to the Breakroom, Richard. Pensa has a unique solution to attacking the on-shelf availability issue. But how did it all begin? Were you searching for a solution to out-of-stocks, or for a useful application for computer vision technologies?
Richard Schwartz: Good question. It began with a bit of both. Pensa had been looking at the problems in inventory and supply chain. There were lots of areas where robots and automation had made a difference already but other areas that were still very people- and eyeball-intensive. In-store inventory was one of those areas.
It is really hard for people to stare at shelves and figure out what is missing and out of stock. At the same time, it is a good application for computers and computer vision, provided some AI can be put behind the application to sense what is right and wrong about the shelves. So, there it is: We matched a painful problem with a new type of emerging technology behind solving it. Somehow things are often like that—a mix of problem and solution that beg for seeing both and putting them together.
Can you put the challenge of on-shelf availability into perspective? Why is it important for retailers and brands to get a better handle on this now?
Believe it or not, an average of between 1 in 8 and 1 in 10 products are out of stock on the retail shelf at any given time. And there is no system of record to automate scanning the shelves to know what is missing. It is the giant black hole of retail—a $1 trillion problem for the industry. That problem translates to serious loss of revenue and profit margin for the retailer and the brand manufacturers—in an industry already challenged to reinvent itself to combat pure online sales and running on thin margins.
Reducing stockouts is “found money” for the retailer—if they can do it with automation instead of having more employees, taking more time away from customers to stare at the shelves. In fact, the problem is only getting more acute as online sales and brick-and-mortar sales blend, with shop online and pickup in the store or Instacart delivery. The stakes go up in this new world. If a customer buys online and goes to the store to pick it up, heaven help the store if the product is actually out of stock and unavailable. Studies show the retailer may lose the customer for life. So, now more than ever, the problem must be solved.
Photograph courtesy of Pensa Systems
We’re now seeing retailers use floor-based rolling robots, ceiling- and shelf-mounted cameras and other tech solutions as a means to improve shelf availability. Why in your opinion is a drone-based solution better?
Retailers and brands have experimented with everything: ceiling cameras and even fixed cameras mounted in the store, in coolers and even on shopping carts. These approaches just don’t work. It is hard to get the right view of the products on the shelf. Computer vision doesn’t do well without a relatively clear view of the shelf and the products. Accuracy is quite low in these cases, and low accuracy means the results aren’t trustworthy. What’s the point of automation if you can’t believe the answer?
Ground-based robots are sumo wrestler-sized electromechanical beasts—large bulky, expensive, and even they can’t get the right view of the shelf. Many of the other approaches also assume impossible-to-get data to base decisions on, like up-to-date image libraries of the products, perfect maps of the store, or of how the shelves should be organized.
Drones, at least the autonomous ones in our case, are small, lightweight and agile. They are nearly throwaway in cost and, frankly, not doing most of the work. They are self-flying cameras that feed computers elsewhere. So the system can be scalable and stand a chance of sweeping through an industry without a heavy lift. The drone itself, as it moves, is able to see the shelves from many angles and perspectives, essentially substituting for up to 100 cameras fixed on every item on every shelf. With our AI approach, this gives our system 100 times at bat to recognize each product on the shelf. That translates to more accurate and trustworthy results.
How do you address concerns that might arise around tiny robots flying through the aisles of a retail store? What’s the shopper reception been like in places you’ve tested?
Our system can sense when an aisle is empty, descend to scan during the downtime, and just fade up and away if a shopper or worker comes into the area. It operates around people by being nowhere near them. It is too small and agile to hog the aisles like ground-based robots. And since it knows to move away from people, it doesn’t have pretend to be friendly with googly eyes.
Frankly, the shopper reaction has been almost disappointing. People don’t pay much attention to it. It doesn’t really look like a drone; it has been described as “flying whiffle balls”—small, light, pretty quiet and not much fun. Roughly half the people don’t notice it at all, even at the other end of an aisle. The ones who do see it are positive, curious and kind of expect to see these types of things coming into retail. And it has an important for-the-customer goal: not to run out of their favorite mustard or other products.
In addition to numerous technology gigs over your career, you also make your own wine. Anything you learned making wine that can apply to creating technology solutions?
Good winemaking is a mixture of art and science. You blend the sensory side with careful measurement, planning and controls. Tech is the same way. You have to think about the user experience, the problem and what kinds of qualities of a solution are practical and will solve the problem without introducing other problems. A good product, like a good wine, is refined, well-balanced, easy to digest and pleasing to the senses.
Lightning Round
Coolest place to shop for food in Austin, Texas?
Food trucks, hands down.
What’s one thing on your grocery list not likely to be on many others’?
Large quantities of yeast and citric acid for winemaking. And empty bottles, lots of empty bottles.
Who’s your all-time favorite computer scientist?
Niklaus Wirth. He said that “innovation is the process of consolidation”—that the best inventions come from learning and adapting from what has been done before.