Daily View: Focal Systems on “Pragmatic AI” in Retail

If you are counting on human-free stores and driverless cars, you’ll want to skip this episode because it’s a downer. To be sure, if you’re big on the romantic potential of check-out lines, tune in! Francois Chaubard, CEO and Co-founder of Focal Systems discusses the investment from Zebra Technologies, profiled in the November 20 Daily Monitor, and explains why human checkout people, and drivers, aren’t going anywhere any time soon. He’ll also discuss where “pragmatic AI” is making a difference in-store, and on-shelves, in retail, right now.

David Zweifler, Editor in Chief, Gordon: [00:00:00] Welcome to another edition of the Gordon Podcast Daily View, an in-depth look at the people and companies that are making news, which are profiled in the Gordon Daily Retail Monitor.

[00:00:18] The monitor covers every announcement, every deal, and every narrative for every retail company across the globe every day. And it it’s completely free. Click the link in the descriptor if you’d like to subscribe.

[00:00:34] If you’ve been following the monitor, you’ll remember our coverage of Focal Systems, which received an investment from Zebra Ventures to scale globally on November 20th. Today to offer some perspective on that deal and explain where the technology is going, I’m speaking to Francois Chaubard, CEO of Focal Systems.

[00:00:56] Thanks so much for joining us today. Maybe we could start by having you explain a little bit about what Focal Systems does.

Empty Shelves, Angry Customers

[00:01:07] Francois Chaubard, CEO and Co-founder of Focal Systems: [00:01:07] We deploy these small little cameras, basically every eight feet throughout the store. So like at a big Walmart, you’d probably need about maybe 300 or so cameras to cover the whole store, but you will have to, you can cover the high-movers or, you know, kitchenware doesn’t really need to be covered.

There are two AI components, One is inferring from pixels, what’s in stock and out of stock… The other is predicting on how to run a better store.

[00:01:25] They take an image every hour. They compute in stock and out of stock. They do product recognition. And then from that, they can infer what’s, in fact, out of stock, (and that feeds into) a system that we call focal operating system, which is essentially the best store manager of all time. It has a complete 360 view of what’s going on in your store. It knows how quickly your, your staff is fixing these issues. It knows where the labor, the inventory is in the backroom or in top stock. So I can tell you know exactly what the best use of one additional hour of labor is and says, “Don’t spend this hour going and restocking mint jelly, restock Tide.” So you rank/sort and do the most important work first, which today is not, it’s not happening. There’s, there’s nothing telling them… rank sorting or anything like that today.

[00:02:19] David Zweifler, Editor in Chief, Gordon: [00:02:19] They’re just brute force. There, they’re attacking all the shelves all at once with a lot of staff to make sure that it’s all without prioritization. That prioritization is the AI component of Focal Systems. Is that correct?

[00:02:31] Francois Chaubard, CEO and Co-founder of Focal Systems: [00:02:31] Yes. Yeah. So there are two kinds of AI components, I guess you’d say. One is inferring from pixels, what’s in stock and out of stock and from, you know, all that, analytics — that’s computer vision being run.

[00:02:45] The other AI component is predicting on how to run a better store. So it’s predicting that, “Hey, you know, you should increase spacing on Tide because you go out of stock by 4:00 PM every single day. So it finds these patterns and helps you get in front of them to run a better store.

Avoiding Low Tide

[00:03:03] If you are constantly just solving, you know, jumping from one out-of-stock to the next, to the next and the next, you’re not actually solving the root problem for an out-of-stock. Ideally, you’d have no out-of-stocks. And the way you would do that is if you measured demand perfectly.

[00:03:16] But you can’t do that always, but you can minimize it. You can do (a) better job. And if your planogram, you know, if your demand is wrong and your planogram is wrong, and your planogram only holds 10, but you go through 10 by it, by noon, then you’re going to have an out-of-stock. And so what you should do is increase your, your shelf space.

[00:03:37]And, so, same thing with ordering. If you only ordered 10, then it’s even worse. It’s not only that you only have 10 on the shelf, but you have none in backstock. So all these things like it’s impossible for someone to be looking at these things per store on 100,000 skews, and then trying to optimize (them).

The staffing schedules that they have today are that way because that’s the way it was 10 years ago, 20 years ago, that that’s really the only reason why they have those staffing methods and scheduling.

[00:03:56] But it is possible for this AI system to be looking at those patterns and predict the best use of your store labor, the best planogram you can use, the best ordering.

[00:04:07] One of the things where we, we just deployed in a couple of our stores is this optimal staffing. Does it make sense to have, you know, 10 people between 6:00 AM and 10:00 AM, or should you have five people at that time and five people from three to seven? And if that’s what all your out of stocks happen, then that’s when you should have people there.

[00:04:28] And very often these, these, the staffing schedules that they have today are that way because that’s the way it was 10 years ago, 20 years ago, that that’s really the only reason why they have those staffing methods and scheduling.

A Marriage of AI and Humans

[00:06:54] Francois Chaubard, CEO and Co-founder of Focal Systems: [00:06:54] Yeah, so it’s like a 10 X ROI. We actually try to break even on the labor side of the business. This is going to save labor in three different chunks.

[00:07:08] One is they’re spending four to five hours per day per store, scanning out, just walking around with a scan gun and scanning all the holes. We obviously do that complete automatically. And so just on that, you’re already break-even.

[00:07:23] Then the next two places where you save a lot of labor is one on what’s called like one-touch replenishment versus three-touch replenishment. A one-touch replenishment is, I take it off the truck, I take X units off the truck and slot directly into the shelf, done. And now I’ve restocked.

[00:07:42] A three touch replenishment, which happens for like 20% or 30% of the SKUs is, I take it off the truck. I think X units. Half fit in the shelf. Half I have to bring back into the back room. Sometime later I had to go into the backroom plot, pull that thing out, and then fill it again. And the only reason why I do that is my planogram is horrible, horrible. And so most problems stem from a bad planogram and that’s what we’re finding.

(Stock-checkers) are spending four to five hours per day per store, scanning out, just walking around with a scan gun and scanning all the holes. We obviously do that complete automatically. And so just on that, you’re already break-even.

[00:08:08] We’re finding these patterns, you know, between two stories that are right next to each other should have different planograms, but that’s not the way that they’re set up today.

[00:08:18]So that’s the second bucket. And then the third bucket is kind of this a Hawthorne effect, if you will, where like, you know, today there’s no measurement.

Taking Stock of Stocktakers

[00:08:27] You have, like, you know, David and Francois are the stockers, in the store. And our job is to get through as many outs as possible. We call it pick velocity. And so if, if Francois’s is at 20 picks per hour, and David is at 40 picks per hour, you know, I’m going to get, you know, flack from my boss… “Why is David, Oh, so much faster than you, why is he getting this done?”

[00:08:51] And right now, there’s no measurement on how effective a stocker is being like, if you ask right now, you go to the (inaudible) and you say, what’s your average pick philosophy in this store? He wouldn’t be able to tell you. I find that actually very surprising.

[00:09:06] Those are the three buckets on the, on the labor side, but we don’t actually, like, I don’t want to prove an ROI on the labor side. I actually don’t think is a good idea.

Right now, there’s no measurement on how effective a stocker is being like, if you ask right now, you go to the (inaudible) and you say, what’s your average pick philosophy in this store? He wouldn’t be able to tell you.

[00:09:15] In the end. It is the retailers’ decision on what they want to do. But I think there are so understaffed today and there’s so much work to be done. If you want to cut a bunch of labor, which they’re doing, you know you can do so, but you’re going to, it’s going to impact the customer experience.

[00:09:30] It’s going to impact sales. And already has. I think everyone is seeing that. I think what you need to do is redeploy that labor just more intelligently or spend labor more intelligently. And the way we think about it is like per hour, is this hour increasing sales or decreasing sales or increasing operating margins or decreasing operating margin?

Using Staff In The Right Places

[00:09:49] If you had way too much labor, then you could be reducing operating profit. That curve kind of does like, you know, an inverted parabola, right? They kind of like, if you’re at zero labor, you’re making zero sales for sure.

[00:10:04] David Zweifler, Editor in Chief, Gordon: [00:10:04] The Laffer curve

[00:10:05] Francois Chaubard, CEO and Co-founder of Focal Systems: [00:10:05] Infinite labor, you’re, you’re losing a lot of money, right?

[00:10:08] Exactly. Like the Laffer curve. And so there’s some optimal point there. And the problem is they don’t know. They don’t know the amount of work that needs to get done and the breakeven of that work. So, you know, you kind of look at like this marginal cost, marginal benefit, as like, you know, if, if I’m out of stock on Tide, you know, you’re losing what, like a a hundred dollars in operating profit per store, per hour when you’re out of stock on Tide in some of these like big Walmarts, right?

[00:10:41] Is it worth, is it worth $10 an hour to go and restock that? I think so. If you’re out of stock on mint jelly and you know, your, it costs you $10 to restock it or something like that. Like maybe it’s not a breakeven. Maybe you should wait for like the most efficient labor, which is night crew and that can get 60 or 70 replens per hour.

They don’t know the amount of work that needs to get done and the breakeven of that work.

[00:11:02] That maybe that’s, that’s fine. Right? But they don’t. Is a measured way to spend labor is not being done today.

[00:11:11]David Zweifler, Editor in Chief, Gordon: [00:11:11] At least you’re approaching it, you know, more intelligently where, you know, what’s on the shelf and, and what you can expect to have on the shelf at any given moment.

[00:11:19]So you guys are kind of in the middle of, of that process that you defined before where you’ve got deliveries coming in the back and then they get on shelves and you just took an investment from Zebra Technologies, which is at the checkout.

[00:11:34]It’s still early days and you can’t really get into the specifics of what that combined technology is going to look like. But it seems like once you’ve sewn up the cash register, the shelves, and what’s going on in the backroom, you could have, a largely human-less store, right?

Stopping Amazon Go

[00:11:54] Francois Chaubard, CEO and Co-founder of Focal Systems: [00:11:54] No, I don’t believe that. I mean, I don’t think even Amazon Go is human-less. I mean, it’s, there are people stocking shelves. You’re not going to replace that for a long time. And robotics is just not there. So, you need people to do work.

[00:12:07]You know, Amazon Go even despite the price point. And if you see my… I have a pretty large case study on Amazon Go, and the cost and implications thereof, you’re still going have some type of auditors and people like that, and especially in big format stores.

[00:12:21]I don’t think that the human-less grocery store will scale beyond a hundred, maybe 500 square feet. It’s just not scalable. I think like the big old box model and things like that will work. There’s a lot of reasons why I believe this, but there’s just not a chance that I believe in this future with human-less grocery stores.

[00:12:40] It would be very futuristic, you know but unless you’re putting RFID tags on every single product, even then it doesn’t really work. and completely not feasible economically. So, you know, if it were, if it were going to be, we’d be doing that business. But it’s just not, and that’s just the fact of life.

I don’t think that the human-less grocery store will scale beyond a hundred, maybe 500 square feet. It’s just not scalable.

[00:13:00] So the piece that you’re describing is our roadmap, right?

[00:13:05] So the three big prongs of our roadmap is, you know, reorganize the back room, with AI, deep learning predictions on what should be done. We put cameras in there as well.

[00:13:14]In 10 years, I think every single grocer will have some type of shelf sensor. And I think that I can make the case that it absolutely has to be shelf scan. We did that. We built the solution because of that analysis. In wasn’t that we built it and then looked for a solution.

Human Checkout As Theft Deterrent

[00:13:31]We’re trying to accelerate the checkout, but making it completely human-less. I don’t think that’s gonna work. I think people will try to steal, I mean, shrink is a huge line item. If you eliminated shrinkage, you’d double Walmart’s EBITDA.

[00:13:47] That’s like a huge deal, right? And that’s where their strength today, spending what 40 billion a year on cashiers. I mean, it’s not even 40 billion and they still have. you know, 3% shrink.

[00:14:00] David Zweifler, Editor in Chief, Gordon: [00:14:00] So really the cashier is, is almost like the bouncer, right? Like the security guard?

[00:14:07] Francois Chaubard, CEO and Co-founder of Focal Systems: [00:14:07] For sure.

We’re trying to accelerate the checkout, but making it completely human-less. I don’t think that’s gonna work. I think people will try to steal, I mean, shrink is a huge line item. If you eliminated shrinkage, you’d double Walmart’s EBITDA.

[00:14:10] And as they switched more and more to self-checkout, right? They’re forcing you through self-checkout. They’re there. You know, people like Kroger and Walmart are spiff based on how, you know, what percent they get in their store is on, through the automated checkout. And as they do that, they’re actually realizing that one they can’t get any higher (usage) two the customer experience — people don’t like it, and three, it increases theft. I mean, it’s just, those are just facts of, you know, does that outweigh the benefit of not having to have a cashier and not having human staff? That’s not for me to call.

[00:14:42] And I don’t know. I’m sure it’s different in different areas, but you’re definitely not going to see a bunch of self-checkout machines in high-theft and high-crime areas. I mean, there’s a reason for that. You will in the middle of New Jersey, you know, really nice suburbia areas, where there’s very unlikely that, you know, soccer mom’s going to steal a whole basket of, you know, a $100 basket.

[00:15:07] David Zweifler, Editor in Chief, Gordon: [00:15:07] I don’t know. I grew up in New Jersey. I think you might be giving them too much credit. You know, I, I saw the, you had done a podcast where you’re talking a little bit about the economics of Amazon, Go and it was, I mean, it sounded like it was almost being built as a showpiece but not for a marketplace that yet existed.

[00:15:32] What are, what are they doing there, do you think?

“Pragmatic AI”

[00:15:34] Francois Chaubard, CEO and Co-founder of Focal Systems: [00:15:34] We are, we’re trying very hard to develop a counter-narrative to Amazon Go piece. We’ve been describing “Pragmatic AI,” — things that can actually happen today rather than things that are not going to happen for another 20-something years.

[00:15:54]You know, we’ve spent some time with the Go team and, and you know, I’ve done these numbers that we’ve built our own demo and in the store, and, you know, you know, one of our core values of the hypothesize, experiment, measure, repeat. We just did a test. We did a store. in our, in our lab and said, what’s the economics that they’re going to cost?

[00:16:13] And, you know, the, a GPU, I put all the math, into that and you can talk to any deep learning expert and they’ll tell you the same thing.

[00:16:20]I worked on self driving car at Apple before I was, before I started this company, and the calculation we did, the power draw that we did, and I mean analysis that we did, is to run the algorithms to make the prediction to drive the car left, right, straight. You know, stop the gas, hit the gas, go forward. This is the same amount of power to push the car down the road. These chips are not efficient, so you’re spending just as much power to run the algorithm as long as to drive to push the car down the road.

[00:16:52] It’s super expensive. So now you’re telling me that you’re going to have hundreds of these cameras running inference all the time. Tracking people, those computers… computing is not free. You know, we take it for granted sometimes, but compute in GPU land is expensive as hell, especially if it needs to be even borderline real-time.

We’ve been describing “Pragmatic AI,” — things that can actually happen today rather than things that are not going to happen for another 20-something years.

[00:17:14] So these systems are not even close to scalable yet. And everyone is talking about, well, in the future, you know it’s going to be zero dollars. It’s going to be free. Well, is it? Because GPUs have been around for about 30 years and that’s not free.

[00:17:28] David Zweifler, Editor in Chief, Gordon: [00:17:28] Yeah. Moore’s law though right? , I mean…

[00:17:31] Francois Chaubard, CEO and Co-founder of Focal Systems: [00:17:31] Moore’s law is over. Moore’s law stopped in 2015

[00:17:36] David Zweifler, Editor in Chief, Gordon: [00:17:36] It’s flattened, huh?

[00:17:36] Francois Chaubard, CEO and Co-founder of Focal Systems: [00:17:36] Moore’s law is done. You can’t fit any more transistors on a chip without violating the laws of physics where we’re stuck at. I mean, literally, there’s (Nvidia CEO Jensen Huang) Jensen got up on stage and held up his P100 processor, and said this is the most transistors you can fit on a chip. You cannot, Moore’s law is done. You cannot get anymore smashed into this thing. It’s just impossible. The transistors would fall off. Jeff Dean did a whole paper on this and why they’re trying, why they’re pushing the TPU and maybe that is that method will work, but you’re not going to see a halving in cost anymore. That that ship has sailed. You might see a 10% improvement year over year. I mean, but definitely not half.

What Is Amazon Doing with Amazon Go?

[00:18:19] David Zweifler, Editor in Chief, Gordon: [00:18:19] So what’s Amazon doing? Like what, what do you feel like, you know, what do you think they’re doing there?

[00:18:26] Francois Chaubard, CEO and Co-founder of Focal Systems: [00:18:26] So, you know, I think that they’re… I guess I would say, they’re doing God’s work. Right? I mean, they are like, you know, they’re, they’re pushing the envelope. They believe that if they invest in the technology, it will bring in, it will accelerate. the GPU, efficiency, advancements, accelerate deep learning and stuff like that.

[00:18:45]But if you look at, I mean, the information leaked. Their internal P & L. It’s like, I’m not sure if I remember, the number is like spending like $3 million per store per Amazon Go store today. Right?

[00:18:59] David Zweifler, Editor in Chief, Gordon: [00:18:59] It’s a really expensive showplace for future technology. But they’re rolling them out all over the place. Right? So…

[00:19:11] Francois Chaubard, CEO and Co-founder of Focal Systems: [00:19:11] Not anymore. What was the last time that was built? Twelve months ago. That was the last one. Even though it’s supposed to be 3000 by 2020 or 2021 or something like that. Whatever the date was. I mean, they’ve stopped and the reason why they stopped is they can, the cost curve is not coming down the algorithm has plateaued. It’s the same reason why Waymo doesn’t have self-driving cars out there anymore.

[00:19:34] My friends work at Waymo. I just had dinner with one last week. The self-driving cars are not… the accuracy caps out. And you actually can’t get down into three, four nines accuracy, and we’re seeing that in ourselves. Getting to four-nines accuracy is nearly impossible. We can get our systems, what we’re at right now, to 96% or 97% accuracy and even getting to that is extremely hard.

[00:20:00] But every single percent beyond that, you’re talking about these edge cases that are insane. Like, you know, trying to say that, a tow truck carrying another car is just one car. It’s like, damn, that’s a hard edge case. I got to recognize there are two cars. Right?

They’ve stopped expanding Amazon Go, and the reason why they stopped is the cost curve is not coming down the algorithm has plateaued. It’s the same reason why Waymo doesn’t have self-driving cars out there anymore.

[00:20:15] David Zweifler, Editor in Chief, Gordon: [00:20:15] It’s an exponential increase in power for incremental increases in accuracy.

[00:20:21] Francois Chaubard, CEO and Co-founder of Focal Systems: [00:20:21] I think what they’re doing is they didn’t know that it would, that would cap out. They didn’t know that the computing costs were not going down as little as they are. They thought it would come down a lot more. And they thought these economics would pan out in 10 years, and I think that they’re seeing that it’s not, the trend is not suggesting that it’s going to pan out and they’re putting a hold on this whole strategy.

[00:20:44] And they basically bought the store in LA, but they had a big format grocery store that they’re going to try to do other stuff and it won’t be Amazon Go. I guarantee you. And you can look, I’ll take a bet with you on that. And it just, there’s no way you get into a bigger format store with Amazon Go.

Checkout Romance

[00:21:02] So the future of retail I think is, is going to be a hell of a lot more pragmatic…

[00:21:08] David Zweifler, Editor in Chief, Gordon: [00:21:08] Human.

I met my wife on a checkout line.

[00:21:09] Francois Chaubard, CEO and Co-founder of Focal Systems: [00:21:09] Yeah, it’s going to be humans in the store. They’re going to be back on the shelves until, you know, probably until you and I die. I think that is likely the case. And that’s unsatisfactory but you know, people promised flying cars in the 60s, and you know, they were let down. And I’m telling you that you’re going to get let down.

[00:21:27] David Zweifler, Editor in Chief, Gordon: [00:21:27] No, no, no, are you kidding? I’d say half of the romantic comedies that are on the screen would have to be rewritten if there were no, attractive checkout people.

[00:21:36]Francois Chaubard, CEO and Co-founder of Focal Systems: [00:21:36] I met my wife on a checkout line.

[00:21:40] David Zweifler, Editor in Chief, Gordon: [00:21:40] Really?

[00:21:41] Francois Chaubard, CEO and Co-founder of Focal Systems: [00:21:41] Yeah.

[00:21:43] David Zweifler, Editor in Chief, Gordon: [00:21:43] Oh so, you’re, you’re definitely pulling hard for the humans in the store. That’s great.

[00:21:50] Francois Chaubard, CEO and Co-founder of Focal Systems: [00:21:50] It’d be mean, it’d be convenient, but yeah, I don’t, I know

[00:21:55] David Zweifler, Editor in Chief, Gordon: [00:21:55] You’re partnering with Zebra. If you make those checkouts too efficient buddy, and you’re not going to, I mean, there’s another universe where you don’t meet your wife. Don’t make them too efficient, okay?

[00:22:09] This is David Zweifler from the Gordon Podcast Retail View. I’ve been speaking to Francois Chaubard of Focal Systems.

[00:22:19] The Gordon Podcast Retail View provides an in-depth look at the people and companies that are making news profiled in the Gordon Daily Retail Monitor.

[00:22:28] Click the link in the descriptor if you’d like to subscribe.

[00:22:31] The reporter in me loves making podcasts, but the marketer in me asks you to go to iTunes and subscribe and write us a review. It really helps.

[00:22:41] Be sure to check out our next episode of the Gordon Podcast where we’re speaking to Jens Levins, CEO and co-founder of Sitoo, a global mobile point of sale, or as the cool kids like to say, “MPOS” solution, who will explain how those technologies are helping to lower the barriers between the online and offline channel and improve customer experience in-store. We’ll also be talking about the dark side of the technology… basically, how he has eliminated bargain hunting from the retail landscape.

[00:23:14] Be sure to check it out on iTunes or wherever you get your podcasts or at Gordon Magazine itself at www.gordonmag.com.

 

David Zweifler

David is the founder of Gordon Magazine. David's experience spans investment banking, journalism, marketing and technology.

David Zweifler

Leave a Reply

Your email address will not be published. Required fields are marked *