Mike Graen dicusses Computer Vision with Pierre Marie Rallu - CEO of 1retail AI - about how CV and AI can help companies know what they have and where it is located, as well as specific use-cases and best-practices.  

Hello, my name is Mike Graen. Welcome to the

Supply Chain Management Research Council focused on on shelf

availability. I'm your host today and I'm joined today by

Pierre Marie Rallu, who's the CEO of 1retailAI. They are a

company that works very closely with SES-imagotag to understand

on shelf availability using computer images. Please join me

for with my conversation with Pierre. Thank you very much and

good afternoon. We are absolutely thrilled to have

Pierre here with us. We're going to spend a little bit of time

talking about our favorite topic, which is on shelf

availability, not a big surprise and sense this is kind of the

focus that we've had for quite a while in this space. But before

we get into the specifics of appeared, we'd love to have you

kind of unmute, introduce yourself to our audience, and

give us a little bit of background on yourself.

Thank you, Mike. Thank you for having me

today on the podcast. So my name is Pierre Marie Rallu, as you

just stated. I've been in the retail technology for retail

world for about 20 years, more or less in the forecasting,

replenishment and supply chain space; in-store, warehouses and

all of the above. I started 1retailAI a couple of years back

to help startups especially focus on AI technology to grow

in the retail segment, the retail vertical in the US. That

has been has been most of my career has been studying or

helping companies grow in that segment. So that has been a core

a core theme.

That's outstanding. Well, you and I have something in common.

I've been doing it 40 years, you've been doing it for 20

years. I guess the good news and the bad news is we're still

employed because there's still an opportunity right? Haven't

quite got figured out. Ever since the first retail store got

opened, we've always got issues with on shelf availability. And

if you are one of the things that I always ask my my my

guests on this on this podcast or this conversations on retail

take your day job away for a second

put that on the side, put yourself in the shoes of a

customer because the all that we can we can we can think about

from a technology standpoint, from an AI standpoint, at the

end of the day, our best experience is going to be being

a customer shopping in a store. So any stories that you could

provide to us where you actually as a customer went in to

purchase a product, and we're disappointed that it wasn't

there you don't have to call off the retail or the brand or

anything like that. But tell us how you felt through that, what

did you do about it, etc, it would be pretty interesting to

hear.

I think that's the key one, there are actually two stories

you want to mention here you want to talk about showing up in

the store with a specific need and not finding what you want or

going online buying something and not getting what you want.

Both of them are painful.

I have I have three daughters here we're

all very specific, where they want to eat and drink and things

like that. And we shop a lot online I mean my wife and I both

work so we tend to you know use Instacart or ship though or

directly the retailer websites and again no names but I would

say almost 100% of the time one or two products are being either

substituted or just removed from the delivery and a best case

scenario is substituted by something which is kind of

similar which is fine or worst case scenario that just moved

the Nutella to another brand and that's a disaster the house

every single exploding out there. So so so I think

everybody has as this as this as experienced already. And when

you show up at the store it's that was a couple of things

which are which are main pain painful main pain tickets for

the for the clients is really like Well, first of all, not

finding your products, and can be the product itself. It can be

the brand you want, it can be audited the fact to go and

search for the product and having to spend 1015 minutes to

look for the exact product you need in the grocery store can be

pretty large. So those are some of the experiences that you see

that that are that are that are an issue currently in the

customer journey.

Yeah, hey, I guess two other things is you know,

typically if you're looking and you think you already checked

the website and they said they have four of them and you go to

the shelf and they're not there. The obvious thing is to ask a

employee of the of that company Hey, do you have any of these?

Well, it says we have four and those dreaded words let me go

back in the back and look for them which basically means

you'll be standing there for a while before that person

probably are actually returns. The other one which is

interesting is either my kids or my wife will look at me go is Is

this what you do? Isn't this supposed to be fixed? Because

you guys work on this stuff? So somehow you get it? Yeah, not

only did I disappointed because it's not there. But you know,

somebody throws me under the bus because that's my job to fix

that. So well, that's awesome. Well, the point is, we all are

customers as well. So we've all experienced either ordering

something online, going to pick it up and store it and not being

there or going through and getting major substitutions or

eliminations of things that you really need in the store. And so

part of what we want to spend some time talking about is tech.

Now certainly there's a people and a process portion of that in

a supply chain, it says there's lots of reasons why things

aren't on the shelf. But one of the things that we do is try and

work against technology to at least least lets people know

about that. So tell me about your company. I know you're

integrated are working very closely with SES-imagotag,

you're part of them now. Tell me exactly what you do with with

Sure, yeah. So I started 1retailAI a couple

of years ago, in order to work with many startups, as many

with the company you have now.

startups as I can in the retail world, and help them grow in

that region. And as you mentioned, you're absolutely

right. I mean, there is technology, there is process and

there is people and actually it start with people and then

process and then technology. And quite a firm and the technology

can be really good. But if the technology doesn't fit into the

process, and if the people don't use that technology, well, this

technology is very useless. And therefore, part of what I'm

doing is looking at, okay, what does the return what needs what

makes sense to me? And where do I see value. And then I start

working with those companies. For example, I started to work a

year and a half ago with with a startup called be live.ai. And

they are specialized in computer vision, AI, and the And so

basically, what they do is you put cameras on the shelf, and

the cameras is watching the opposite shelf, and then taking

pictures all the time in near real time. And then there are

big AI algorithms behind the scenes that are analyzing the

products, looking at doing deep learning, understanding the

products exactly as you and I recognize a product on the

shelf, the AI does the same thing. And then they look at

okay, what is out what is partially out what is out of

place, or upside down all these kind of things, and then help

the employees to guide them through the store. The real time

basis had prioritized and step six. So I really liked the idea.

And we will come back we both lost the pension. But we'll get

back to it. But I really liked the idea. And it happened that

we live has been acquired very suddenly by SES-imagotag. And to

join the the business unit of computer vision of ACS. So so

now I'm very close to SES. Understanding how we are merging

the solutions, and how we were building the building that

vertical especially in North America. That's only a bit

though. And I just, I just spent the week actually, I just spent

a week actually with with with the SES leadership team in the

retreat in France, not too shabby. The chateau.

It only took you 10 minutes to tell me you just got

back from Paris. That's just not right.

We will, we will skip the champagne referenc

.... But, the company has a vision, which is very

interesting. And and I think it fits pretty well where the

retail retail world is right now.

Yeah, just and just for our audience. SES-imagotag

is an incredible company, I've worked with him quite a bit

actually worked with him when I was at Walmart with electronic

shelf labels, etc. Bringing bringing basically digital

intelligence to the shelf, right whether at shelf labels, whether

it's cameras, in your case, taking that camera methodology

and being a look at the opposite side of the of the aisle and say

it because it's really important. Everybody say well,

we should be able to look at a shelf and say there's a label

and there's no product. That's pretty easy. But there's also

there's a label and there's product, but it's the wrong

product. We call those incorrect products or plugs. We see times

where the price at the register and the price actually at the

shelf in the paper label world environment are different.

That's a problem. So talk to us a little bit about your method,

that methodology of literally using computer vision. I'm

saying I'm assuming it's real time computer vision to to

identify what the opportunities are on the shelf.

Yeah, absolutely. Let me let me give a

step back, which is one thing which is very important to

understand as well as maybe understand some of the forces of

the market and why that makes sense now. I mean, you mentioned

this is a good it's a good thing which is actually 30 years old,

which is in the technology word like a dinosaur. But in the

meantime, it really, really behave as a startup or multiple

startup, I would say. And, and so the company, as you

mentioned, as being has been, has been known the number one in

the world for ESL, so the electronic shelf labels, which

is their historical job. But they have recently transformed

under CEO Thierry, Thierry Gadou and to become a real offer

around digitization of the physical store. And what we see

right now is, I mean, ESL has been in Europe, on the shelf of

every grocery store for 20 years. I mean, I grew up with

ESL, and not 20 years old. And the the, and surprisingly,

arrive in the US in early 2000s. And I didn't see a single ESL

anywhere. And as of today, I mean, it's very, very rare to

find the essays on the shelf, when we're talking about we're

talking about two or 3% market penetration, which is compared

to 6065. And in Europe, and what we see is that the market has

evolved itself, the US market has evolved, I mean, under the

under the global pressure, the rising labor costs, the energy

costs going up the and then the e-commerce, the pressure of

e-commerce, to align the experience, customer experience

in the store. All of that is bringing a lot of pressure on

transforming the brick and mortar experience. And a lot of

that is, and this is why the office is makes a lot of sense.

Because the fact of digitizing the store - basically realign a

little bit - the econ and the brick and mortar stores, stores.

Got it. That's great. That's great macro.

And they are and they are and then after of

course, we got to talk about the ESL, which is blooming in the

US. And so being being in that market right now makes a lot of

sense. But then above and beyond just ESL, which are driving

dynamic pricing and promotion policies and things like that

much faster and much easier. There are a number of

technology, including computer vision and AI that are also

going to help you digitally that story. Yeah, digitalizing that

store.

I know that I know, the electronic shelf labels, I

know that's not your specific area of focus. But not only does

it provide you consistent pricing and correct pricing at

the shelf, so that shelf price and the price at the register

are the same price versus a different price. But also you're

building intelligence into those now we're literally, if I'm

picking items for a customer, as I turn around the corner, and I

have two of these that I'm looking for, they have LED

lights that can blink, blink to me and say, Hey, I'm right here,

pick right here. Rather than, where's that item, there's a

whole bunch of items. I mean, when you're talking about great

big, you know, things that paper towel, it's not that hard when

you're thinking about cosmetics or things that are very SKU

intensive, that that that will be called picking operation

becomes easier. And from a stocking standpoint, allowing an

associate or a employee to pick out an item, scan the UPC and

have it blink to where it goes, rather than trying to find in a

big four foot section where it's supposed to go, I think is a

game changer. Let's focus back on the auto stock work because

that's the most exciting part of this. So it's exciting time to

be part of retail. So where is AI and computer vision today?

And where do you see it kind of going in the future? So starting

with kind of the US market? Because I think you're right is

it's a little bit behind where Europe is? Where is it today?

And where do you think it could go in the future.

So, curiosity what we see, so if you

think about the supply chain in the in the retail and most

supply chains, the purpose of the supply chain is to bring the

right product at the right place at the right price, right. So in

the written world, that will mean that when you want to have

the product on the shelf, when the when the consumer and the

customer is coming in look for it. Well interested in almost no

retailers can actually know that the product is on the shelf at

the time they know it. They might know best case scenario

they might know they have it in the store. But they don't know

if it's on the shelf when the customers looking for it. And

you think about all the for the messy recesses the debate

expands in terms of supply chain logistics to the store the

store, or the or the or the automation of the replenishment

or the planet prime automation and so on. And still I mean, we

can't say if the product is there, so I think this is what

This is that vision that we're bringing with that technology,

when I say vision based, but this computer vision that we're

bringing, the idea is to have an AI on the shelf, that looking at

those products permanently, real time or near real time. And that

either to address the information that the product is

out is low or, or short. And then being able to drive the

store operations and the supply chain in order to improve that

stock position. So they are coming back just a little

background. And going back to what you said. I mean, I would

say Europe and the US are about the same level on this one esse

a different ballgame for sure. But Computer Vision we're

talking about and emerging technology. And we will come

back to it. And so I would say the players in Europe and

players in the US are tip-toeing at the moment on where we are.

Yeah, so just got a text message from somebody. So

I'm going to ask you a question that's kind of off script here a

little bit. As you think about a big mass merchandiser like a

Walmart or a Target or somebody, you got some product on certain

parts of the store that computer vision makes all the same. You

have other product categories, like apparel, and maybe

electronics, etc. That may make may be different, right? So how

does computer vision? How does confuse you? Because I don't

think computer vision is a one stop solution for everybody. I

don't think RFID is one stop solution for everybody. How do

you see these technologies working together for different

kinds of categories, like apparel versus boxes of cereal?

That's a very good point. So computer

vision is good for what you can see easy. I mean, when you have

the you have a box of cereal on the shelf on now they have. So

and they will divide that. And then there are areas as you

mentioned, it's very difficult to recognize a black T shirt

size L from a black T shirt size average just by vision. So that

is a little bit of it's a little bit of the limits to certain

degree. But just just computer vision itself has been evolving

pretty fast. Meaning that, for example, I mean, over the last

six months, I spent a lot of time on produce. And the

capability to actually recognize the produce, making a difference

between the banana and orange and Apple understanding the

level of the lottery ever produce on the shelf. With this

limitation, of course, I mean, but we will, and understanding

your so the directness of the product is a product, I mean

mature, not mature, how long time has been standing on the

shelf. So looking because the cameras are looking at the

product, I mean, we can see if a product that has used by a shelf

life of a couple of hours has been sitting on the shelf for a

couple of hours. And then there are a lot of business cases that

are being being that are being developed above and beyond the

standard grocery grocery or GM merchandise, or GM categories.

So there are there are a number of number of categories that are

that are good, very good candidates for for computer

vision. And then you mentioned I think those are technologies

around the store and, and the store digitization and that's

when it'll be the idea of ACS is not a one stop shop of one

technology, you're going to have a number of technology that are

going to be complementary. For example, you can have smart,

smart brochures that are counting products on the shelf,

you can have scales that are weighing the product on the

shelf. And then when you start having this mix of committed

this mix of technologies, we start having a very much more

accurate visibility of what you have on your in your store.

Okay, well so you just opened up a very

interesting conversation because typically people say, Well

computer vision is great it tells you if the products on the

store. But unlike things like RFID computer vision can't count

I can see the first one but I can't see the ones behind it.

You just uncovered something those pushers the weight sensors

talk to us a little bit about the role that they play in not

only the product is available, but how many are on the shelf

and it also could generate a signal to a store associate

saying hey, by the way, a new case can now fit it couldn't fit

a few minutes ago but it can now fit walk us through that.

What this example of the case shows that

you know retail pretty well. I mean understanding that you have

one or two cans of soup is the one thing. Understanding you

have room for a new case of soup is another thing, right? Because

when we're thinking demo replenishment, I mean we don't

buy it by the SKU right we buy by the by the PCB we buy by the

case or the pack. And so and so they they are the this is the

this is the this is the this is where this is where the

replenishment is understanding replenishment is very important.

It's not just a number of understanding the exact

quantity, which chef rather that, is it the time to order a

new pack on that. And so, when you think about just computer

vision, what can the computer vision do work, computer vision

is going to look at the horse on the shelf, you're going to look

at, if a product is, is low on that, we can see that if there's

one facing of two facing or three facings. And the computer

vision will be able to, to understand if a product is not

at the right place or not. Okay, if your your planograms, you're

supposed to have to face things, and suddenly you're three or

four because you're missing a product on the side. I mean,

that makes sense. And so those are some of the business cases

that computer vision can solve. And then it's going to guide the

employees toward what actually matters. Because at the moment,

if you get a standard grocery store, why would you go you're

gonna go and scan your gaps and check if I had my zero. Do I

have quantity on hand, is it going to be delivered today? But

you have to go check your your, your your scanning your gaps on

a regular basis. The problem of doing that is even if you manage

preparatory inventory, when you look at the shelf, you don't

know if there is definitely inventory until you check it.

Well, that's what computer vision will do for them, we're

gonna go to receive the file of your balance on hand, and

compare that with what you have on shelf and just manage the

exception. So definitely don't certainly don't have to scan the

2, 3, 4 thousand holes you have on the shelf, but just maybe the

five or 600 that are actually never in the store. And when

they say actually, maybe they're on the back room, maybe they're

on the riser, if you have an inverter on the riser, or maybe

the the quantity is wrong. And then you need to adjust the

quantity. And when you do that on the daily basis, what you'll

start to you start to to have a much leaner replenishment when

every day you correct your balance on it. And we can talk

about why the middle southern discrepancies. But when every

day you are cleaning your your stock in the store, but your

punishment gets better. Your punishment gets better gets

leaner, your your other stock goes down, but your inventory

level goes down as well, because you'll have excess inventory.

And all of that drives a much leaner supply chain because your

your demand, which is the store much more accurate. Okay,

perfect. Second part of your question, just finishing you

asking you about the pushers. What the pushers are going to

RFID ideas? Well, everything which is attached to the unit

will be able to actually count the product. And counting the

product is a very important, it's very important information,

especially in specific categories. When if you're going

cosmetics with very high margin high price products, when a case

is expensive. So you want to make sure that you're ordering

at the right point. So knowing the number of products you have

in this category, pharmacy, cosmetics, and then we can go in

after hours. Well, I mean, we have different way of counting

the counting the product on the on the on the hanger. And so and

so those technologies provide a much more granular visibility on

the accounts. And that enables more precision in your replenishment.

Got it. So JW has asked basically a really good

question. I think from from my perspective, he was interested

in, hey, give me a one give me a one or two or three use cases

for computer vision AI, versus kind of what the typical rain

RFID which is basically UHF RFID that a retailer might choose.

What are those use cases from a computer vision that you think

stand out as a differentiator?

Well, first of all, as far as I know, I know only one chain that

has enabled RFID across all its product in its entire chain. And

that's Decathlon, a sporting goods chain, and their

experience is fantastic. And when you pick whatever you want

in the in the store, everything 100% of the products are RFID.

And then you go through through the gate, and then it flashes,

the RFID and then the building. So you still have to go and

bidding in the case but no counting no need to burden. I

mean, this is a fantastic experience. But they it was

never for it. They were able to deliver it for a couple of

reasons. First of all, sporting goods company. So I mean, we're

not talking about you're talking about fairly high price. We have

a [?] and they're almost - I don't know the exact number but

they have a very high level of privacy but they manufacture

probably 50% more of their goods. So since they manufacture

both embed the RFID from the manufacturing process, and then

spend a lot of time for those are 50% to RFID themselves in

the warehouse. So that's an investment that seems to be

paying at the store level. So RFID is great. But if you take a

Walmart's that have, I don't know, 120,000 SKUs, a lot of

them are very low cost. And enabling the RFID is very

complicated, it's actually not done at all. And therefore,

therefore, what we look at computer vision, because it's

much easier to enable it's much more realistic in the store at

the moment, and provide a mass solution. So ideally, every

single day RFID, and then life is great, well, reality of the

supply chain is different. So, of course, I need the business

case, what was really what I'm trying earlier, is really like

being able to be super specific on how to get organized in the

store, based on the knowledge of the data we have, what

computation does, that brings a new set of data, which is the

presence of the product on the shelf, constantly. And, and the

idea, the idea is really to be able to change the padding love

having people which are maybe dedicated to grocery or

dedicated to GM, that have to do their product, maybe with a more

versatile workforce in the store, which is more task

driven, rather than rather than schedule driven.

Excellent. So again, I got a text from somebody. So

so this may be a little bit a question that you weren't really

expecting but I think it's an important one: there are other

solutions that are computer vision solutions that either are

crowdsourcing, ie go and take the picture and do the analysis

there. I'm thinking of things like Field Agent, Trax, etc.

There's also a suite of solutions, which are robotics

base, you know, Badger robot, Brain robots, Simbe robot, some

of those folks who are literally scanning items on the shelf. And

reporting that same back, obviously, fixed cameras, is a

different solution, walk us through the differences between

somebody taking the picture versus a robot catching the

picture versus fixed cameras, from your point of view.

Yep. So there are there for them comes

in every technology, I think I've been playing with all of

them. So an island, the fixed cameras, but so when you talk

about when to take pictures, using your mobile phone, and

things like that, the advantage of that picture that you have

the flexibility to have a very neat picture really informed of

what you want to focus on. So So and then and it's very, it's a

very light solution, there's not a lot of investment upfront,

because you don't have to install anything. But you need

people to do that. And the fact of having the the folks doing

that work that has a recurring costs, which is extremely high,

which is the folks taking the picture. And from from most of

the the brands that have worked with Coca Cola. Cheers to those

guys. My wife works at Coke so we have some common things

there, disclaimer. But a lot of the feedback I had is the lack

of consistency of the pictures. I mean, you gotta take, you

gotta take the big because yeah, you have an army of people.

They're just taking pictures. So you need to train them. Taking

your pictures easy, but taking the right picture. So the AI can

analyze the right data. I mean, do I have the right frame? Do I

see everything? Do I Do I Do I Do a corset architect did I miss

one of the shelf one of the shelfs or something like that.

And a big one of the issue I've heard about is consistency.

After consistency. It's not a big deal. If the business case

is zero, for example, merchandising, because

merchandising doesn't require as much velocity as inventory. So

you can have a view of your shelf once a week, once every

two weeks, and statistically you can start looking at okay, am I

compliant I might not comply with and things like that. And

so, for merchandising, especially for the brands, I

mean, that's what's your show? That seems to make sense. Now.

There are a lot of brands that we talk to that want more. They

want to have much more detailed KPIs, they want to understand

the inventory position of their products. They want to

understand the dynamic of the product like a heat map. How

fast is it for like moving during the day, because that's

what what fix camera can do right there taking pictures all

the time. So they can analyze, even if the product is out where

the product moves, whether the product removed and put back and

creates a heat map of the interaction between the clients

and the chef. So the effective the effectiveness of the

merchandising, if you will. So that's one thing. The other

thing is you mentioned the robots, which on same day, which

was on my brain, so the robots are so have a great, great value

in the sense that when they are, first of all, they are very

close to the shelf, they have a very net and again, they're

usually they have very nice sensors with good lenses, they

can, they can then either read the barcode and everything. So

there are a lot of information associated with orbit and they

can write continuously in the store that say, a robot will

give you a picture position once or twice a day at the maximum

unless you have an army for I don't think that's the point.

And so and so it's going to give you one or twice a day in auto

position which is good and as well as merchandising position.

So I think the part of the part of having the robots is

interesting is interesting for that perspective, after there is

the there is the customer perception of the robot in the

store. Some like some dogs. I mean that that's one thing and

and the fact that you also have a potential single point of

failure that doesn't start work at all today. So if you're

looking for like, if you're looking for like some solution

that gives you really looking forward to take care of your

other stocks. I personally think the shelf the shelf eight

cameras are the best. But if you're just looking for to take

care of a merchandising, pain points or a branding thing for

the nose, a solution might work.

Well I hope you enjoyed that conversation with

Pierre from 1retailAI. Join us next time as we resume the

conversation to discuss more about leveraging computer vision