Mike Graen sits down with a variety of guests to talk about the importance of Computer Vision (CV) product recognition software to take digital images and transform them into on shelf availability issues.

Joining Mike are Anand Prabhu Subramanian (Infilect), Angam Parashar (ParallelDots), Renish Pynadath (Snap2Insight), and Matt Greene (TRAX Retail).

Hello, my name is Mike Graen. Welcome to today's

podcast on on shelf availability hosted by Conversations On

Retail and the University of Arkansas Supply Chain Management

Research Center. We are going to spend some time today talking to

leading edge product recognition software providers. Whether you

have a robot that is scanning products in a shelf on a store,

or you have fixed cameras, or you have a service that allows

people to go in with their own devices and take pictures. All

of that is requires a ability to be able to say what exactly was

seen. Product recognition in store condition software does

that, and we are we're very fortunate to be able to see from

the leading product recognition software providers. So let's get

started with the podcast.

Well, hello, everybody my name is Mike Graen and it is a

pleasure to be with you. We're actually recording this on a

Friday afternoon. So for those of you who are joining us live

happy Friday. We've got a really, really super exciting

panel that we're going to spend some time talking with today.

The topic is continuing to be on shelf availability, and

specifically being able to use computer vision software product

recognition about exploring in store conditions. So I'm going

to go over a couple of kinds of things that we want to cover

first, and then we're going to introduce our tremendous panel

who's with us today. The first thing I want to kind of walk you

through is just a little bit of background on me, I've been in

the retail and supplier area for over 40 years. I've been with

both Procter and Gamble, Walmart and a third party merchant

called Crossmark. And now really work with suppliers and giving

back to the University of Arkansas and Conversations On

Retail for information about on shelf availability kind of

capability. So one of the things that I think it's important to

know is just kind of what the ground rules are. Conversations

On Retail is helping us to co host this along with the

University of Arkansas, we want this to be very interactive. So

we're going to ask you to be prepared to answer questions or

ask questions, there's a couple different ways you can do it.

One way you can do it is after each section of this

questionings that I've already got for these folks, we'll go

ahead and come off mute and if anybody has wants to act, ask a

question live and interactive, you're more than welcome to do

that. If you'd rather stay and just be anonymous, you can text

the actual questions to our chat function and when we get a

break, I will certainly ask those as well. When you're

asking the question, we'd asked you to keep the video on, use

the chat function. And then we've got, you know, the obvious

thing is we're gonna run this under antitrust guidelines,

we've got, you know, four incredible companies and leaders

who are all basically direct competitors of them's of each

other, but they're all in the same basic space. So the

purposes here is we're not talking about pricing and

margins, or anything that would look like competitive or

antitrust violations. We're looking to educate the industry

about this role that this starts. So a couple of things,

just just a couple of introductory things that to kind

of set this up. First off, I want to thank Conversations On

Retail and I know Matt Fifer is actually hosting this as we

speak. Matt's a great resource and Conversations On Retail

continues to provide education for suppliers and education for

the retail industry around that. So I've encouraged you to, to,

to follow the Conversations On Retail link on retail to see

more about that about getting involved with that. And then

obviously, the University of Arkansas Walton College, which

is the number one Gartner supply chain university, a part of the

journey university. We're excited to have them post both

the audio and video podcast to that. So we appreciate them very

much. So a couple of background kind of thing. So this all is

about on shelf availability. And we typically will see the Sam

Walton quote about making sure you're meeting customers needs.

I actually found this the other day, this is Doug McMillon,

who's the CEO of Walmart. And I thought this was pretty, very

aligned to what Sam Walton was saying, but it's really pretty

simple. If you're not meeting the wants and needs of the

customers, you're done. There's not a lot of loyalty there. So

if customers go in and they want to buy something from a store,

and the store doesn't have it on the shelf, guess what they're

going to get the product and they're going to find whatever

way they can to do that. They're not loyalty to a particular

retailer, their loyalty to the product they want. And these are

a couple stats we've shared before, I'm not going to read

them out loud to you, but basically, if you don't have the

product on the shelf, or if you have out of stocks or whatever,

customers are very resilient and they keep you know Amazon. Well

thank you very much for the additional business of pushing

folks that way. On shelf availability, and this is why

these group is so important is you look at a forefoot section

of, of shampoos and conditioners, this happens to be

head and shoulders at a Walmart. You obviously see the one

particular case where you have an out of stock, what you don't

see is there's a lot of things like missing labels, suppliers

typically call that distribution void, you've got pricing

situations where the price at the register and the price

actually at the shelf are different. And then you have

opportunities like plugs, and these guys will sure talk about

that, but that's where you have a label, you have the product,

but they're different. The product, a product that's there

and the label is there is different. So as you step back

and look at it, there's a lot going on with the shelf that's

really wrong, but most people will only focus on that red box,

which is the out of stock. And that's where these guys I think

really will come into place. Because if you think of a

retailer like Walmart, that's got several 100,000 items, and

several hundreds, probably thousands and thousands of

stores, that's a lot of things that can go wrong. And so we

think that in store conditions and product recognition can play

a very important part. And I'm gonna cover these really, really

quickly because I know we've talked about them before how to

measure OSA algorithms, but in store audits in store audits

where somebody's actually going in like a Trax or like a Field

Agent and taking pictures. It's great that you got the pictures

and we've actually got Matthew here from Trax, going to talk a

little bit of what's the behind the scenes sauce that actually

turns that picture into information. Then we got cases

like its shelf scanning robots like Woodman's has, they have a

shelf scanning robot that will go into the store and literally

scan up and down. But something has to take that information and

turn that scanning into something that's actionable,

like what where's the out of stocks, or where's the plugs.

And then a couple of others ones that I think are appropriate are

things like Focal Systems and some of the companies that we've

got, like InView, and Focal Systems and SES-Imagotag who

have fixed readers, same thing. I've got a fixed camera that's

taking a picture, something behind the scenes has to take

that information and go, Okay, what's out of stock, what's

incorrectly priced, etc. So all three of those I think are very

relevant to the conversation we're gonna have to have,

because what we're going to talk about now is, okay, what does

this data capture mechanism, whether it's a robot or a fixed

camera, or somebody doing an audit in a store, we got to do

something with that, and point out what the issues are and

that's what this group is going to talk to you about going

forward. So without further ado, let me go ahead and introduce

this group and we're going to have them unmute, and give

themselves an introduction. We'll just do it in the order

that we've got here. We'll give them a few minutes to give them

an introduction and a little bit about their company. A couple of

them have sent some slides so we'll I'll advance those when

you're ready. But the first one is Anand from Infilect, and,

wow, I believe it's what you know, almost 12 o'clock

midnight, on a Saturday, you get you get to travel, the farthest

travel award. Congratulations. Go ahead and unmute yourself and

tell us about yourself and your company, please.

Great. Thank Thank you, Mike and Matt, for

arranging this podcast. I mean, I've listened to some of these

before, and I'm quite excited to be part of this panel. My name

is Anand Subramanian and I'm the co founder and CEO of Infilect.

Infilect is a retail intelligence company. We

specialize in computer vision and image recognition technology

and help global CPG companies and retailers with in store

retail execution insights. So primarily, we work with visual

data, and then convert the visual data into intelligence

and actions that that can that that can be taken in the stores

so that the reps can improve the install execution. As the slide

says, we are a visual intelligence platform, right for

the worldwide CPG brands and retailers. Mike, just move on to

the next slide, please. Yeah, so just a quick understanding about

the problem that Mike just introduced right OSA, on shelf

availability and in store execution. Pretty much every CPG

supplier would want to understand what exactly is

happening in thousands of retail outlets where their products are

placed. A lot of these CPG suppliers have been using

traditional manual audits, right? I mean, all of us know it

is slow, a lot of human errors can happen and not very

efficient, right. So we work in this space of completely

transforming this audit process into a simple click and go

process and use image recognition technology to help

the CPG suppliers, retailers, and even merchandising companies

to understand and execute in store execution in a better way.

Yep.

Thank you so much.

Got it. Maybe I'm next.

Thank you. Thank you, Mike. And thank you, Mike,

Correct.

for organizing this this podcast, really appreciate it

and glad to be a part of it. Good afternoon, good morning,

good evening, everybody, wherever you are. I am Angam, co

founder and CEO of Parallel Dots. At Parallel Dots, we are

building a computer vision platform for CPGs and retailers

globally. So essentially, a very similar problem statement that

Anand just talked about. We again work with the global CPG

brands, and retailers to help them understand what their

visibility is on the retail shelves, help them minimize that

out of stock, help them maximize their OSA numbers, their

compliance numbers in general. We do it by analyzing pictures

of retail shelves by passing them through our proprietary

computer vision algorithms through which we can determine

important retail execution KPIs such as you know, of course on

shelf availability, share of shelf, planogram compliance,

pricing compliance, those kinds of things. And we also help our

clients adopt these solutions at a large scale so that they also

start getting benefits, you know, such that it either

increases their sales or decrease the cost, helping them

improve their, their, their numbers. Right, so we also focus

a lot on that aspect.

Awesome. Renish, your next.

Thank you, Mike. Hi, good morning, good

afternoon, good evening, everyone. This is Renish, co

founder for Snap2Insight. We are headquartered in Portland,

Oregon. And again, operating the same space. A retail analytics

provider leveraging AI and computer vision to help CPG

suppliers and retailers and to measure and improve their shelf

execution. We also operate in the space of helping some of the

big suppliers to do space service, which is in some sense

a precursor to space planning, and followed by space retail

execution. So we we today use AI and computer vision to help

these suppliers mostly and sometimes their third party

merchandisers or brokers, or their distributors who go into

stores to better execute against their plan. So yeah, in a

nutshell, that's what we do, again, using computer vision and

image recognition technology. Nice to meet everyone here.

Awesome. Thank you Renish. Last but not least, Mr.

Greene. Matt, you want to unmute and tell us about yourself and

your company.

Yes, thank you. Thanks, Mike, for organizing

this, and thanks to Anand, Angam, and Renish, for your

contributions to what I consider a really, really cool field. I'm

excited to talk to you guys about our perspective from Trax

in artificial intelligence and image recognition. This panel, I

think, is going to be an awesome way for all of us to learn from

each other and I'm really excited kind of for the next

steps. Our companies are really aimed at the same main

objectives, I won't bore you guys and tell you exactly what

it is that we're out to solve. But in a nutshell, it's

optimizing your in store field execution using artificial

intelligence. I've been with Trax for about five years, I

came to Trax by way of an acquisition in 2018. And prior

to that worked at a company called Currie, which kind of was

one of the early pioneers of in store crowdsourcing. Previous to

that started in CPG, at Nielsen. Learned a lot about what

customers were looking for in store and that there was a real

market need in this industry. And as I said, I'm really

excited to talk to you about what it is that we've learned

since then. If you wouldn't mind, Mike just advancing next

slide. So a little bit about our company. We have a very global

reach, but we like to think it's very local impact. We are in

over 90 countries worldwide. Our, America's headquarters,

where I live is in Denver, Colorado. Our R&D headquarters

is in Tel-Aviv, Israel. And as I mentioned earlier, we have a

couple of companies that we have acquired over the over the

years, one of which is where I came from, that really allowed

us to look at retail execution not just as AI but as kind of an

end to end service model. And we'll talk a little bit about

that in a sec. Next slide, Mike.

Oops, sorry.

Yeah, so we work with 30 of the top 50 CPGs or

there about and as I mentioned, kind of partner globally. You

know, we feel like our global reach and impact is definitely

something that many CPGs of this side need. Next slide. And this

just gives you a little bit of understanding of kind of what

our core technology does. So we have really two main components,

we do production or I guess image processing of a shelf in

the cloud. We also have on device recognition, both of

which are kind of served for the purpose of a certain use case,

whether it is a customer that is in field that is trying to be

more accurate and efficient with their day, or whether it's a

headquarter call that needs more information, more understanding

of what's going on in the field. We have two kind of primary

methods of recognizing the SKUs that are on a shelf. Next slide,

Mike. What happens here and there's there's a bit of a build

is these images come back, they are then stitched together

through our engine. And you can see basically simulate what's

going on in an aisle, allowing anybody from a headquarter

perspective or a field perspective to see exactly

what's what's happening in that aisle. And part of the value of

this is putting together disaggregate images that come in

from the field and turning it into a real picture. So you

could virtually see walk and kind of understand what's

happening and in multiple different, you know, regions,

cities, countries globally, and it leads to a very cool

experience. And finally, next slide. And then obviously, one

of the most important things, if not the most important thing

about our company is we try to take all of this information and

turn it into a really translatable asset for you to

react to. And I think that if you talk to any of the other

panelists here, they would say the same thing that data is, you

know, information overload is a real thing right now and if you

can't translate that to something that matters for the

customer to make a real decision on, then you're probably not

serving as much of a purpose as you need to. And last slide,

Mike. I want to just talk a little bit about where we're

seeing image recognition going and I'm sure this will come out

kind of in the further panel discussion. But we'd like to

think of image recognition at Trax as a backbone, almost a

commodity of what's going on in store and the value of image

recognition today, it's very different than the value of what

it was five years ago, ten years ago, when it was a very new

early stage technology. So of course, it's the table stakes

you need to do, you need to annotate what's going on in the

store. But what really matters is what you're doing to react to

it. So we're very much invested from a company perspective in

building an on demand flexible labor source that actually

executes on the results of image recognition to close holes as

quickly as you know a customer needs them to be when a shelf is

not looking like it needs to. And then also, we're investing

very much in shopper activation. Mike made a comment earlier

about changing consumer behavior, and how that's a real

thing. How if a product is not on the shelf, you're almost

instantly going to lose them to a competitor. And we're very

much focusing on maintaining consumer relevance through a

shopper activation field force that again, both of these

technologies are leveraging image recognition, to maximize

what's happening in store. And again, that's a little bit about

Trax, looking forward to the next stage of the discussion.

Awesome, thank you very much, Matt, appreciate it.

I'm gonna I'm gonna leave these particular things up here,

because what you'll see is QR codes for each one of their

LinkedIn profiles. If you have any questions that you want to

ask specifically to one of them, then that'll give the folks

online opportunity to do that. Got several questions here, but

actually, I think some of you guys have already answered them.

So I'm going to do some combination, little questions

here, which is the role of computer vision okay, you've

showed me what you've done. But what are the real things that

you're trying to solve? What are you guys solving for? And the

second kind of follow up question is, who is the

customer? Is this a retailer that wants to know how they're

doing? Is it a brand owner or a CPG company that wants to know

how they're doing? Is it a third party merchandiser? What

exactly, so what is it doing, and then exactly who your

customer is. And I'll just, I'll just open it up for all of you

guys, and just come off mute and try and address that question.

Okay, I can take the first question. I think

computer vision again is a means to generate or digitize the data

that we generate from what from the stores and the most of the

value comes from the data, right. And the data that is

coming out of this algorithms is needs to be analyzed and

insights and actions need to be generated and that is where the

key is. And it is again, just measuring the data is not going

to help anyone, right. I mean, it's probably the first step but

taking the data, taking the insights and actually making

them into actions and then executing them in the store is

what is going to help suppliers or retailers or even the, the

merchandising agencies that you mentioned, to improve their

outcomes, right. So that is where I feel and but but

computer vision is a great tool, it has tremendously improved the

overall process of collecting retail execution data from from

stores, right, because of all the automation and the accuracy

advantages it brings them right. So that is my thought.

So what I heard you say is it allows you to

understand what's going on in store. And I heard I think I

heard you say as both retailers, CPG, brand owners, and

merchants, merchandisers potentially are customers of

this. There are probably different ways that data is

collected that we'll get to in a second but all of them want to

know what is exactly going on on the retail shelf. Anybody else

have any builds to that?

Yeah, I can add something to it. So yeah, I

completely agree with and on this, you know, and what we have

seen from our experience is that there's so many different

players who can who can benefit from it, right? We have seen

well CPGs, one of our first clients to begin with. And of

course, there's a clear value prop for them right how they can

make their sales reps or their other agent field reps in a more

efficient, but we have seen increasing interest from

retailers and all format retailers beat from the largest

supermarkets, hypermarkets or even convenience store chains

all across the globe, being interested in this where their

primary objective being they want to reduce out of stock,

they want to increase their availability and reduce out of

stock. And to also merchandising agencies all across the globe,

they have seen, they have also they're also realizing that how

they can save their reps by their stores, right? And make

the entire piece, entire supply chain much more efficient,

right. And everybody benefits from it right. Not just

retailer, the CPGs, and agencies, everybody benefits

from it. So this has this has been what, and merchandising

agencies and retailers are seeing increasing adoption from

from these two different layers.

I would agree with everything you just said. I

mean, I think we look at it like this is a really, really dynamic

industry where consumer behavior is changing on the fly. And

honestly, merchandising behavior is changing on the fly too. If

you're walking at least in the US, if you're walking around a

retail store, even during peak shopping hours, it is not a

strange thing anymore to see five different people and five

different badges doing five different things at any given

time, merchandising really in real life. And our goal is to

make sure that we are informing that merchandising activity on

behalf of a brand with the most updated relevant in store

condition information as possible. So that we can put a

product on shelf for on behalf of that brand or the

merchandiser can put it can put a product on shelf on behalf of

that brand with as minimal downtime as possible. Talking a

little bit about changing consumers, Mike, you made a

comment at the beginning that is so spot on that if a product is

not on shelf, the consumer will not be brand loyal, and they

will pick something else. And anytime that that experience

happens to your everyday shopper, the brand is basically

missing an opportunity to keep a customer and there's a decent

chance that they're going to find something else in real time

and ultimately make a different decision that could impact their

their behavior in the in the, you know, future purchasing. So

we aim to make all of that a more seamless process, make sure

that all those merchandisers that are always running around

the store, especially now, have the right information and

merchandise in a really effective way.

Great. So

I want add one more thing.

Yeah, go ahead Renish. I was going to call on you next.

Yeah, agree with all the points that I heard here. One thing

that we also started to notice is that, using the power of

computer vision, we are able to see the shelf conditions, but we

also see a role for technology to help the merchandiser or the

field rep to take that next best action on the shelf. How can we

reduce their cognitive load by using technology. Suggesting the

next best alternative in terms of which product is stocked, if

the shelf has an out of stock, right? How to make real time

decisions, right? Without, and without data, it's very, very

difficult, right? And sometimes the folks at the shelf are in

some sense, not that skilled to be able to take those heavy

cognitive load. And they are sometimes working against very

aggressive timelines, right of covering many stores across

several categories. Right. So how do we enable them to take

those right decisions on the shelf? Just like we are helping

a shopper to take those right shopping decisions, right? So we

see that as an addition extension to what we do, in

addition to looking at how the shelf is.

Okay, thank you. Great answers from all of you.

I'll put a couple of you on the spot here. Matthew, I'm gonna

start with you. How are these images collected? Obviously,

what your model is somebody go in the store, whether it's a

retailer, or merchandising, or brand or take a picture. We also

talked about earlier about robotic cameras, or I'm sorry,

shelf scanning robots in store fixed cameras, etc. What's the

most predominant way you see today the images being captured?

And do you think that's going to change over time?

Yeah, it's a really good question. Thank you. You

know, I think the most predominant method certainly

will change. But from my perspective, it is still human

with a mobile application. And that you know, that human is

somewhat agnostic. It could be someone from a mobile crowd,

right? Someone that has an app that you are signed up that is

using, you know, to go in store, get paid and collect, collect

data. It could be a merchandiser themselves, it could be someone

from a traditional broker, like Advantage, like Crossmark, like

Acosta, like any of the above, or it could be an actual sales

rep. And so that is certainly from Trax's perspective where

most of our business is, is with a human using a mobile

application. But the human might kind of look and feel very

different based off of based off of you know, who they work for,

and how they work. Now, going forward, we also have fixed

cameras, that has been part of our company strategy for some

time. And the holy grail is to basically have this action done

through, you know, through not having humans. Whether that's a

drone, whether that's a camera, whether that's a dog that walks

down the aisle with, you know, a phone attached to them. There's

there's a lot of need for automation and continued

automation in space to make sure that the cost of data collection

is as low as it needs to be for a technology like this to really

scale and make sense for our customers. Does that answer your

question, Mike?

100%. Yeah, because you know, labor, whether it's

some merchandiser, or brand owner or retailer, you really

want those folks actually taking action on the output of your

solution and not being a data collection device, if at all

possible. So I believe, I don't know about the drone thing. I

don't know if I want to get buzzed in the store with a drone

taking pictures, but I get your point. And I've never seen a dog

with a camera on it, that could happen, I suppose. But to me,

certainly seeing shelf scanning robots that are very prevalent

on the industry, starting to see fixed cameras, starting to see

more and more of that technology. And then okay, I got

all these pictures now what do I do with them, which is was part

of your pieces. Renish, question for you. So how does this

processing occur? Is it like, feed, feed all these pictures

into something and you guys crank on it for over the

weekend, and on Monday morning, you give me the results. So you

know what is, what is actually the processing behind the scenes

look like in this thing?

Yeah. Thanks, Mike. Again, let me take this in

two parts, right, like, where does processing happen, right,

and today, I think like others mentioned, like these pictures

are being captured by different types of endpoints, whether it's

a device, whether it's reps phone, whether it's shelf edge

camera or shelf scanning. And with the recent advances in this

in the compute capability of these devices, what we are

seeing is, technology is able to do a lot of these, the heavy

lifting on the device, which helps us avoid network transfer

cost and giving faster results, right. So we see that the

processing itself is moving more and more closer to where the

action is happening. And when you look at how does it happen,

these images are being processed by computer vision technology,

where you're trying to extract features of what we see in those

images and then compare against what is expected, right. And

that helps give actionable information whether it is

comparing against a distribution list expected or whether it's a

planogram, where, for example, if you have prospects filed for

a particular shelf, you can compare against that. In certain

cases for a supplier you're trying to measure against their

contracts for a contract compliance. A supplier has a

particular contract with a particular chain, looking for

compliance against that. So there are different aspects of

shelf execution that is being compared against with the data

that was extracted from the majors. So that's, that's the

processing part, right.

Got it. So let me let me ask you guys, I want to

go back to a slide that I shared before. It's this one here that

basically kind of lays out the different challenges and there's

probably more I mean, just to be clear, there's probably more,

but as you look at this thing, what kind of data and images and

information do you need to basically tell a retailer or

brand owner I see a label, I see no product, there's obviously an

on shelf availability issue. What data do you practically

need to be able to do that? I'll start with you Anand. What data

do we need for your algorithm to be able to tell us that?

Yeah, I mean, I think the basic information is

the master data from the brand or, or the product category,

right? So what are the products that need to be recognized

because AI is in a state where it needs to be still taught

right to understand what product is what so there is some kind of

data that is required. Of course, images from the shelf,

which is basically the images that we capture from the stores

with just these two set of data, algorithms can be customized,

and they can be as accurate as, as ground truth, right? Like,

like a human would see it. These are the two set of information.

And as we go, work with more and more customers, we see that the

amount of data that is needed from the customers is, is not a

lot, right, a set of pictures from the stores, as well as the

master data from the customer, is typically what is needed to

customize algorithms which are very accurate. And that can

digitize these visual images from the stores and then

generate data out of it.

So you need a picture of the shelf, and you

need a really good quality product database that you can

match it up against and go yeah, that's a problem. That what I'm

hear you saying? Alright, so if we expand that, beyond this, and

I'm going to put Matt on the spot now. Pricing, this one is

very important. There are some places in places like New York

and New Jersey, if there is not a match of compliance, retailers

can get fined a lot of money because the price at the

register and the price at the shelf doesn't match. So in

addition to what what what Anand just mentioned, what else do you

need to be able to be able to say there's a pricing discrepancy?

Yeah. So pricing is is probably one of the most, I

don't know if the group agrees with me here. But from my

perspective, one of the most misunderstood components of

image recognition. OCR technology has been around for a

long time, and one would think that you could use OCR

technology to basically validate or invalidate a price fairly

easily, and then correlate that back to what SKU is available or

is not. And that's, that's true to an extent. The reality is,

though, is that especially kind of because we're working with

customers that are in a lot of different places of the world,

and if you really look at pricing at some of the smaller

stores like bodegas, let's use your example of New York City.

There's a lot of handwritten prices, there's a lot of things

that are not easy to recognize, even for the human eye. So I'll

start by just saying that pricing, while the technology is

there, the real kind of special sauce is using the price,

getting it at scale, and then correlating it to the product

within the photo that it comes back. That really is a hard

thing to master, and it's something that when you start

looking at a lot of images, it becomes very real and apparent

that pricing is not as simple as you might think. So that's the

first point. The second point that I would say is that we we

use pixelation to basically determine where the price is and

where the product should be to then create almost a an

artificial link to what that hole actually is. And that's how

we would tell a customer what the hole on the shelf is, is

through basically correlating hole with price and then how

close in proximity it is. And then obviously, there are

processes around you know, quality assurance and cleansing

that makes sure that you know, that's working effectively. But

to answer your question, Mike, price is an enormously impactful

thing. And it tells a real story when things are priced correctly

or incorrectly that you know becomes a lot of value to a

brand and then also a retailer to see at scale.

Got it. Perfect. Anand, I'm gonna put you on the

spot, plugged item. Plugged item basically says there is a label,

there is a product, but your tool is going to tell me that

that product is not the right product for the shelf. And that

causes two problems. Number one, the products not on the shelf.

Number two, when the person comes to stock that product

cause it finally shows up, they go to stock and they can't find

out where it supposed to go because somebody put something

else in the way. So, what data do you need to be able to say

that item is a plugged or an incorrect item or sometimes they

call it they've expanded the facings of it. What do you need

to be able to accurately tell that to a retailer or a CPG

company?

Yeah, so that the image of the shelf from the

image of the shelf itself the pricing stickers can be decoded

like electrical. How Matt was mentioning using OCR technology.

Of course there are challenges because these stickers are very

small and handwritten sometimes in some of the stores these are

not very in standard format, right. But I think the two

inputs that are needed is the record recognizing the sticker,

as well as recognizing the product that is plugged. Right.

So and then match them together because the the sticker usually

has the SKU name as well as the price, right? And and the

product recognition will tell what is the exact product that

is placed there. And using both of these information, we can

identify whether it's a plugged item or not, right. And these

are the two sets of information. And again, with with computer

vision, just by scanning the shelf, these two informations

are extracted automatically with algorithms and then compare it

and then these alerts can be generated.

Got it. Perfect. All right. Angam on the last one is

probably the hardest one. So I figured I'd leave the last one

to you. So these yellow boxes, basically say the product is not

there nor is the label there. So there is supposed to be a label

there that is not there, it got skipped. In addition to the

picture of the shelf, what do you need from the retailer to be

able to say there is a distribution void, or there's a

missing label that should be there that's not?

No, I think there's not other input required

from the retailer, we need to the CV, computer vision

technology needs to be smart enough to identify gaps in the

images, right? To understand that, okay, there is supposed to

be a product here, there is supposed to be an SKU here, but

there's a big gap, the particular red one that you see

right out of stock box that you see it's the CV technology needs

to be smart enough to identify that gap. And also see that

there is no price label we need that right. The first step when

you're doing pricing is to segment each and every price tag

written right? Once you segment it, then you read it. So if

there is nothing to segment, you know that there is no price tag,

and ultimately, you tell that, okay, this is just an empty

space, there's supposed to be something and nobody knows what

was supposed to be there, right? So you can go to the planogram

image and try to figure out what could be here. But otherwise,

you know, you have to detect the empty space and the missing

label that there is no missing label to come up with this

information.

All right, so I'm gonna, I'm gonna put you guys on

the spot a little bit, and then just just to let the audience

know, this will be my last kind of question before I kind of

open it up for any questions you guys might have on this kind of

section. Every one of you have been mentioned product database,

in one way, shape, or form. You gotta have a product database

that you're matching it to. Talk to us about that because my

sense is each one of you are probably using your own product

database, you're expecting it from the retail or the CPG

owners. I would imagine that this is probably one of the

biggest challenges in your space, which is, brands spend

millions of dollars a year just changing packaging to make it

look like new and improved, right. So how in the world, and

you certainly get products, while this particular example is

very clean, you'll get things turned to the side, you'll get

things knocked down, you get all those other kinds of issues that

you have to deal with. Talk to us about the importance of the

product image library to be able to do this kind of work. And you

know, what's the roadmap in the future look like for product

product databases?

Me, I can go first.

Sure go ahead.

Yeah, so this is, as you rightly said, this is

a huge problem, right? To maintain the product library,

and almost to a certain extent some moving goalposts, right.

It's, you're never going to achieve that. Right? Every time

there's a new SKU, you can achieve it for a particular

region, the button, then you move to another market. And then

again, there are newer SKUs, right. So it's a it's a big

challenge. I think this to a certain extent, can be solved

using AI technology, how strong your AI technology is,

particularly how much data do you require when you want to

train your AI? Right? If you require thousands and thousands

of shelf photos, it's going to be a nightmare, right? If you

require only few tens of shelf photos, the problem becomes much

more manageable. But what you're always trying to do, you're

trying to manage the problem, right, unless somebody can form

a universal database, which can be applicable everywhere.

Another thing to do this, and this is something that we have

been doing with other clients as well, is to have very deep

integrations with the clients, right? So the moment there's a

new SKU, or a new packaging of a particular SKU, right, you

quickly get it within your database and you can start

training it right. So this is something which has been helping

us where we have really integrated our technology with

their, with their databases to ensure that we are we always

remain aware of what's the latest SKU is and ultimately

what you also need to do is really reduce your AI training

time so you cannot take weeks to train your AI because by then

the packaging will change again right. So you have to be on your

toes. It should happen within a matter of a few days, if not

hours, right and and then you know, you should start you

should start delivering a very high level of accuracy right on

day one because you know that's when the value, that's when the

client will get the value. Right?

Perfect.

Yeah, one point adding to what Angam

mentioned, new product launches as well as custom design changes

are definitely happening. And they happen at different rates

with different customers and also depends on the channel as

well, right. So integrating into the customer database or

understanding the launch calendars can give a heads up,

right. So that is one of the key ways we also try to understand

what is going to come right. And usually brands have a launch

calendar, like what are the new SKU is or is there going to be a

combo pack or some kind of an offer? Right, all these are

planned by the point or by the by the suppliers, right? Or even

by the retailers. So with that, and then very, very

sophisticated algorithms, then these can be handled in a better

way. Right. Another approach is, there are I mean, algorithms are

now smart enough to identify some of the new products that

are appearing on the shelf, which they have not been

trained, not trained for. Right. So let's say there are some 10

SKUs of head and shoulder here, and then the suddenly we see 11

to 12 SKU, algorithms can pick and say that this is something

new, right? So so we sometimes go proactively to the customer

and say, Hey, this is something that is these are some couple of

escapes that we see new in the the field, do you do you want to

add them to the recognition, right? And then they come back

and say, yeah, these are some new launches. And then yes, of

course, right, and then then we start tracking them. So there

are multiple ways in which new product launches, as well as

design changes can be handled. And definitely, the technology

is very sophisticated now to handle these pretty much with a

very, very fast response time.

I would agree with everything that's been said.

This is a huge, I don't really want to call it a pain point,

but I think it's something that if an image recognition CV

solution does not take it super seriously, then they'll fall

behind against technology and others that are one of the most

valid points about why having a updated image library and master

database is so important is because when you have the right

information at a SKU by SKU level, and you can recognize in

the photo exactly what that SKU is, you can then tie back very

interesting custom attributes to basically what will build the

end outcome or KPI for a customer. And in order to do

that, you have to tag it at a SKU level. In order to tag it

correctly at a SKU level, you need to have the right image to

train on. So exactly like what these guys said, I think the

solve right now is through creating you know more

autonomous processes within our individual companies that will

allow for, you know, a certain queue to be created of things

that need to be looked at and retrained as possibly a design

variant, and acting on that as quick as you humanly can. The

long term, you know, big utopia for this industry is an updated

relevant master syndicated database that you know, everyone

can look at as the source of truth, and that just doesn't

exist right now. So

Yeah, great point. I mean, it feels a little bit like

you guys are almost differentiating yourself based

upon the quality of your product recognition library, which is

great, I suppose. But honestly, that to me is not the secret

sauce, I'd much rather you'd be focusing on different ways to

look at the same problem and solve different problems. I

wonder if there was some kind of industry and industry standard

would be better. I'm gonna open it up to a guy by the name of

Mike Price, who always has really tough questions. This

Mike is easy on you guys, but Mike price, I'll give it to you.

He's gonna give it to you. And he's also running on European

time so he's probably a couple couple of hours away from

sleeping too. So Mike, I'm gonna open it up to you. Hopefully,

you'll be able to talk if I've given you the permission to

talk. So go ahead and unmute and ask your question for us.

Thanks, Mike. I'm sure I'm surprised you haven't

banned me from your sessions with the questions that I've

been asked over the

I keep I keep saying disinvite but it doesn't do it.

It just comes back and gives me more questions. So you're ok,

you always ask great questions, Mike.

Well, I'll speak to the guys that stay in front

because they've managed to disable me from all their

webinars. Anyway, moving on. Moving on. I never I'm never I

never get knocked off of this. Anyway, moving on. So my

question to you guys is, I've been involved with this or was

involved with this when I was at Unilever since 2014. There's a

lot of you in the space now I keep a list. And I've got 43

partners, as I would call you on my list across mobile, fixed

cameras and robotic solutions. So my question is, how the hell

do you stand out and what makes you different? What is your USP

because most CPGs go through a pitch of five or six potential

partners, and it's really difficult. So that's my

question. Thanks, Mike.

Yep, thank you, Mike. I told you I was gonna ask

you tough questions. Please do me a favor and don't

differentiate yourself via price because I don't want to get into

antitrust things. What is your what is your point of

differentiation? I think that's a very fair question. We'll

start, Renish, why don't you go ahead and kick it off.

Yeah, so I think, I think like, like we all

agree here, the ability to kind of get the training right in a

very cost effective ways, one of the biggest challenges in this

industry, right. So we talked about master data, about

products, master majors, right? So we differentiate ourselves

being being able to train up data through both stock images

that we, for example, what we see on Amazon product listing,

and real world shelf images, right. So it's a mix of both

stock images and real world images, that helps us to be very

up to date when it comes to being able to recognize products

through our models. And that also gives you to actually

differentiate yourselves when it comes to cost that you can offer

because the moment you can train up without too much of real

world images, which are costly to acquire, especially the long

tail part of it, like I think Angam was talking about it, you

go to any new market, any new part of the country, you are

going to get new products in the same category that you're

operating at. How do you minimize that and be able to

give a very cost effective solution. So that is one right?

Where we have built our platform to be able to use stock images,

along with real shelf images to make training faster, and cost

effective, right? The second is, how fast can your system do do

the recognition and then suggest the next best action, right? So

we, we again, differentiate ourselves on speed, right? For

some of the largest snack manufacturer in the world, we

audit about 300,000 stores, and a million fixtures are there and

today, we are able to get the results back to them in under 30

seconds for every fixture that they audit, right. So speed,

again, speed to speed to results, right? That is the

second part that where we differentiate. So those are, I

would say two pillars on which we differentiate in terms of

being fast to be able to train up and cost effective. And the

second is being very fast at the shelf to be able to give that

recommended next best action to the rep who's standing in front

of the shelf to fix the issue of the shelf.

Awesome. Anand, you want to go next? What

differentiates you?

Yeah, well I would have told the same thing

that what Renish would've said. I would I would assume that what

Matt, as well as Angam was also going to say the same thing. So

I think the key key here is, of course, think there are

different levers in which we try to stand out right, for example,

speed at which we can customize algorithms and then get things

rolling, right. As well as the next best action part, and there

are a lot of these key things that are there. But I think the

differentiating factor will be to what degree we are able to

do, right? And I think that is where. I mean if you ask feature

by feature, I think anybody who any one of us can build a

feature that is not there with any one of us, right? It's not a

major thing, but at what quality and what precision you are

giving it right. And then I think that is one. The other

interesting part I would add, and maybe others would also say

that that's the differentiation is on the operations part. It's

not just the image recognition capability alone, right? How do

we operationalize this? How do we scale this? And then how do

we make it into a product that the CPG stakeholders are able to

ingest into their regular systems, right? Because unless

we do this in a in a very seamless way, it becomes hard

for them to ingest the data, and then it just becomes a data

overload, right? How are we going to help them to ingest

this into their day to day operations, and then help them

take quick action. So these are some of the key things that we

focus on because we see that of course, image recognition, like

it was in the beginning is it's just a means to an end, right.

But that helps us to get things in a faster, accurate way. But

on top of that, definitely there are a lot of other aspects in

terms of operations, in terms of the analytics capabilities and

other things that differentiates us. Yeah.

Awesome. Great perspective, great perspective,

Angam, Parallel Dots, talk to us. What's the differentiator

for you?

Yeah, and you know, just to, as Mike Price

said, right, there are 43 companies in his radar building

the same thing and the market is crowded, right? So technology

too can only be differentiated to a certain extent, right? And

for the customers today, it's incredibly hard to differentiate

on the basis of technology for them every vendor is essentially

the same tech right? So while of course, we focus a lot on the

things that you know, Anand said and Renish said and on the tech

side, but to build on on what Anand said further right.

There's more to it than just a technology piece, right? We have

seen a lot of our clients really struggle to piece all these

things together when it comes to retail execution there different

technologies here, right? You're talking about computer vision,

you're talking about, you know, their field operations, you're

talking about their ethos data, you're talking about so many

different pieces here. Ultimately, everything should

combine together and result in something which helps the client

increase their sales or reduce the cost, right. It all boils

down to two things. And that's where, you know, our approach

has been, we have been trying to understand the requirement of

our client in a little better way, and trying to forge our

approach in a way which helps them get this benefit, right.

It's rather than just selling technology, now, we are trying

to sell sort of this, our our understanding as well, that we

understand the space, and we're gonna help you make these

changes in the way you do retail execution, and image recognition

is a part of it. So it's more of a consultative approach now that

we are effectively choosing to take that to our customers. So

that I think is turning out to be a small differentiator in

this crowded space for now.

Gotcha. All right. Last but not least, I know what

Matt's gonna say, I think. Matt, what's your differentiator?

Yeah, so I guess everybody is spot on with what

they're saying. And I think the question, Mike was a really,

really good one. You know, we know from from a Trax

perspective, we know and accept that IR image recognition,

computer vision, is a known commodity, and is very much

advanced in the last 10 or 12 years. It wasn't always this

way, but it certainly is now. And I mean, all you have to do

is talk about the 43 competitors to tell you that it's clearly

there. And there's there's a very saturated market. So what

we from a company perspective, have aimed to do is to not think

of ourselves just as a computer vision provider, but as an end

to end tech enabled solution that drives and aims to drive an

enormous ROI for our customers using our service. And what does

that mean? It means that we don't want to just tell you

what's on the shelf or not on the shelf. And we don't even

want to just tell you what you should do based on what's on the

shelf, we want to make sure that you have the ability to leverage

a one stop shop to understand what's going on in the market,

in the stores on the shelf, and then also close the gaps as you

need to. Merchandising as a whole is extremely ROI driven

and any change to a service or a service model with merchandisers

needs to come with a very accelerated return. So said

differently, if there is someone that wants to change their in

store execution practice, they need to prove to their own

company that they're going to drive an increase in sales lift.

And we believe that our ability to recognize a gap on the shelf,

and then close it in a very cost, efficient, expedited

manner with our labor resources and merchandising resources is

our true differentiator. We also I would say, have a very global

presence. So it's a pretty compelling conversation for

someone at a brand's headquarters to look across 50

different markets at once and measure kind of metro size how

their perfect store looks in execution and at scale. And

that's something that, you know, we've seen a lot of success with

some of our biggest customers where they can basically point

and click across different geographies and tell a very

different story than than they have before. And then, you know,

last but not least, we do have a couple of very tailored packages

based off of our 12, 12 and a half, 13 years in the industry,

where we know we've taken a lot of best practices from customers

that keep telling us the same thing in different ways about

what it is that they want to see in store, and how they can

execute differently or what they need, and we've built packages

and solutions around that. Where we think it's kind of leading

with industry expertise, because a lot of people are saying the

same thing. So

Great, awesome. Y'all did well, and you did it

without fighting. That's awesome. That's exactly what

we're hoping for. Appreciate that. I've got

Great panelists.

Yeah, I got two more questions. And I'm going to open

it up for any other questions that people have, because we've

got about seven minutes left. The first question is kind of a

two parter. There's a lot we've had a lot of podcasts on here

from people like crowdsourcing last time, we had Trax and we

had Field Agent. We had companies that are responsible

for fixed cameras like Focal, SES-Imagotag and Invue, and we

have shelf scanning robot companies on as well Badger

Simbe, Brain, Zippity, just to name a few. If you are going to

be talking to these solution providers, so no longer about

retailers and suppliers, but solution providers. Why should

they come to you and talk to you about your product recognition

solution rather than building their own because what I've seen

most of the time, is they build their own. What's the advantage

to and again, this is for all of you. What's the advantage to

using you guys versus building their own proprietary image

recognition software, whoever wants take that one first.

Yeah, I will take it. So when we do see these

solution providers as partners, we have in fact, partnered with

some of them, especially the crowdsourcing companies,

definitely crowds I mean, all of them bring they bring different

perspectives, different strengths, right. I mean, from

from a core computer vision perspective, again, a focus for

our company has been on fine tuning this technology to a

very, very sharp way, that way, image recognition and the data

digitization is done in a very accurate way, right. So, we have

seen, from a technology perspective, our algorithms are

a lot more superior, at the same time, the other players are

bringing in other strengths, right, for example, crowd

sourcing company brings in the field force a cheaper ways of

the conflict case, right. And that is extremely important.

And, and if we are partnering with them, and then it the

overall solution becomes a lot more comprehensive, and then

valuable, right. So that way, we have seen, we do see these as

definitely complementary partners, and an approach where

a combination of strengths work together, and then you better

value to the end customer is, is what what we have seen and

definitely, that's an approach that we're taking.

Yeah. Awesome. Anybody else?

Our company vision is very holistic, and not

dissimilar to what was just said, you know, image

recognition is one really key piece to that and it's kind of

the backbone of everything. But if you were to look back on our

company acquisition strategy, we have acquired and partnered with

many companies, both from crowdsourcing to robots to

shopper activation. And I think there's a certain culture within

our company that acknowledges and knows that there are certain

players that are always going to be better than us in certain

things. And our goal is to be the one stop shop for retail,

not just for image recognition. And that lends itself very well

to partnership opportunities and overall kind of growth within

That's great. Thanks. Last question. What's

the industry.

the roadmap look like? Where are we going to be two to five years

from now, I'm gonna say in store recognition, I'm not gonna say

product anymore because I think there's other opportunities

besides just product. What's the next two to five years look like

in terms of in store, you know, recognition of what's going on

at retail? Each one of you got about a minute or so to close

it, and then we'll open it up for any final questions.

Maybe I can go here and share my thoughts. I

think next five years are going to be crazy and super exciting

as far as the technology is going to concern right? With the

amount of innovation and we can always look at what the past

five years have been in the trajectory that we have been on

in the past five years, right? Not just we have, the underlying

technology has become far stronger. We have now large

scale case studies dealing with helping thousands of SKUs at

scale, doing analysis on in, in within seconds, right, detecting

them in seconds. All of this was probably not possible five years

back. And what I mean by this is that five years later, I think

that kind of things that we're talking about are only going to

get more bizarre in terms of the final output. So for example,

you know, right now, we used to talk about image recognition

through photos, and we have done it on device, right. I think the

next step is going to be videos, right? Where we can do, we'd be

processing videos, possibly videos on device as well, with

very, very high accuracy. Maybe more KPIs on device itself, you

know, some of the more difficult KPIs can also be done on device.

All I also think that, you know, a different pieces of

technology, as Anand said, would come together in the next five

years, right. You will see a more holistic solution either

beat by one vendor or beat by one supplier, or multiple

suppliers coming in together, and sort of attacking this

problem end to end, resulting in a much see more seamless

solution than what we have today, right for the for the end

clients. So I think those are the few things which are

definitely happening in the industry and definitely, you

know, something which via suppliers need to be need to

know as well as the clients need to know what's what's happening

next in the industry.

Great. Anybody else want to take a shot?

Yeah, I'll add one more point. I think this

is primarily around image recognition in very low light

and very complex scenarios. So a lot of the recognition a lot of

companies have worked in and developed market where there is

nice arrangement of shelf, larger stores, good lighting,

right. But I mean, as all of us know, the complex traditional

trade and the developing markets is very, very large, right? And

image recognition has not been playing a major role probably

for the past few years, but in last few years, we have seen

like the technology evolving, especially recognizing products

in very low light conditions and very products that that are not

arranged well, products that are inverted, right? All sorts of

complexities that come in, when when you go into smaller stores,

traditioned rates stores, developing market, right, and it

opens up a big, very big opportunity, because the number

of stores that are huge. The value for the brands are also

huge, because these are the areas where there are growth

opportunities for brands as well, right. So I think image

recognition is improving there. Definitely huge opportunity to

digitize not just the large supermarkets and hypermarkets

but across the globe, any type of stores, any type of

conditions in which the store environments are present.

I think from a macro perspective, retail is

going to see a lot smarter demand forecasting and

replenishment systems, there's a lot of really smart companies

out there that use AI to really calibrate and get better at what

products should they be ordering, and you know, putting

in the back room and then putting on shelf and we can

reduce overall weight, shrink, etc. So I think with that,

there's going to be a very interesting dynamic from, you

know, there's shelf holes everywhere during the pandemic

to we might get better and better at closing those shelf

holes, because they're getting smarter with their ordering. I

think everyone that said there's a rapid technology advance in

image recognition at a lower cost is spot on that's happening

in front of our eyes right now. We'll continue, I would say to

advance in autonomous data collection methods. So less

people and more either robot or fixed hardware. And then I think

the last thing that's probably going to happen here is around

predictive analytics and we're going to see them really

increase in terms of getting really smart and calculated

about using a lot of different variables to do the next best

thing in store with the data that you already have. And I

think that that opportunity is just huge to kind of take one

big leap from where we are today.

Yeah, this is Renish here again, I agree with

what all my fellow panelists had to say, including the evolution

of technology, which is gaining pace, lot more data available

now, which makes it more actionable. One big, interesting

thing that that we see happening is a lot of suppliers are, are

able to move to what what we call as a store precision

planogramming, or being able to kind of plan for every single

shelf, in every single store, in every market that you operate

in. Today, a lot of suppliers use common planograms across

formats or in a large region, but we see that some of our

existing customers are at a position with the amount of data

that we are able to get them to be able to drive towards a

planogram for every single store that they operate so like highly

customized planograms based on the demographics based on the

shelf based on their supply chain. So we are able to bring

together a lot of data to be able to offer the ability to

kind of plan for every single shelf at every single store. So

and obviously other advancements like what Mike was, Matt was

talking about in terms of predicting in terms of more

automation in data capture, more automation in even in the model

training, right. So today, AI model training is probably the

is a heavy lift right now, how do we bring in more automation

there, make that faster, make that more responsive? That's

where we see this going.

Awesome. Awesome. Did I get everybody's last kind

of vision, I think I've covered everybody, Angam did you respond

on to the five year vision.

I did actually.

Okay, that's what I thought. You were the only one I

wasn't sure about so. Well, we're gonna go ahead and wrap it

up. We don't have any additional questions from any of the

participants. I just want to thank each one of you for taking

time, in some cases, early Saturday morning. Anand, thank

you very much, go to sleep and you can wake up on Tuesday, if

you want to. Appreciate you guys doing this. This is the future.

I really believe this and I do believe automated data captures

cameras, fixed cameras, robotics, but that that how do

you take that information and turn it into insights so Matt,

to your point, somebody can actually do something about it,

I think is the future and I think I think the more we

leverage this technology to make sure products are on the shelf,

I think the better off we're gonna be. I thank each one of

you for your time. Continued best of luck of your in your

business and keep on developing, I think I think the future is

very, very bright in front of all of you. So thank you very

much. Have a great weekend and we'll talk to you later, bye

bye.

Thank you.

Thanks guys.

Well hope you enjoyed that podcast on product

recognition software. Please feel free to reach out to any of

those suppliers if you have an interest or follow up

standpoint. Next time we are back talking about RFID and the

specific topic is the ID in RFID. What companies are taking

advantage of that serialized item level whether it's in an

RFID tag or a 2D barcode. The ability to uniquely identify

every single selling unit is a key unlock for the future of

technology in retail. Please join us then.