Well, hi everybody and welcome. I'm Peter Curran, Chief Product Officer at Coveo and I'm joined today by Ray Iyer, one of the founders of Anglera. Hey Ray. Great to be here Peter. Thanks. So we have some moderators. If anybody has questions, please feel free to ask questions as we go here. But Ray, thought we'd start with a little bit of the founding story. Tell me a little bit about how you guys, got started at Anglera. Yeah, it's a great question. You wonder how two two people like us end up deep deep in the weeds of product data and enrichment. Well, my cofounder and I actually met each other, because we were working on the technology team at Uber Eats. And Uber Eats was making a big push, they still are, into grocery. And so we were looking at the top 10 search queries that customers were coming into the app with. And one of them was, as you can expect, milk, right? Bananas. Simple things, one word. One word, simple groceries, right? Staples that people want delivered. And then we started to notice organic milk, gluten free snacks, high protein snacks. And these are things that we would always be monitoring on the tech team because you wanna figure out where users are getting tripped up. Right. And we started to realize a lot of the time users would type in organic milk and what do you think they'd get back? Not organic milk. Not organic milk. And how do you think that makes our customers and users feel? Not good. Not good. And when we started to dig into the root of this, we realized the problem wasn't our search, our ranking, it was the product data. It was, we simply didn't know if this milk was organic or not. Maybe if we did, it was deep in some description field somewhere that wasn't getting pulled out. And this is something we ran into in our jobs and and and then we started thinking, this must not be a problem specific to Uber Eats. In fact, it must not be a problem specific to just grocery. It must be specific to any product that is sold anywhere. Yeah. So then fast forward a couple of years, you guys work on this and, and what did you learn? And what we learned was very much so. This is this is this is a problem almost universal to anyone in the space of retail, of distribution, especially people that end up selling products that they don't create or manufacture themselves. Folks with very, very large product catalogs. Maybe folks like some of the folks that are joining us on the call today. And you and you realize... What we realized was was not just that this was a long standing problem, but actually the tools and technologies that that, retailers were using to attack this previously just hadn't cut it. No. And I've been in this space for a long time, a lot longer than you, but we've both been in a lot of rooms. Nobody has this problem solved. It's universal, right? Yeah. Who who have you met in your... In the last, ten, twenty years that says I have a perfect product catalog? Zero. Nobody. They all hate their product data. Yep. And it had been one of the things that people tolerated, right? There were tools that attacked around it. Maybe they helped move things around a little bit easier, right? Maybe they helped management get a little bit easier, but it was all still manual, it was all still difficult, it was all still messy, Right? And that's what really led us to build Anglera because we saw emerging capabilities of AI getting better and better and better over the last few years, and we're like, there must be there must be a better solution to this problem. Yeah. So we've seen the same thing. Let's look at an example of what this might look like. What's the product data problem? And the problem is what you see on the screen here. Right? For everyone on this call that's been in this industry for a while, you know it. Right? The product data problem manifests really in three ways. First, product data can be incomplete. Right? Which means you're missing key information that buyers need to make their purchase decisions. I don't know if it's organic. Yeah. The second is it could be incorrect, is in some ways even worse. It's even worse. Yeah. It's embarrassing, right? They call it organic, but it's not. But it's not. Yeah. Imagine how a buyer feels, the Whole Foods buyer would feel if they order organic bananas and they get normal bananas. They get normal bananas. Not good. That's worse. That loses trust. Right? So incorrect might in some ways be worse than incomplete. Now the third one, inconsistent. You might think that that's the least important, but don't know. Mean, take take this water bottle. Right? For example. If I ask you, what is the size of this water bottle? I know it's eight ounces. Could be eight ounces? Yeah. It 500 milliliters? Could be. Could be small. Small? Yeah. Medium? Yeah. Reusable? You know, everything in the kitchen sink. Yeah. Right? You could get any answer. You could ask 10 people what the size of this water bottle is. You'll get 10 different answers. All 10 might be correct, but they're not consistent with each other. And that's why consistency is is is so is also part of the problem. And you know what makes it worse is you get a 100 brands that make water bottles and they all call it different sizes. Something different, yeah. And that's what retailers, distributors deal with. And then you have a 100 buyers typing in searches on your website that all are expecting to search for a different size Yep. By the way they think about it. And so then that, you could imagine how that blows things up. Yeah. There's an infinite number of ways to ask for a small subset of things, right? Cool. Well, let's look at a demo and see how you guys have solved this problem we've identified. So, let's get more concrete. Let's take an example of a batch of fridges, right? Let's say that we have a batch of refrigerators that we're trying to launch and we need to collect complete attribute data about these refrigerators so that we can actually put them on our digital website. On your screen here you see an example of the starting point, right? A lot of you probably have come to know and love, maybe hate this ERP style of writing where maybe you receive just very limited information about the products that you're trying to prepare. You have no image, you have no documentation, you have no manuals, and you might have a 140 character title that's abbreviated that tells you basic information about what this refrigerator is and maybe who the manufacturer is. And what I want to show you in the course of this demo is how we define what attributes we're going to collect, how we define our sourcing, right? Where we're to pull data from, what sources we trust, and what sources we do not trust, more importantly, And how we're gonna use those unstructured data sources that are out there that are trusted to then populate and fill in all the attributes that we would ideally like to collect. Right? Which are like product titles, product descriptions, even FAQs that are targeted and tailored to that consumer. And all the structured attributes that this consumer is gonna use to make a purchase decision on a refrigerator. Right? And so right now you're seeing the before state. And so, next what I wanna show you is what attributes are we actually gonna end up collecting? And that's actually really flexible in Anglera. And so, you can actually define any attribute which would typically correspond to your taxonomy. And you can apply any naming convention or guideline as a process to be able to define what good data looks like. And so that's really step one is is what does good look like? What do I want to collect for every refrigerator? Super flexible. Usually this is gonna come in from your PIM or your ERP. Right? Think about the structured fields and columns that you're collecting. That's gonna be replicated one to one in Anglera here. And it spans not just structured attributes, like depth style and installation type, but it also spans things like imagery, PDF documentation which we can clean up and source. And what's really interesting here is that we tackle a really important consistency problem like we were talking about with the water bottle. Because now when we talk about something like total capacity, what we can do is we can actually say total capacity is defined in cubic feet always. So whether or not I have a 100 manufacturers, some say c u f t, some spell out cubic feet completely, some leave the unit out altogether and you have to go reference another piece of documentation to figure it out. We can always normalize it and make everything very consistent. And so these are the attributes that we end up collecting and every single one, it can be defined with a unit or, for example, it can be defined as a set of selectable values where you have a very specific set of values that correspond to how buyers are maybe searching or faceting on these products, right? So when it comes to insulation type, there's really only three possibilities. Is it freestanding? Is it built in? Is it under counter? Etcetera, etcetera. This is how you define what good looks like because this is how we're gonna determine what comes out of Anglera. Right? The output side. And so now that we've actually set up the attributes, wanna talk about where we pull data from. And so, generally, what we're able to do is pull information from brand and manufacturer websites directly. And so what we're doing is we're searching by model number and brand. Right? And so in this case, we're actually looking at Frigidaire and Frigidaire's own website to actually pull this information from. And this is how you know you're not pulling data from a location that you don't trust. So And this is something that's very important to us. Right? As we'll talk about, knowing exactly where you're pulling information from, knowing that the Frigidaire information and specs is coming directly from a validated source, and LG is coming from LG, Samsung is coming from Samsung, this is very important to us. And so once we set up all of this web search and set it up so that we know exactly where we're pulling from, we know that we're gonna be extracting information from PDFs, from websites, from imagery to be able to populate the rest of the empty information here. And so now, I'm gonna move on to my favorite part which is where we're actually gonna trigger this enrichment. And so to trigger the enrichment process, all I need to do is select a certain amount of rows and click the enrich button. And so now, these products are gonna get queued up to get enriched in parallel. And so this is really important because even if it takes, you know, a minute or two per product, we can process hundreds if not thousands or tens of thousands of products in parallel. And this really allows you to get to that scale of product catalog enrichment which would be very, very difficult if not impossible to do manually. What you're gonna see in the next one or two minutes is coming back would typically take a person maybe thirty minutes per SKU. And now we're gonna get that back in the span of minutes. And so what's happening under the hood here is that Anglera is going out to a trusted data source like Frigidaire's website, pulling Frigidaire's imagery that's trusted and validated, as well as the PDF documentation, which includes spec sheets, user manuals, line diagrams, and installation guides. Things that are really important for consumers to be able to understand whether this is the right fridge that fits in their right opening with the right installation and tools that they have on hand. And you're seeing all of that information get crunched and processed and put into your standardized fields here in real time. And so you can see that we're pulling the specification sheet and not just pulling the link to the specification sheet, but using that information to populate all the rest of these attributes. And these are the attributes that you know and love. Right? These are the attributes that are gonna power your search and discovery. What is the... Is it fingerprint resistant? What is the depth style? What's what's the defrost type? Total capacity. And you'll see that this solves the consistency and completeness problem because every single value here in this column is is very clean because it's applying the rules and the governance that we set up in the attribute screen. So regardless of how that supplier or a 100 different suppliers describe the product, it's all gonna get put into a single clean consistent column here. But now you might be wondering, okay, we filled out all this data. So where is that data actually coming from? How can we trust it? And so that's really important to us too. So you'll notice we've pulled all this information. We've pulled the spec sheet. Right? We've pulled the instruction manual. And if I show you an example of what the output looks like, you'll see that we've now written a description. We've now written key features. And now we have these rich FAQs for how your buyers are actually trying to discover these refrigerators which were never present on your page before. Can I install this refrigerator in a garage or utility room? How much space should I leave around it? Does it need a waterline? How much energy does it use per year? These are questions that if someone had come to ChatGPT or come to your own website and asked, they would have never found an answer. But it goes beyond that. Because when it comes to this question, does it need a waterline? That's a very, important question. And if you get that wrong and the buyer receives a fridge at their house and it's the incorrect fridge that doesn't fit, then you're gonna be on the hook to pay for that return. And this is where providence comes in. Because when I... I can interrogate any representation or fact that Anglera returns about a product and I can tell you exactly where that information came from. And so in this case, does it need a water line? This actually came from the owner's guide, the water supply connection section where it explicitly requires access to a cold water line, I can tell you with very high confidence that this came from page 13 of that manual. And this is where Anglero's provenance and citation capabilities come in and you'll be able to see that for every single fact that we fill in about every single product. So when it comes to citations, knowing where your data comes from, when it comes to a $2,000 fridge that's hundreds of pounds and costs maybe hundreds of dollars to ship to your door, knowing where that information came from is really important. And so knowing does this refrigerator need a waterline, we can cite that back to exactly where that information comes from. But data is not always this clean, and we acknowledge that at Anglera, and that's why we route really important source conflicts to humans to monitor and review and resolve. And so what Anglera does is when it detects a value conflict, it leaves a note. It leaves a note for a human to actually review and triage. And so in this case, it's actually left a note here for one of these refrigerators. And it'll leave an explanation saying the source data actually conflicts because the depth with the door open at 90 degrees, the spec sheet table says that should be 43 and a half inches, but the installation diagram actually shows forty two and six tenths inches. And again, this is really important to get right because some people are choosing and buying specifically on the depth with the door open at 90 degrees. And so if I actually go in, I can see exactly what attribute is impacted and I can see exactly what value was chosen and why. And this is something that you have complete control over as the customer. Because some of our customers will say, hey, we should always trust the installation guide, right, over any other source. We should always trust the PDF over the website. And so in this case, what's happening is 43 and a half inches has been chosen because that's coming from the official PDF. And so the installation diagram has been cited as a secondary source with a different value conflict, but the confidence score is low. And we will always leave a note when we detect two different values to be able to determine where human review is most necessary and adds the most value so you can make sure that the data that you're putting in front of your customers is trusted and accurate. And so that's a quick tour of Anglera. At this point, you can actually export this. And the cool part is you can send it to your PIM. You can send it to your ERP. You can send this as a feed to ChatGPT. You can send this to Google Merchant Center. You can send this enriched product data to every single consumer and source that needs it to power discovery, ads, and other use cases both on your site, but also off your site. And that's as easy as exporting this as a CSV or an Excel file, or you can export it using an API to send it directly where it needs to go. Awesome, Relevance for that demo. Let's talk a little bit about how Coveo and Anglera ended up together. So we obviously saw this same problem that you guys have solved and we saw emerging number of companies in the market who have built out this kind of capability, and we knew we had to have it. So we started a first principle search looking for, you know, what are all the different companies that are out there and how are they different? Because we know that if we make better product data, we'll have a little bit easier time handling more difficult searches and helping people find the things that they want to find. So what we learned as we looked in the market was that there's really two different kinds of companies. Most of them are only focused on cosmetic enrichment. In other words, they're looking at photography and they're only looking at that photography in order to enrich the record. That's not only what you do that, but that's not only what you do. So tell us a little bit about the difference between what you guys do and what everybody else does. Yeah. A 100%. And I think, yeah, cosmetic enrichment is key to what we do too. It's looking at the image and adding a bunch of stuff, right, to to how you describe that product. Which is important because of the 100 buyers and the 100 different ways that they're expecting to search for dishwashers. But I think what really sets us apart, right, is is is knowing where information came from. And also, looking for every possible place where that information and data could live. And I think these are two very key aspects of of what we do. Knowing where information came from is particularly important in B2B. Right? As as as we can imagine. Especially in B2B. Right? Because, if that thing needs to fit a particular machine or it has to have a certain kind of specification in order for us to be able to use it in that application. Like those details really really matter in B2B. Yep. And if I show you a picture of a dishwasher here or you know, a picture of a glove, you know, and I wanna know if it's fire resistant or cut resistant or what the rating is. Yep. These are really important factors for Yeah. For safety and for for well-being. And you need to know not only that it's trustworthy Right? For the buyer to know they're getting the right thing, but also know that that information could be on page 15 of a of a PDF somewhere. Yeah. And that you're able to cite it. You know where the PDF, have provenance back to where the information Where that information came from. And that's true across any type of product you might be looking at. So Coveo and Anglero working together means that customers are going to have their source data that you're going to enrich. You put that back in the PIM so it doesn't just go straight to Coveo where we hang on to it, but goes back into the PIM. You resolve as we saw in the demo all of the conflicts because the PDF and the image may sort of tell a conflicting story for example. Or you may have two different spec sheets that both say different things. So you resolve all of that, put it back into the PIM. Now you have a clean record, but then that also goes into Coveo where it helps people do searches. I think the more interesting thing for me is the business value. Like does this make people more money? So tell us a little bit about, what you guys observed at a at a big retailer. Yeah. And this was an incredible experiment that we ran over the past year with a retailer that we personally know and love. And, it really boils down to, I would say, two things. Right? So TVs were what we experimented on and TVs are a great example, right? Because TVs can be thousands of dollars. You know, I don't know about you but I could take a month or two to make a decision. Yeah, it's a deal, right? Like, It's don't centerpiece of your living room. Yeah, totally. And it's really important. And, you know, there's so many nuances to how you might want to discover it. And so on the left what you see there, right, is more shoppers end up finding the perfect TV. Because manufacturers put so much time into every nuance and every attribute, and you wanna structure that in a way that if I ask, you know, an 85 inch TV that looks like wall art, looks perfect in my living room, you end up driving more visits to your product pages because the data is structured in the right way. But that's only one half of the story. Because, you know, the reality is more eyeballs doesn't necessarily mean more sales in every in every context. Right? And so that's the other half of the story is we're getting more eyeballs. We're getting more people to that right product. But more of them end up buying. And the reason is because we're matching so much more closely to that search query, the product to that search query. And we're targeting users and buyers, for example, that use faceted search and filters to find exactly and precisely what they want, that when you do get them to that page, they're more likely to convert because that product is the perfect match for what they want in their living room. And you you bring that together and it almost compounds in a way. Yeah. It's a compounding advantage. So people are finding the products on traditional search surfaces like, you know, they're Googling for it. Or they're in Claude or ChatGPT and they're asking about what is the best wall art style television. They find their way into the site and they they they actually end up buying it because the things they learned about in that AI surface or in in Google actually turn out to be true when they get on-site and and it's time to buy. Yeah. Yeah. So it's a compounding advantage. But what's cooler too is that in the future with Coveo and Anglera, we see this being a closed loop. So buyers are going to do searches on-site and conversations because we can do multi turn conversational experiences. Coveo is gonna learn from that but then we're going to pass that information to Anglera. Anglera is going to use that to further enrich the catalog. So let's help people understand. Let's give them an example of a case where we would observe something in search behavior that would result in another improvement to the catalog. Yeah. And this happens every single day. I mean, about social media and trends. You see your favorite celebrity wear a red dress with stripes at a football game, and you're gonna go to your favorite retailer and start searching for that product, but you're not gonna say red striped dress. You're gonna say Taylor Swift dress. Today, Taylor Swift dress, Chiefs. Always Taylor Swift. It might always be that. It might it might be Chiefs Taylor Swift dress. And and and these are signals that... And these are very, very valuable signals that your buyers are feeding you, and that's step one. And what we realized at Anglera, why we realized it's such a perfect fit, is that we we were blind to what users were were sharing on the front door of your website, and now we can bring that in. Right. Yeah. Bring it in and then create this continuous improvement. So it's not about just enriching that product once and you're done. It's an evergreen process of continual enrichment that's tied to signals that we're getting from consumers and and B2B buyers. Yeah. This is the dream for a digital merchandiser, having something like This is responsiveness. This is this is what we want. So, hey, this was awesome, Ray. Thanks for taking time. If anybody on the line is a customer and you wanna learn more, please feel free to reach out to your account manager. Or if you're not a customer yet, come to our website and we'll get you a demo and show you more. Thanks for coming today.
De meilleures données produit, une découverte plus intelligente : ce qu’Anglera et Coveo rendent possible ensemble


