Abhishek Vora: We are the technology that actually extracts and generates this oil or data which can then be provided to other sort of engines. We realize, hang on, this is a better problem to be fixing so we pivoted the business. The need for data will only expand and I think how you then leverage and use that data will become quite critical.
[Voiceover]: Welcome to ADD to cart, Australia's leading e commerce podcast that express delivers all you need to know in the fast moving world of online retail. Every week, Nathan Bush from E Suite and an E commerce industry expert will share the news, research and insights that you need to know to keep you at the top of your game and of course, keep your customers Adding to Cart.
Nathan Bush: Hello and welcome to Add to Cart. My name is Nathan Bush, host of Add to Cart and director at e commerce talent agency eSweet. Now put your hand up if you actually enjoy writing product descriptions and those pesky product tags. Yeah, look, there's not. This is a visual medium so I really don't know, but I'm assuming there
Nathan Bush: aren't many hands going up.
Nathan Bush: Today's guest is looking to automate product
Nathan Bush: tagging and descriptions so so that you
Nathan Bush: never have to do them again. Abhishek vora is the CEO and the founder of AI platform Ocular. Founded in New Zealand in 2019, Ocular helps retailers by taking images of products and automatically generating the product tags that can be used in categorization, filters and sorting, SEO and personalization. They're even working towards creating automated product descriptions so you never have to write them again. Already they they have leading retailers on board including Blue Bungalow, Maya and Princess Polly. Today, Abby talks us through the nuts and the bolts of how Ocular generates those product attributes. He gives examples of how other retailers are using them and explores the way that AI will be able to speed up some of the manual tasks that we dread in E Commerce. So thanks to our partners Shopify plus and and Paclio, here's our conversation with Abby Vora, CEO and founder of Ocular. Abby, welcome to ADD to Cart.
Abhishek Vora: Thanks Nathan. Thank you so much for having me, mate.
Nathan Bush: I've been looking forward to this conversation. We've been talking about it for a little while now, but we were held up because you were recently at Shop Talk a little bit earlier in this year, which is very exciting. To getting back to international events.
Nathan Bush: How was that trip and did you come back with any learning from the
Nathan Bush: US market that surprised you?
Abhishek Vora: Yeah, that was phenomenal. I think the things that caught me were definitely the size of it and the number of people who were There There were about 14,000 people at the Shop Talk Vegas event and a lot of it was driven from. They were all stuck at home because of COVID and now they'd had a chance to get out. So we had people from all over the US and, and the way they'd organize it was really, really amazing. I think I really enjoyed the, the energy over there. There were a lot of discussion about reaching us, saying, look, we've not been able to do a lot of these things for the last two years. What do we do now to get forward? And they were not looking for just sort of quick fixes. I think a lot of them were thinking, okay, now we realize that we need to really put in good effort into things. So a few things that stood out were learnings from China. There was a lot of discussion on what China has done in E Com and how globally people can learn around it. And that was one. The second was a lot of talk about personalization and understanding what does the customer want, how do you actually understand intent, how do you deal with providing it at the right time, et cetera. So those kind of things were very, very topical, but everyone was very, very engaged. I think that's the things that I took away from there.
Nathan Bush: Absolutely. And I feel on a very different scale. We had Retail Global a couple of months ago and yeah, felt that energy of everyone being back in the room. It was fantastic. You've obviously got Imedia coming up as well, so really great to be getting back to those in person events and hopefully I'll see you at one of them soon.
Nathan Bush: But we are here to talk ocular.
Nathan Bush: So you talked about personalization and intent, and I know ocular plays a role in that, but it is not all you do. Can you share with our listeners what ocular is and how it helps retailers?
Abhishek Vora: Yeah, absolutely. And the best way to describe it would be. So we are a retail technology company and if I give you maybe an example where, you know, there's a lot of talk about data is the new oil, right. So everyone refers to that. So if I use the same analogy, then we are the technology that actually extracts and generates this oil or data which can then be provided to other sort of engines which can then create better product discovery and better personalization, et cetera. So there are lots of sort of engines out there which require good data. And so we are the company that actually generates this underlying data that can be fed into these engines. So what sort of data is there? At the moment we are really concentrating on product data. So if you are an E comm retailer right now you are generating your underlying product data or product information. Majority of them are generating this manually. So that think of an E commerce retailer who's selling say apparel, fashion or furniture sort of products. So we have built a product which can automatically look at the image of the dress or the furniture item and it can generate the product attributes like it's a blue dress, it's you know, polka dot, it's flowing, it's a vneg. And we can now actually generate the product copy as well, the description as well, which again automates the process of something which is being manually performed. And that's what we feed into a lot of our partners that we work with like chemistry, like Silkspring et cetera, which can then use it to create more personalization, et cetera.
Nathan Bush: So does Oculus store the data or is it generating the data so they can be stored elsewhere?
Abhishek Vora: So we generate the data and it can be stored anywhere else. So one of the examples is so we would connect into the product information management systems could be like chemistry or in a Kenya and we would then as soon as the image has been posted into these portals we just need the actual image and from there we can generate all these different attributes and tags and the product copy which the retailer can actually audit it so they can go in and say no, this is not quite square neck, it's boat neck as an example. And with every correction because we using machine learning, it starts learning and getting better and the accuracy starts improving even more. Of course there is a base case accuracy which is there out of the box and then with further training it starts getting better and starts improving even more.
Nathan Bush: Ever scrolled through an E commerce packaging website for fun? Nah, me neither. Until today. Paclio is putting the joy into the packaging game. So let's play a game. I'll tell you the name of the Paclio product and you have to try and guess what kind of product they are. Fairy floss, Compostable mailer, Queen bee Honeycomb padded mailer.
Nathan Bush: Here we go.
Nathan Bush: Gummy shark water activated tape. Now if my jaded self thinks that this packaging is fun, imagine what your customers will think. Paclio is also eco friendly Australian owned and operated with same day dispatch and 14 day returns. Now that's pure joy for everyone. Check out the Paclio range of E commerce packaging options@paclio.com that's Paclio P A C K L E O paclio.com
Nathan Bush: is it as simple as a retailer just uploading a bunch of product Images to get started.
Abhishek Vora: Yeah. So right now what we do for some of the retailers is literally we take the product feed. And the product feed can have the image URL, it can have this QID and the category. So we need the category because sometimes if you have the entire image of the product, you know, say I'm wearing a jacket and a T shirt and pants or something, then our models need to understand what should we attribute? Are you selling the jacket or the T shirt inside or the pants? That's why we need category information, because then we can go in and do the attribution and the product description appropriately. But yeah, those are the three things. We need a simple feed that gets started and then we've provided our own sort of web portal which connects into this. And when as a retailer use join in, in the beginning, we actually go through what attributes are important to you. And also the taxonomy, because as you would know, you know, taxonomy is all over the place in this, in the industry. Someone calls it a. Someone calls the exact same thing a plus one. We help set that up so that if it's an error, it's an actual genuine error. It's not an error because of what we call it, something else. You can set the naming correctly in the beginning. You can also set up lots of things like synonyms. We're trying to generate automated synonyms so that when you add, say, polka dot, it automatically adds dots, spots, old cars, so that the additional data can then be leveraged by the search algorithms, et cetera, to generate better discovery.
Nathan Bush: And I want to come back to
Nathan Bush: how retailers are using this data because it would be a next layer of data that a lot of retailers just don't get around to adding on to their basic product data. But just before we do so, we're talking about. So do most retailers give it to you from a, like a Google product feed? Is that the main kind of feed that you get?
Abhishek Vora: Yeah, pretty much it's a Google feed or they can set up through APIs as well. So we've got both sort of abilities in there to sort of pick up the thing. But typically, yes, a simple feed is what we need.
Nathan Bush: And is there a time frame that the machine learning has to learn your brand or is it building, is it straight in and tagging straight away?
Abhishek Vora: So out of the box. For the type of retailers that we work with in, within the fashion vertical, which could be, you know, apparel, shoes, jewelry, etc. We've already sort of trained it out of the box. So if you have, you're connected to some of the companies that we're already got connectors into, say like a Shopify. Then you can be up and running in sort of couple of hours because we've already trained it. You don't need to do any specific training for that to get started. Of course there will be certain categories where you might need a bit more effort in there, which we can definitely cater to, but otherwise it is out of the box to get going from there.
Nathan Bush: That's incredible.
Nathan Bush: And you talked about writing your own descriptions. I knew you did the tagging, but
Nathan Bush: I didn't know you did the writing of descriptions as well. Is that a new feature or how have you developed that?
Abhishek Vora: Yeah, absolutely. So it is a new feature and we've sort of just recently launched it with the beta with some of our existing customers. And what we realized was that of course there is a whole manual work being done for the product tagging, etc, but writing descriptions is tedious process. You know, you don't want to hire creative talented people and make them write description. And sometimes you need to write, you know, five, six, ten different types of descriptions for the same product because you're selling it through different marketplaces. And each of those marketplaces are saying, well, we want a unique description so that it doesn't interfere on SEO and AdWords and those kind of things. So you literally have people sitting there writing the same thing ten times. Of course, they've also sort of democratized it in some way, but it's not the best use of their time. So what we did is we said, look, we obviously generating the tags and attributes, we can overlay that with understanding how we can write the entire copy of it. So we are now testing it out with our retailers and we've sort of come up with a few types of tones because obviously brands will have a specific flair they want to go for. And those are the things that we are testing out, saying, okay, what is the brand play that works, what should we add, what should we remove etc. From? Yeah.
Nathan Bush: Oh, that's very exciting. I can hear a few product managers and merchies like screaming with delight right now.
Abhishek Vora: That's the part. So we are saying, and sometimes when we think of automation, sometimes we think of 0 or 100%, right? We think, oh, it should be fully automated. So what we are saying is, look, this will already provide you a base case. You can use that to further edit it. And I think that's where technology is best used, where it has got it is doing your base case baseline amount of Effort. And then you overlay that with, instead of writing from the ground up, you're saying, okay, it's 70% there, 75% there. I'll edit it instead of doing the whole thing. And then it'll start learning and getting better with time. So I think that's what we try and do. So it's more sort of human and AI working together to get the output that you really want.
Nathan Bush: It's always better starting with populated fields and editing fields rather than starting with a whole bunch of blank fields that you've got to like, work every time. Now, you mentioned some of your clients and obviously I've had a look around and on your website you've got listed Blue Bungalow, more. Maya Kukai. Some brilliant Australian E commerce retailers in there. Can you share how your retailers are using this data, this attribute data, this enhanced attribute data and descriptions as well to enhance that customer experience?
Abhishek Vora: Yeah. Now, we've been very fortunate to have these amazing brands and we love working with these retailers because we learn every day on what else can we add into our product suite. So some of the ways that they're using it is, of course the two things that really help from an AI perspective is one, you end up saving time. So there's a clear sort of time arbitrage that happens. But also what we are seeing is retailers can then generate quite granular filters because now they have the extensive data set to, so they have X number of products and now you can sort of slice and dice it into different sort of attributes. And effectively that allows you to have more filters on your page. That data can then be fed into your search algorithms that can create product discovery better. It can be used by the merchandising team to understand, okay, what's coming next in terms of fashion trends. And if we need pink dresses with butterfly prints on it, how many do we have today and how many do we need to order going forward? Those are the kind of things that they are seeing. So it's increasing engagement as well, because a lot of the times because you have more filters, the users are able to really pick out those granular filters because ultimately when a user buys something, they buy based on certain attributes of the clothing item. So if you only have three, four filters on the page, it's hard for them to get to that unless they sort of scroll through all the products in there. So being able to discover the product is one big thing that they can
Nathan Bush: see, makes a lot of sense. And then with the filtering as well, does it help with consistency of tagging and labels. Because one of the places where filters go wrong is when you've got black spelled four different ways or with capitals, and all of a sudden there's a whole bunch of filter options there that really don't exist just because the data isn't quality. Does ocular help maintain that consistency?
Abhishek Vora: Yeah, exactly. I think so. What we do is we generate this underlying data and we are also seeing how best the retailers can use it in there. So they can actually come up and specify that these are the filters that we really want in there. Because some retailers we have seen have got a number of filters, but it's actually what is in those filters. So you can have the broad category, but you have enough products in there that have been categorized. So where we help with is categorization of those filters so that when a user goes to a specific filter, they don't just find one item in there. So that is definitely there. And also the search part, because ultimately if you don't have good underlying product information or good product data, the search algorithms cannot really pick up the right cues or the right words to showcase to the user. So many times, and we've done this test repeatedly, we have said, okay, if we do a formal blue dress search on a website, it sort of gives you white shirts, black pants, et cetera. So we're saying, clearly something is broken over there. And I think everyone agrees that having more data will be good. But then the question becomes, how much time and effort do we put in to get that fixed? And which is why conversion rates are really, really low. Because if someone looks for A and they get Z, of course they'll be a bit frustrated and they leave.
Nathan Bush: That's it. Especially in the fashion world where items might be in and out in three months time. And it's like, how much effort and manual time do you put into creating those descriptions and tags?
Abhishek Vora: Yeah, and also it's also the ROI on, because somehow the other people are spending money on Facebook, Google to bring traffic to the website. So that's paid traffic coming there. They're looking for a product that you have, but when they search for it and they get something else and they leave, you've actually paid for them to come here now and you could have sold them something, but now you can't sort of thing. Yeah.
Nathan Bush: Is that enhanced layer of data that you're adding for your retailers helping them showcase their product in different ways, like being able to create more category pages or being able to personalize recommendations better for customers?
Abhishek Vora: Yeah. So they can actually generate Granular filters, they can create different types of sort of categories in there. So for example, one of them, because we can generate all the colors for the products, so they can create a particular color palette which can help showcasing the product more than once sort of a thing. And also different collections that can be, that can be showcased. And also what we do is we end up generating the SEO tags as well, so which can then be further leveraged to drive traffic to the website as well. So what we are also figuring out is that look, from all this data, we are also seeing new ways of how brands and retailers are actually using their underlying data and can be showcased in different ways. And that's why we end up working well with other kind of tech stacks over there, because they can all end up using some amount of that data to work. There's so much discussion on customer data and understanding the customer. So we're saying that's important. But if you can really understand your product information as well, which are the two, obviously you know, that's the lifeblood for any E Com companies and you can join those together. I think that project or that part actually will have a huge merit over time.
Nathan Bush: And you mentioned the tech stacks that you work with and we've already mentioned things like Shopify, Commerce Street. What are the most common tech stacks that you find yourself working with with your clients at the moment?
Abhishek Vora: Yeah, so what we have seen is on the cloud side we've got aws. Then we work with a number of different PIM companies. Obviously we work very closely with commistry. We've done a number of events together as well. Same with certspring. And I think all these things together actually gives us a good story as well where we can generate the data, one of the payments can obviously store it and that information can then be leveraged by the search company to further enhance product discovery on the website. But also equally, we have seen a lot of brands who are sizable are using Shopify. And I think there are some good inherent capabilities within Shopify that are helping them stay with Shopify and work through it. But again, we are exploring other areas where we can really integrate. For us it is more about having the client where they are. And we say right now these are working itself. We are a small enough team, we cannot integrate with too many partners just yet. So we said let's work with the ones that we have got.
Nathan Bush: And it's one question that I find retailers really grapple with a lot is when is the right time to move from having all product in an all in one platform to having a dedicated pim. Do you have a view on this? I know you've got to be very neutral around platforms because they're partners and the environment, but from a retailer capability, do you find any signals or signs of when a retailer should be considering having products in a dedicated pim?
Abhishek Vora: Yeah, that's a good point. And I think we have seen across the board because generally I think the tendency is that, okay, you have to be big enough to have a pim, but I think now the reality is because as a retailer you're actually sending the product to multiple different channels as well. Where the PIM is also helping is obviously storing the information in one place and then you can obviously use that into different marketplaces, marketing channels, etc. The type of BIM obviously can differentiate between something which is easy to onboard and, and get started with. But I feel most retailers, if they are of any, you know, any size, should be looking at that because it actually helps them even from a governance perspective to hold all their products in one because otherwise you'll end up having, you know, multiple Excel sheets and something is here, etc. And as you grow you realize, oh, it's too late, sort of gone all bored in there. So the best answer to these, what I've seen is, is an ability to test it out. I think the. Because they will then know is it really for us or not? But also testing it out with, giving it enough time to test it because sometimes we just test it for, you know, a couple of weeks. These things don't show any benefit straight away that way.
Nathan Bush: Yeah, but I love your point there too, around having it as a structured environment too, when you're selling into multiple channels or distributing across multiple channels and treating this product data as an actual business asset that you look after in one place.
Abhishek Vora: Yeah, and also it's similar to some discussion we were having earlier. We said even if a small company, they would have some sort of a CRM. So if you're having a CRM of some kind, you should, because obviously you want to keep the customer information together so you can market it properly and you know, you can really use it and leverage it in the best way. The same way they should be thinking of the product information as well. If they can all have it in one place together, even if it's not significantly large, I think that would definitely bear fruit because they can start thinking of new ways to sort of start using that underlying data when it's all together in one one place.
Nathan Bush: And when you are first speaking to new customers. Where do you see most of the pain points come from? In the way that they've structured or managed their product data historically? Yeah.
Abhishek Vora: I think one of the things that we definitely see is most retailers or brands are performing that work manually. So if we, when we go and speak to the E Commerce coordinator, we can see the pain because they are the ones who are actually doing the work on the ground and realize. And it's also because of the sheer volume. Right. Because there's just so much volume that is going through. And this process hasn't changed, you know, in the Last sort of 15, 20 years since E Commerce became a thing, but the volumes have gone up 100 times, if not more.
Nathan Bush: And when you talk about on the ground, you're talking about them rocking back and forth under their desk with their laptop, trying to get it all done.
Abhishek Vora: Pretty much. Pretty much. And I think they're facing the real pain of getting it done. And I think I've also seen that if someone is a bit too senior in the organization, sometimes they don't feel it because they've not gone through it. And you know, E Commerce directors, et cetera, who were perhaps E Commerce coordinators before, they have felt it, they've gone through that pain before. So they know the amount of effort there is, otherwise it can become. Oh, that's. And I think that the best way I sort of mentioned to them is, look, if they're all thinking of, everyone's thinking about personalization, predictive analytics and all these big things out there. I said if the underlying information is not good, nothing will work. Because then it's just, you know, the old saying of garbage in, garbage out. And it may not be the sexiest thing at the moment sort of thing, you know, but it needs to be done and it needs to be done well. Because if it is not done well, then you have the problem of, you know, you look for a formal blue dress and you get, you know, white shots.
Nathan Bush: And that's going to be awkward when you turn up. Yeah.
Abhishek Vora: And just imagine if you walk into a physical store and you tell that you want this and someone comes up with white shots. You know, you'll say, so why is that okay, over there? And we see it on a daily basis in the E Comm. World. And I think this thing, if it gets fixed, it will definitely. Of course, there are other challenges that need to be sorted out to help final conversion, but this is one that we are seeing repeatedly that can be fixed now. And I think the answer is also sometimes that many are not aware that this can be automated. I think when we take someone through it, then they realize, okay, that's how it can be done.
Nathan Bush: It's funny to hear you talking through that because I put myself in the position of a salesperson and I go, okay, if I've got a product that the decision makers don't necessarily feel the pain, it's the junior team members that feel the pain. It's not necessarily the sexiest thing, but we know that it makes a difference. But it's not necessarily sexy. And most people aren't aware that there's. That this can be done, that it can be automated. How do you go into new organizations that may not realize it's a problem and sell them in that there is this brilliant solution available for them?
Abhishek Vora: Yeah, that's true. And I think what we refer to is. So, for example, we show the case studies that we have done, and for example, one of the retailers saw sort of a 25 to 31% increase in their conversion and what they attributed that to waste that, hey, you know, we just were able to generate a lot more granular filters and our searches were getting better outcomes than before. So clearly that's a big win for them. But also timing wise, for example, one of the retailers needed almost 100,000 products attributed and tagged relatively quickly. And if they had to do it manually, and they said if we were doing this manually, to have taken us five minutes per product. I mean, you're talking like, you know, 10 people working for five months and still not being able to get that done. And we were able to do it in sort of less than 40 hours. And to the real lift is in conversion, it's in saving time. And it's also engagement. As I said, conversion can be hard because there are lots of other things that can get attributed to that as well or that result in there. But engagement, it's a very clear to be able. We can show them before filters being used and after filters being used, because now users can trust those filters because users have also stopped using filters on certain websites because they think, oh, it doesn't work. So for us, that's what we've been doing. We've been going and showing that this is a really painful, mundane process, but it's very critical for your business and we can actually automate that process. And obviously the early adopters that we're working with have recognized that, look, we don't want to be doing this work. And then of course, some of the partners have also played in and said, look, we need this data. If you don't have good data, we cannot really create better experiences for your end customer. And I think those things help as well in getting us through the door.
Nathan Bush: I was about to say, if you get the roi, then it means your. Your partners, like your searches or your product recommendations, your personalization, they all get the wins as well. So I can definitely see where you bring the community together, because it's not a.
Abhishek Vora: Exactly. As I said, it's not a. It's this or that, man. You can say, we can actually generate this, which can then be fed into lots of your current tech stack, which will also benefit from it.
Nathan Bush: So what was the moment, Abby, where you went, I've got this idea and I think there's a business in it and I'm going to go after it?
Abhishek Vora: Yeah, that's a really good question. So we used to run a marketplace before. So we have basically four founders, and three of us were flatmates in New Zealand when we were starting together, doing our undergrad. And, you know, one went off and did his PhD in computer science. One moved to Melbourne and I moved to Sydney, and we met another founder who was based in Melbourne. So we were running this marketplace and we were, you know, uploading the products ourselves and making sure that, you know, we have tagged it, attributed it, and at some stage we said, look, man, this is just painful. I mean, getting all these products, but we need to make sure that the end customers can find it. And if you don't do this work, it's, you know, and it was just taking up so much time. So we sort of felt the pain ourselves in a way. We were the ecom coordinators collectively. And so this flatmate of mine at that time, he said, look, I'll build you guys perhaps something which could help in automating this process. So when we started exploring that so effectively, it started off as an internal tool to solve our problem. And as we went through it, we realized, hang on, this is a better problem to be fixing. So we pivoted the business, we shut off the marketplace, and we said, look, this is what I think we should go after. And then we sort of came up with the broader vision that once we can, everyone has the same problem. We went and spoke to a lot of the other people and they said, look, the data becomes a crucial bottleneck for all sorts of things. So if you guys can figure that out, you know, people will tell you how in different ways they're using that underlying data. You don't have to give them different engines. There are lots of engines. You just generate that oil. And we said, okay, let's. So that's how we sort of started off and, and then sort of moved in full time in getting it going.
Nathan Bush: Now they may do 1000 burpees before breakfast and bench press pallet racks for fun. But when it came to Black Friday, lskd needed some spotting. Introducing the well built Shopify plus look out. LSKD reaped platformed onto Shopify plus between the 2019 and 2020 Black Friday sales period in order to keep up with the huge growth and automated processes that were running through the business. And the results were a PB in Black Friday 2020, LSKD handled 50,000 orders. That's a 1100% increase from the year prior, even serving 19,000 shoppers at the same time. Talk about some serious gains. To read more of LSKD's story and see other case studies, visit the customer section on shopify.com auplus. You can also hear the full LSKD story and its incredible growth story back on episode 76.
Nathan Bush: But that's not a bad side project, right? To be able to create a machine that can look at a picture and then pick out, start picking attributes. And I'm assuming it was a lot more basic back then to where it's advanced to now, but where do you and the team start with creating that technology from scratch?
Abhishek Vora: Our main thing was one way to look at it is why has this AI become topical just generally, right? So and there's sort of three aspects to that. One is just around the world you've seen an explosion of data, right? Just any type of data, you know, you know, from sending emails to messages to searches to anything of that sort. So there's a lot more data that we are generating. The second is the processing power and computing power cost has reduced significantly because of the likes of AWS, Google's, etc. You can actually process a lot more faster. And third is really the advancement in the algorithms that has happened over the last 30, 40, 50 years. So these things together has allowed a lot more data to be processed to create maybe data sets and outputs that can be done, which is what has helped AI become more topical and something that can be done in there for us. The best way that I can explain is think of it like a, you know, it's an artificial brain, right? So we have sort of curated a data set which would be filled with V neck, square neck, boat neck, etc. So we are basically teaching this Brain, okay, this is what V neck is, square neck. And that's where our sort of ability has come in on how quickly can you train it? Can you train it with limited data set and data set which is without noise, because if you give it too much noise and it gets confused on what exactly is it? So we've been able to teach it and then we sort of partnered up with a couple of fashion schools as well to build a taxonomy of what is relevant to fashion, what would be appropriate. And based on that, we then created data sets. Then we taught this artificial brain and then it starts sort of splitting out the output and as it spits out the output, if it is incorrect, that's where we are getting others to audit that information and say, no, no, this is not quite right. And that becomes the feedback loop that goes back into it to further train it. So yeah, it's been, it's been that process in there. And sometimes it can be considered, you know, you know, the robots are taking over kind of scenarios when AI is involved. But really where we are around the world in terms of the AI practice, it is effectively automating certain manual heavy tasks and ketos. I think there's a lot more to go to get to that sort of general AI where, you know, they are, they understand everything. And there's, I mean, there's been huge progress made in the last 10 years in those sort of areas. But I think there's a long way to go before the robots take over.
Nathan Bush: Are there any product attributes or product details that you wish you could train these artificial brains on that you haven't been able to manage yet? And that you go, when we do that, that'll be the holy grail.
Abhishek Vora: Yeah, I think, see, we are also overlaying visual with text based. So certain things we cannot do it from a visual perspective. For example, material. So just by looking at it, if a human cannot say what type of material this is, it becomes hard for us to train it to learn something. So, but what we are doing is we are learning from involving text based recognition as well, because sometimes the supplier is providing what is the actual material, but it is still someone having to take that information and put it into the PIM or any other ERP as such. So we are building ways where we can recognize that because then it can become more complete than what we have today. So that someone, as an E Comm coordinator, you don't have to do half here and half that. You can actually complete the entire set in there. But also we are also trying to figure out how best can we leverage this underlying data to provide better personalization using the description styles. Because ultimately, what's important to the retailer is, is this helping in making discovery better? Some of it is, you know, you need to have good filters, etc. But as we progress from there, if we say, okay, we can achieve that, no problem, how do we then take it further and overlay that with the personalization, et cetera? Yeah.
Nathan Bush: Nice. And in terms of the categories that you're in, we've spoken a lot about fashion. You mentioned furniture before. Are you looking to expand out into more categories?
Abhishek Vora: Yeah, I think, like I said, we can train that brain to learn anything. We just want to be laser focused in these areas right now because we're saying that's where the visual plays the biggest role at the moment, is the fashion vertical, which includes, you know, apparel, footwear, jewelry, bags, accessories, etc. And the furniture homeware segment. But down the track, definitely, we'll be keen to explore other categories. We've got some categories in mind which we haven't trained for just yet. Just because we said, look, if we want to go after it, we want to make sure that it's done well. We just feel that we may not be able to give it the ample time that it deserves just yet.
Nathan Bush: Yeah, very smart. There's still a whole lot of human involvement in machine learning, isn't there?
Abhishek Vora: Yes, exactly. Exactly. So we need to make sure that because it just. The work moves further down where you're actually curating those data sets as well.
Nathan Bush: Yeah. When I was doing a bit of research for this conversation, I actually spoke to one of your retailers, and I won't name them because they might get embarrassed. And they were raving about you, obviously, but they said that they thought that you had the opportunity to disrupt the world of online dating by tagging humans on platforms like Tinder and Bumble so
Nathan Bush: that people get better matches. What do you reckon about that idea?
Abhishek Vora: Right, right. I can see one day there was a few drinks involved before coming up with that idea.
Nathan Bush: Maybe a few bad dating experiences.
Abhishek Vora: Yeah, I think I can see. Based on the quality of what fashion they prefer, matching them based on those attributes. You like abc? We like abc. Yeah. Look, I think there are different use cases on being able to use the underlying information, which is absolutely there. And what we do is, you know, once a month, we actually discuss a whole heap of use cases that have come out from customers, partners, teams, and then we sort of go and rank it and then we sort of store it. Into a data set and we say, okay, let's keep these ideas to come back and sort of work on going forward and how do we actually take it forward. So but the key thing is it's great to be able to see that the retailers are seeing value from what we are doing. And now it is just a question of how do we further enhance this. And I think that's a continuous process to continue to do that.
Nathan Bush: And look, 10 years ago we probably would never have thought that you could upload a product file with an image in it and something else would automatically populate all the attributes and even the description. Where do you think 10 years from now, Even if we take product out of play, but where do you think machine learning and intelligence, what role do you think that will play for retail say in 10 years time?
Abhishek Vora: Yeah, I think a lot of these things will become just business as usual. It'll become something which is quite new and cutting edge right now, will just be, hey, that's just part of bau. And then. And also it'll be more on sort of hyper personalization, understanding that intent of the customer, what do they really, really want and almost sort of surfacing. I know Amazons of the world do that, but I think being able to break it down to, for smaller organizations to be able to leverage all that technology will be there and I think that's where it will move towards where you can sort of understand and there'll be a lot more on sustainability as well. Where you know, even now people want to really want to really understand where is the product being sourced from, where the material is being sourced from, who's working on it. So I think through technologies like blockchain, etc, that's already happening, but those things will definitely progress this. The other part is because of the sheer volume increase that is already happening and is going to happen even more. The only way to solve a lot of these things will be to not just throw technology at it, but also build intelligent teams who can then leverage technology as an input to do it. Because sometimes a lot of the AI projects internally fail. Because as I said, the intention is either it works 100% or it doesn't sort of thing. And if it doesn't, we're saying, okay, we'll just default to this. And also culturally teams have to be prepared to go through that trial and error and see, okay, what really is appropriate for our business. Because you don't want to go after, you know, just to shine. Because you know, sometimes these names can be thrown around, right? AI Machine learning, it can be thrown around and it becomes like, oh, here you go again. Those are things. So, and I, and I think that has to be factored internally from the teams on what is appropriate for the business, what is the real problem? And can technology be an enabler in doing that as opposed to, can I just do it 100%? Ideally, yes, that's nirvana. But sometimes getting to that is not going to be in a, in a week's time sort of a thing.
Nathan Bush: Makes a lot of sense when, you know, we're already having trouble finding great e commerce people and encouraging them and moving them on a, on a path if they're spending their time getting creative and enhancing data rather than trying to reinvent the wheel every time from scratch. It's a nice solution from that perspective. But I love where you were going with blockchain because, you know, it got me thinking around, are we going to a world where it's not going to get to this point where we've got this final glossy photo of a product and then we tag it, but the tags actually build up along the way and the product data builds up along the way and gets carried along the chain so that we're not reliant on this one flat image towards the end to tell us everything about it. Yeah.
Abhishek Vora: And also, like we've been just discussing with people about the whole metaverse and how the use and the need for data is only going to expand exponentially. It's not going to decrease because the way we're going to search and the way we're going to find what is relevant to us is not going to be just by, you know, right now it's mainly text based. Right. So it won't be that. But at the same time, you want to be able to use those images, your Personas or what you've actually bought previously and you sort of carrying that along with you through whichever sort of world you might be in, physical or virtual. But the need for data will only expand and I think how you then leverage and use that data will become quite critical.
Nathan Bush: The metaverse plus dating plus Ocular. I think you've got a new product there, Abby, go register it.
Abhishek Vora: I'm changing business straight away.
Nathan Bush: Abby, it has been so great to speak with you and we could have gone down the rabbit hole even further, I think, on a lot of those questions. But brilliant to get the high level view on what you're doing at Ocular. It's fascinating and I think a lot of retailers will be listening to this going, we want to know more. So how can people get in touch with you and the team?
Abhishek Vora: Yeah. So no, thanks again. Appreciate it for getting in touch, obviously directly with me on LinkedIn thing or obviously from our website ocular IO. We've got our normal form set up over there, but I'm always curious to engage and chat with retailers because every time we have a conversation, it actually helps us think of new ways to help them and obviously new type of features, you know, some more basic than what I think Nathan, you were going with. But still, those are, you know, you. Those are really helpful conversations. So. But I really enjoyed this conversation and it's given me a lot of food for thought myself. So thank you for engaging on this.
Nathan Bush: Thanks, Abby. It was brilliant talking to you. So if you're listening to this podcast
Nathan Bush: while you're mapping out all that product data, know that the robot help is on its way. Just hang in there. Here are the three biggest takeaways I took from our chat with Abby today. Number one, multiple channels usually require a pim. I often get asked the question by growing retailers of when is the best time for a dedicated pim. That's Product Information Management System. I think Abby articulated pretty well when he said that as soon as you are selling in multiple channels, such as websites, marketplaces, third parties, where they all require different types and formatting of data, a PIM becomes crucial for maintaining that one source of your very, very precious product data.
Nathan Bush: It's a simple response to a difficult
Abhishek Vora: question, but I liked it.
Nathan Bush: Number two, AI isn't perfect and don't expect it to be. One of the key messages that Abby kept coming back to is that he doesn't expect Ocular to be perfect every time. Instead, he would rather it take 80% of the grunt work that you used
Nathan Bush: to do and leave you with 20%
Nathan Bush: of the refinement and creativity.
Nathan Bush: I made up the percentages, but you get it.
Nathan Bush: It's better than doing the whole thing from.
Nathan Bush: From scratch.
Nathan Bush: I've seen AI tests fail when they don't perfectly replace humans. That shouldn't be the task. We should be asking how they optimize humans. And number three, train the machines. Extending the concept of AI not having to be perfect, Abby gave us some insights on how they've trained Oculus AI robots, even taking the data to fashion schools to help train the brain on different categories and the features of fashion. The nerds can't do it all. This training side of AI is often overlooked, and it's important to spend the time upfront training the technology to give long term efficiencies and intelligence. It doesn't just happen automatically. To get the highlights of today's episode, head on over to add to cart.com and sign up for our free newsletter each Tuesday. We will send Monday's episode summary links and discount codes for you to go next level on. And if you're looking to explore your next e commerce opportunity, come and Visit us at eSweet. We're a dedicated e commerce talent agency connecting the best e commerce talent with the fastest growing brands in Australia. Head on over to esweettalent.com where you can download the free e Commerce Salary Guide and sign up to our weekly e Commerce job emails. Thanks for listening and until next time, keep those customers adding to cartoon.
