Synopsis: In this AI Leadership Insights interview, Amanda Razani speaks with Bernadette Nixon, CEO of Algolia, about AI technology in the retail industry, especially as it pertains to customer searches.

Amanda Razani: Hello, I’m Amanda Razani with Techstrong AI, and I’m excited to be here today with Bernadette Nixon. She is the CEO of Algolia. How are you?

Bernadette Nixon: I’m doing very well Amanda. Thank you.

Amanda Razani: Good. Can you explain a little bit about the company and what services does Algolia provide?

Bernadette Nixon: Sure, absolutely. So Algolia powers discovery. And we do that with our AI search and discovery engine. So when I say search and discovery, everybody gets the search part. It’s the search box that we all type into multiple times a day. When I say discovery, people don’t always relate to it as much unless you’re in the industry. And so what that means is, when you’re on a website – let’s say it’s an e-commerce website, and you’re browsing around – I might be having trouble sleeping, I might be on a, you know, a luxury handbag site. And I might be looking at crossbody bags, or shopping bags, or a clutch. And as I look around, and I browse around that website, and I click on something, and it’s not quite for me, but I see the recommendation carousels at the bottom that we’re all used to now. Well, search technology powers all of that, which is why we call ourselves a search and discovery platform. And we get the honor of serving 17,000 customers at this moment, and over 1.7 5 trillion searches on our platform. We’re a private search engine; we take over when Google or Bing drops you into a website. And so when we do that, one way to think about how big we are; we’re probably second only to Google as far as the data that we can get our hands on. So we’re four times the volume of the top five public search engines, if you take Google out of the mix. So you know, we’re pretty big in our own right.

Amanda Razani: That’s fantastic. Thanks for sharing. And so you have some experience with artificial intelligence. And I want to ask you, what advice do you have for companies as they try to harness the benefits of AI and implement it into their processes?

Bernadette Nixon: Gosh, well, I mean, AI can have so many different meanings. 2022, as we all know, was a major watershed point with ChatGPT and DALL-e and generative AI in general. And so what I would say is, if you’re looking to, as an enterprise, if you’re looking to leverage generative AI, and build generative AI applications, leveraging your own data, then obviously you won’t be doing it with ChatGPT, because of security and privacy, and therefore compliance issues. So you will be, you know, you can still do it. But you have to then get your own data into a repository, into some sort of a data store, where those generative AI apps that you’d be creating, where they can access the data; because at the end of the day, data is the fuel of anything that is to do with AI. And so when you’re looking at making your own datasets accessible to your AI apps that you want to build, you’re gonna have to vectorize your own data. And some people would say you need a special vector database for that; we would actually say you don’t really want two databases that you’ve then got to keep in sync. We need to help – we as the industry – need to help those enterprises by just vectorizing the data institute. Where it is doesn’t matter – if it’s a SQL database, or, you know, a no SQL database, any of the different flavors that are out there – graph databases. But essentially, the you know, the words of wisdom are from everybody in the industry, data is the fuel to AI. So being able to unleash your data is of paramount importance, but doing it in a secure way with an eye to privacy and compliance.

Amanda Razani: Absolutely. What are some of the roadblocks and challenges that you see businesses facing?

Bernadette Nixon: Well, I mean, I think some of this is just uncharted territory. So I think there’s a lot of experimentation that is going on right now. And I think that’s good. I think that is very good. You know, the governments will figure out how they want to shape the legislative legislation. I mean, Sam Altman was up on Capitol Hill, what, two weeks ago now. So you know, we’ll see that landscape change over time, I’m sure. But I think, you know, experimentation is the name of the game right now. And not to be afraid of it, but you know, you’ve got to do it in a, I guess, a responsible and somewhat contained way.

Amanda Razani: Most definitely. So Algolia has introduced a powerful AI solution recently. Can you talk a bit about that? What is NeuralSearch and how and why was it built?

Bernadette Nixon: Yeah, so as I mentioned earlier, last year changed everything, you know, with the democratization, if you like, of AI, but what it really did for us as human beings is, we’ve all, just as humans, we’ve always wanted to be known and understood. That’s just, you know, part of being a human being. But what last year taught us, or I should say, probably the beginning of this year as well, what the – you know, the generative AI movement and ChatGPT coming out towards us is that you know, we no longer have to put up with the crappy experiences and the poor chatbot experiences that really don’t understand us; we now expect to be understood – and from our context, in that search and discovery journey that we’re on. And so that caused us to say – and this was, you know, way earlier last year – we believe that there would be a significant evolution in how we as humans want to interact with our mobile phones, or iPads or laptops, whatever. And as a consequence of wanting to be understood, in a search paradigm, that means not searching in this way that we’ve been trained to so far, which is very staccato, red sweater, black shirts, you know, blue shoes – instead, we want to be able to search as we think. So what that requires from the industry is to be able to take the traditional keyword search: Black dress, blue shoes, and merge it with a natural language search capability based upon vectors. But when we decided that, it wasn’t an either or world; we wanted to merge both together and make both of those things, both of those different types of search modalities available to our customers for every single search that consumers are making on their sites. So we bring together keyword search, and the natural language vector search, which is getting intent and all of that juicy stuff, and we bring all that together in one product called Algolia NeuralSearch.

Amanda Razani: Fantastic. And so is large language modeling used for this product?

Bernadette Nixon: It is; we use several LLMs to train our product. And so that in conjunction with the algorithms that we use, really enable us to deliver some amazing results. So we’ve had beta customers, and now production customers, and some of the results that they’ve been receiving have been quite amazing. I’m happy to share a few of those if you would like.

Amanda Razani: Yeah, so I was about to ask you, if you could provide some real world examples.

Bernadette Nixon: Yeah, sure. So we’ve had some customers like Uniqlo, who’s in the top 10 fashion brands in the world, Everlane – here in the States, House of Fraser, and their various sub brands in the UK. So we’ve had various different global brands on the platform, NeuralSearch, already. And obviously, the top thing that any retailer is looking for is how do they increase the conversion, because that increase in conversion results in more revenue to the top line. And so we’ve seen results as high as a 23% increase in conversion to revenue. It’s been, I mean, that’s an astronomical number. And on the other side of things, it’s funny, I was reminded of this myself over the Memorial Day weekend – I was searching for something online. And I got twice, I actually in the same session, I got no results returned to me for a search – not even a recommendation as to something that might be similar. And there’s no excuse for that these days. And so with our customers that have looked at this technology as well, they’ve had anywhere between a 45% and a 70%. decrease in the number of null results, as we call it; zero search results returned to a customer. So what you’re seeing with neural search is both an increase in the conversion to revenue, as well as an impact on the customer experience and the overall customer journey.

Amanda Razani: Wow, that’s significant. And I believe I read something about, with Algolia, you enable retailers to adopt this AI enabled search without any changes required to their existing code.

Bernadette Nixon: Yes, so for example, if you – Algolia has been around now, we’ve in our 10th year. And so if you’re an Algolia existing search customer, you don’t have to touch a thing. There is full forward compatibility, and you don’t have to make any changes at all. Now, over time, you may choose to make some changes that would be for the positive and you may decrease the number of rules that you want to have, or you don’t need to have the number synonyms that you used to; you know, I come from England originally. And the things you wear on your feet to go running are called trainers in England and over here, the are sneakers. That would be a synonym that nobody would need to manually or even artificially, you know, AI curate any longer, because it’s just part of the LLM, and therefore, the system understands a sneaker is a trainer. So it’s only things that are very specific and proprietary to your brand; your industry that you would really need to do that for. So you can make some changes for the better as you go forward. But yes, you know, there’s a very seamless migration path.

Amanda Razani: And you mentioned it a little bit earlier. But can you explain how does this differ from other generative AI technologies like ChatGPT and other ones?

Bernadette Nixon: Yeah, so it’s interesting in the retail field. Right now there are certain junctures on the customer journey, where you do want generative AI to give you examples; let’s say you’re a spice manufacturer, and you had recipes on your site. And if you knew that I had turmeric, and cardamom in my cart, you might want to suggest different recipes. And that might mean that I would have to buy three additional spices from you. Well, that’s a great place to use generative AI in the customer journey. However, when I’m searching in the search box, that’s where you have the highest intent of anybody who is on your your website. And therefore, you don’t want to – if I’m on a Louis Vuitton site – you don’t want to imagine and generate this amazing handbag that I could go buy if I can’t then click, you know, buy now. So you don’t want, you know, you don’t always want it – what you need is the best possible retrieval of what is going to be most suitable for me because you want me to click buy, and not just leave it in my shopping cart. So therefore, I think there are places where generative AI absolutely has a place in retail. When you look at what we have today, it is retrieval, not generation. However, we are also coming out with a conversational commerce product that will give you the best of both worlds, it will give you some generative AI capabilities, it will give you the chat modality without it being annoying, as most chat bots have been up until these days. But yet, we’ll be running that against your product catalog, for example, that we have in our system. So kind of the best of both worlds at that point.

Amanda Razani: Oh, that’s so interesting. So more like real time intelligent chatbots?

Bernadette Nixon: Yes, I would say non-annoying chatbots.

Amanda Razani: In this day and age, when we are seeing a lot more customers ordering remotely, just due to COVID. Ever since then, there’s just been a lot of stores turning more to e-commerce and more customers shopping online. So what advice do you have to businesses when it comes to that customer and the end user experience – how to improve that?

Bernadette Nixon: So, when you take a look at, for example, the things that you and I might be on a website looking for – you know, we might be searching for, like I said, blue dress, black shoes, but we might want to actually search for something much more specific. I was looking for hardwood flooring recently. And I was not liking any of the colors that I was getting. So then I thought, hmm, let me put blonde hardwood flooring in because that was you know – blonde, obviously is more of a hair color than a wood color. But I wanted the lighter hardwood floorings. Well, I got weird responses. I got olive wood chopping boards. Totally not relevant. But my point here is that some of us go to websites. And we have what, what the industry calls longtail searches; that’s a search that’s anywhere from three to five words long, and it’s what people call the long tail. And all of those long tail searches are, they’re low in volume. So they’re not you know, it’s not like somebody coming to a website, a sneaker website and looking for Air Jordans. No, they’re looking for something that they’re putting several descriptors on. But the combined volume of all of those searches can make up as much as between 50 and 70% of the searches that a brand or a retailer will see on my website. So cumulatively they add up, whereas each one search might not drive a whole heap of revenue, but on mass, those long multi-word, three to five word searches, they really add up in terms of potential revenue. But the problem is that no company has the amount of merchandisers to be able to then merchandise for that long tail. And so as a consequence, the only way that you can improve that is with AI. It is weird because we have the large language models that we train on. And so we can enable you to get behind the intent, the real intent behind the question that somebody is putting into the search box. And so that’s the, if you want to ask me, the low hanging fruit on the table is for companies to use, you know, NeuralSearch, for example, to be able to monetize that longtail, because you could just never have enough bodies in your merchandising department to throw at doing that. So it’s money that – it’s the low hanging fruit. But up until this point, it hasn’t been the low hanging fruit. So that’s the one place that we see a lot of interest right now.

Amanda Razani: Absolutely. So, this technology is advancing so rapidly, and it’s hard to keep up. And looking at the future – and I’m going to say just one or two years – because of how quickly it is advancing, what do you see for the retail industry as a whole?

Bernadette Nixon: Gosh, crystal ball – I always wish mine was a little clearer. You know, I think like every industry right now, if you’re not using, or at least looking to use, AI in your business, then you’re going to be looking at an extinction event. And the problem with that is that in order to be able to leverage the AI, you need data, as we said before. But you also need normally, you know, very, you know hoards of data scientists, which are a very scarce resource, and they’re tough to keep once you have them, because everybody wants to poach them from you. And you’ve got to be a pretty technically adept company. And whilst there are many of those out there, there are lots that aren’t. And so most companies are going to need to put a cadre of partners together that are AI-as-a-service companies. like us, think of us in that manner. For search and discovery, you don’t have to run it, you plug it in, and then you know, we do the hard work for you. I’m not saying that there’s no work that you have to do. But if when you look at it, the reduction in the work that you have to do is far less. So I think that’s one key sort of message. I think the other messages that you know, we’re gonna hear a lot about AI everything. So in our case, AI search, AI discovery. But the problem is, with the underlying technology, as I mentioned, the vector databases, they’re great. However, they’re slow, and they cost a lot of money to make them performant. And you can see that on Twitter, right now – you’ve put in pretty much any vector database, and you go on there, and you’ll find lots of stuff about how to cost optimize, because you know what, it’s costly. And it’s not easy to do. So that’s where I’ll go in NeuralSearch is different to everything else that’s on the market, because we have a technique that we have developed that basically shrinks those vectors whilst maintaining the fidelity and then cutting a lot of the costs out – almost 90% of the cost – out of the equation. So you know, I think it’s also, that’s why customers and companies out there, retailers and others alike, need to have good AI as a service partner company, because fishing through all of these different elements is just going to be really hard and you need somebody there to help guide you.

Amanda Razani: Yes, partnership is going to be critical. I feel like that as well. Well thank you so much for sharing your insights with us today and I look forward to the future and what it holds.

Bernadette Nixon: Thank you for inviting me Amanda.

Amanda Razani: Thank you.