Latest / Inside Commerce: Ecommerce Strategy, CX and Technology Podcast / EP172: Ecommerce Product Discovery Trends & The Shift To AI-Driven Personalisation, with Attraqt
Transcript
- 0:02Thanks for listening to the re-platform podcast sponsored by
- 0:05tractors hypersonics and hosted by me.
- 0:06James get and Rogers warm, welcome.
- 0:08If it's your first time on we hope you like it.
- 0:10We put a lot of time and effort into bring this content you
- 0:13please help us keep the podcast free for everybody.
- 0:16Yo refers other people in your network let them know exist.
- 0:19Share our posts on LinkedIn. Twitter and also give us a
- 0:22rating on Apple's Spotify, YouTube or wherever you
- 0:24listening, really helped boost visibility and get our name out
- 0:28there. So thanks and hope you enjoy
- 0:29this episode. Code.
- 0:30So let me set it up for introduce our guess.
- 0:32We're talking about transition into and getting the most from
- 0:35an AR I brought discovery engine.
- 0:36So quite a important and topical discussions day and we're also
- 0:42trying to one of our sponsors, but don't worry.
- 0:44This isn't the Tories hijacking. The BBC.
- 0:47It's going to be packed with useful and practical e-commerce
- 0:49insights. So, there's my my political dig
- 0:53in done for the for the week. But the reason why we talk about
- 0:57products covery, it's because e-commerce, Above my people who
- 1:00are searching browsed recommendations personalization
- 1:03as if they were distinct capabilities.
- 1:06And now, focus on what products cover is, how people find and
- 1:09how your personalized experience across all of these different
- 1:11user Journeys and attracts has long, been one of the market
- 1:14leaders. It's the convergence of bread,
- 1:16Hopper indicator, and they've got impressive customer base, a
- 1:19Brands, like a sauce, fanatic self, just group EXO, wealth of
- 1:22Knowledge from Real World. Produces what we're trying to
- 1:24bring to you today we're going to cover key trends in
- 1:27e-commerce products, coveri cultural shift to a wider Even
- 1:30personalization and the complexities of products covery
- 1:32at scale. So with that, as our position,
- 1:35let's introduce Our Guest. So start with nigga, who's the
- 1:38chief strategy Innovation officer?
- 1:41Welcome nigga, how are ya? Good with didn't think you're
- 1:43doin? Will they still be here after?
- 1:46Yes, I'm I'm Nicholas McDonald. And she said, together tracked
- 1:50it down that track four years ago following the acquisition of
- 1:53early builds. It was a personalization says
- 1:55that for my funded of Danielle's ago.
- 1:58So I think in some personalization Market Since a
- 2:00couple of years now. Fantastic.
- 2:04Yeah, I'm really looking forward to having this conversation with
- 2:06you and our other guests. Stay in run, is the VP for
- 2:09go-to-market. Same room with you Luke and talk
- 2:12about what you do, but right. And then also, do you wanna give
- 2:15us that that kind of introduction for those who don't
- 2:17know attractive, what of what the product is, and what the
- 2:20business is? Yeah, yeah, sure.
- 2:22So, in ranch, Airdrie BP Gate Market.
- 2:24So I look after how we're taking our products are solutions out
- 2:28to the market that involves working with sales marketing
- 2:32product, the Whole site. All different areas of our
- 2:34business was quite nicely. The gets quite a lot of time.
- 2:37We've both the product team and customers.
- 2:39So hopefully some of the insights to share it going to be
- 2:43valuable. The introduction is quite hard.
- 2:46Now boj's can show you. Did it a, pretty Sterling job, I
- 2:49have to say, but yeah, you're quite right.
- 2:52The market has historically been distinct in terms of search
- 2:56merchandise and recommendation tools and with that you get
- 3:01disparate systems Consoles, configurations and then side
- 3:06they data and we all know that the customer Journey when you're
- 3:09going through Discovery, how you find something, I discover
- 3:12something, you know, how you get to see the right product at the
- 3:15right time, through recommendations or
- 3:17personalization, you know, it is hard to create those experiences
- 3:21if you aren't joined up. So, you know, they're the
- 3:24solutions, but we look to join up and provide a bit more, the
- 3:27consistent by Discovery Journey with regards to product.
- 3:32And content. Lovely.
- 3:36So am I loved us first question? And warm up, everyone brain a
- 3:39little bit around my product Discovery and econ and so
- 3:44obviously there's a lot of I guess skepticism and concern
- 3:48around the economy at the moment and globally.
- 3:51But particularly in the UK and Europe, how is this impacting?
- 3:55The kind of buyers mindset and how is this impacting I guess
- 3:59you guys and your market and demand?
- 4:02Yeah sure sure. I'll take that one if that's
- 4:04alright because so We actually launched two reports recently
- 4:08the obvious one, Pete Trader report, everyone kind of does it
- 4:12donate. And then we commissioned a
- 4:15report with the London Business research group on the state of
- 4:18product Discovery. So, I kind of got a few stats,
- 4:22so not to throw too many. A qubit is quite interesting.
- 4:25So when we look at Pete trading last year, P trading did dip.
- 4:31A lot of people have reported that.
- 4:33So did by about one and a half percent Sent.
- 4:35Also, what's interesting, though, is Black Friday.
- 4:40So Black Friday accounted for only 11% of that total week.
- 4:46That's that's actually the lowest we've seen for about five
- 4:49years. And when we look at all requests
- 4:53from January to kind of November, there's 11 days
- 4:57probably any accounted for about three and three, three and a
- 5:00half percent or so, we'll request.
- 5:02We saw throughout that year. And that again, was lower than
- 5:06we've seen before because typically, the average is about
- 5:094.5, 4.8 percent. So we are starting to see verd,
- 5:14you know, last year. What, with the cost of living, I
- 5:16won't get political like, you James, but various political
- 5:20things were happening. The time people were worried
- 5:22about mortgages, their disposable income.
- 5:25So it we did see that as an effect and you know, our
- 5:30Merchants are now dying to look at.
- 5:33How do we rebalance things? Obviously disposable income is a
- 5:37lot lower cost of living concerns and their kind of
- 5:41drilling down on the tools that help them sell and engage more
- 5:45so that they can try and rebound some of that, that impacts a the
- 5:49London Business research report on that state of Discovery was
- 5:53quite good to see forwards in a way.
- 5:55But, you know, in some respects we saw about 60% of merchants.
- 6:00Now, looking this year to really focus in on product Discovery,
- 6:04purely to try and improve those conversion rates and average
- 6:06order values. So that's I guess the revenue
- 6:09side of it, and then the other side of it was around again.
- 6:13Another sixty percent of those folks, but we survey.
- 6:16So we serve about 300 odd, Merchants about 60% of them are.
- 6:21Now, looking at efficiencies and really looking at those
- 6:23efficiencies freeway are so they think Merchants are trying to
- 6:28find a Strategy to help rebalance, but for sure the
- 6:32cost-of-living crisis is, is have an effect for sure.
- 6:36Yeah. Right.
- 6:40And then next question. So AI is ever a moment and has
- 6:45been to the last few months, obviously result of partly as a
- 6:48result of chat to be T and then more recently, GPT for up.
- 6:53And lots of people are talking about, like, use cases and the
- 6:57overall impact. I'd imagine a lot of your
- 7:00customers would move from manual or kind of rule based
- 7:02approaches. Can you talk us through the
- 7:05operational impact for E carbon teams?
- 7:09And maybe kind of your view on how you get the optimal balance
- 7:14between the two approaches. Yes, true.
- 7:17So, so, the first big impact fossils team is in term of
- 7:22change of mindset as they are used to control everything to
- 7:25create another prudes, and many times they end up with a
- 7:29website, which is a controlled by hundreds of different rules
- 7:34for each category and things like that, and they need to
- 7:37change their mindset and Trying to apply to a full liberal base.
- 7:41Configuration is a very challenging because it does not
- 7:45give room for the AI to learn. Basically, and AI needs to have
- 7:49route to express itself. And so it's better to start by
- 7:53removing as many rules as possible, activate the AI and
- 7:57then apply rules where you want to add more corrosion or drive.
- 8:01A little more with will be presented to your Twitter
- 8:04Shopper. So it could sound a little like
- 8:07something complex or that will Take time.
- 8:09But actually it's a so good opportunity to to do a big
- 8:13cleaning. We see many clients where they
- 8:16will do the rule. They jailed in your other nails
- 8:18with different people working on it and not everyone, understand
- 8:22everything that is configured on the website.
- 8:25So there is also, the AI isn't so good opportunity to clean
- 8:28that to stop fresh and to to concentrate on, what will be the
- 8:35game changing rules for your business.
- 8:37So really to work on what is strategy in terms of fruits and
- 8:41not things, which are just there to to, to cheat, to decide how
- 8:46to present products that without Redeeming, the is that it also
- 8:50allows you to maximize a ride because you will free up some
- 8:54time from Naturalizer. And you will be very focused on
- 9:00what is business-critical. And from your experience in
- 9:06terms of how people need to think about making the adoption
- 9:10successful, is it about being sensible and how you roll out if
- 9:16you're moving from Rules base, where you said?
- 9:18Yeah, there are some signs up, hundreds, and hundreds, and try
- 9:20to unpick them is quite challenging where you have
- 9:22Global rules, local category-specific.
- 9:25How would you recommend they approached it rather than doing
- 9:27a big bang? Is it about take in category by
- 9:29category, how would you approach?
- 9:33Yes, it's so that you could definitely do that category
- 9:36bucket to read for all International play and sometimes
- 9:40they will start on on the country or a market, which is
- 9:43probably nodes. It is a big wild.
- 9:46So they will start with the smaller Market to, to do some
- 9:48tests you can do. You can also approach it by by
- 9:52use case. So, perhaps, for example, you
- 9:55can start with some simple use case like product similarity of
- 9:59complementarity. It's quite simple because it's
- 10:02the It's something really mature but so it's value will find a
- 10:07lot of are, right? So it's a good way to start
- 10:10because you will never age it quite easily and quite quickly.
- 10:14So so you can also take it so bad, peruse case, do category
- 10:19for markets or is different approach which is I totally
- 10:21agree with you. It's not a good idea to do to do
- 10:24just like a big bang and so on all your website with web stuff
- 10:28by use case can be a good approach shape.
- 10:31Yeah and I guess the use case I like the way you think about
- 10:34that because then it makes it more relevant to the specific
- 10:36business rather than just taking a generic right to it.
- 10:39I'd love to get a complete search specific context because
- 10:43for some site, search is the number one user Journey for
- 10:46Revenue driving. The conversion rate can be a
- 10:48two, three, four times higher than on Serge.
- 10:50I know there are exceptions where that's not the case.
- 10:52But for a lot of retail, e-commerce, what AI features.
- 10:56Dear dear provide on-site search.
- 10:59And what? Well, I want the measurable
- 11:00improvements of benefits that you've seen in.
- 11:03Agents using it. Yes.
- 11:05A if I can just keep back one, if that's our James, just to the
- 11:10use case piece, I will come to search.
- 11:12Don't worry about it. This just in terms of what I
- 11:16guess, Nick, I started with the mindset that is possibly one of
- 11:19the greater challenges but we see of those organizations that
- 11:23are moving from, you know, traditional manual, way of doing
- 11:27things to been dying to think of using AI, that mindset and that
- 11:31trust is kind of one of the big challenges.
- 11:34Judges, but we see. So you want the the idea of not
- 11:38only around the use cases to rho into but the use cases to also
- 11:43grow confidence and Trust, that's kind of what we also see
- 11:48as kind of one of the big things we work with customers on.
- 11:51It is a change, you know. So it's kind of little little
- 11:55and often and been kind of rhetoric as you start to learn.
- 11:58But also how you start to men build confidence and okay, they
- 12:03searched, I think. Let me search perspective.
- 12:07What we're seeing is Merchants are trying to uncover revenue
- 12:12and they're doing that by. They've been doing it for many
- 12:15years in terms of optimizing search for many years, but now
- 12:18they're kind of looking at focusing on the long tail, that
- 12:22a long tail of search. So, that's kind of where you're
- 12:25seeing a very low volume of search queries, but the type of
- 12:29search queries that you've got a highly unique.
- 12:32Yeah. So because it slow Volume and
- 12:35it's very unique and every one of those search queries often,
- 12:40you know, there's no real optimization or focus on them.
- 12:43So, typically, that's where you start to see those zeros search
- 12:46results. Will you see a result?
- 12:49Which is kind of Highly irrelevant.
- 12:51So, one of the examples we saw with waitrose where as you know,
- 12:55historically you go type in Oyster Bay wine and you'd get
- 13:00oyster sauce. Yeah, which you don't want to be
- 13:03drinking, copious amounts. Of that, that was kind of, you
- 13:07know, a fringe example which really able to deal with that
- 13:11now thick, which is good. But it gives you an example of
- 13:14what people are trying to search for but maybe not being picked
- 13:17up in chania Trend reports or in high volumes.
- 13:20But there's Revenue that we we roughly see about twenty percent
- 13:26of all search queries kind of falling into this, this long
- 13:30tail. And what we know is there's a
- 13:33commercial value to every sir. Search query, but you make.
- 13:35It's roughly about 4 pounds on average.
- 13:38Okay, so in an area, but if you can try and figure a way to
- 13:42improve it, there's some Revenue that you can start to get back
- 13:45the challenges, you know, it's a lot of effort and a lot of time,
- 13:50which actually in traditional ways, it just doesn't pay off.
- 13:54Yeah, stay. That's kind of where we stay low
- 13:57hanging fruit, whereby a I can actually take on some of that
- 14:00burden for you and that's where you can start to get some value
- 14:03back. So, So, we typically find that
- 14:07when you do it right and you're implementing a, i to try and
- 14:11focus in on this long tail, we can typically on average, see
- 14:15about 97 percent of that long tail being removed.
- 14:18In terms of there is more relevant results or
- 14:21recommendations but have been populated.
- 14:24So you kind of uncovering and taking the taking that missed
- 14:27opportunity, actually turn it into something that can
- 14:29hopefully convert and kind of help you.
- 14:34Attract, you'll Complete product Discovery growth engine, create
- 14:38relevant shopping experiences that convert into sales and grow
- 14:42online Revenue with personalized, search
- 14:44merchandising, and recommendation Solutions powered
- 14:47by a I find out more at a tract.com.
- 14:52I've got a sub question related to that and I'm very large and
- 14:54she do before the spare. I haven't couple of people on
- 14:57LinkedIn or me about this question specific.
- 14:59Is it always comes up is is when you've got more natural language
- 15:03based or Why multi-attribute driven?
- 15:05Which is again, it's the it's the low volume, it's all the
- 15:08edge cases. But you often find that with a
- 15:11standard setup that you'll get the zero results and then you'll
- 15:15just get default, you know, fullback presented which isn't
- 15:17right thing. A lot to know how the AI
- 15:19specifically tackles this case and how it automatically is able
- 15:22to react versus you having to programmatically set things up
- 15:25is like a sizing. So fashion, and you work at all
- 15:30and big fashion brands, you might have like, a floral maxi
- 15:33dress. Is 10, and you're only going to
- 15:36get a handful of those every month.
- 15:38So it's not a massive Revenue driver, but it's a CX, but a,
- 15:41but a lot of time, the default approach, is it will look for
- 15:44Maxie will find Maxie, or will know that floral my be an
- 15:47attribute pattern, but it's not looking for a size 10 and
- 15:51passing that is looking for that in the title and it doesn't then
- 15:54have the added intelligence and known.
- 15:56If this item is out of stock in Sloan DD boost that in the ROM,
- 16:01it's on, I'd love to hear how the AI from you A point of view
- 16:04is set up to solve that more challenging scenario, quite
- 16:10nicely hand that one over to you class.
- 16:13So the fitness thank you very red.
- 16:16So the first thing is traditional style should look.
- 16:20It's their filthy @g W. So it will decompose the query
- 16:24and look at each keyword and see if it can find it in the
- 16:27different products. It doesn't work for that typical
- 16:30example with a mix of keywords from product that So sighs, he's
- 16:35intent. And thing like that.
- 16:36When we use AI, we will base all the learning on how previous
- 16:42users interacted with the search online.
- 16:45So, based on the way they looked at product.
- 16:49They so they did the search query.
- 16:51And we will look at what did they do after that search query,
- 16:54and what can we detect from that?
- 16:57And so query that will include the size of thing like that.
- 17:01So yeah, I Will Learn by itself. To properly react to that
- 17:05because that's exactly what we're looking previous users.
- 17:08So, of course, it requires some traffic to do that.
- 17:12What we did is we depending on the clients, but with some
- 17:16clients who have less traffic, we can also share the learning
- 17:19between different play and so on the typical queries that will
- 17:23include some size requests. The AI we learned a will have
- 17:28learned that kind of queries on all our clients set basically.
- 17:31So that's how we end. On that.
- 17:34So we use more user Behavior more than keywords when we talk
- 17:39about AI. Undone from a point of view of a
- 17:43business that CA is going from zero and doesn't have like a
- 17:47massive historical data. You want to, you know, as a
- 17:50business that size as an attribute and you want to help
- 17:53the the AI in the ml learn quicker, how what the ability to
- 17:57like, give it that knowledge in advance and say, well,
- 17:59basically, let when we have a word like this, you need to be
- 18:03looking for this specific attributes and yet.
- 18:06So first mini business, have more they do than Later, if they
- 18:11have a Google analytics, like the in the website, they are
- 18:14already starting to collect some, some data about the query.
- 18:18So users are doing so it's not like, they are starting from
- 18:21nothing. Usually they have data, but they
- 18:23don't always know. They have that data and how to
- 18:27exploit it, but, but we have that knowledge and we can bring
- 18:30that to them. If the data is that enough, we
- 18:34can also share a. We can also share Knowledge from
- 18:37the probe, the verticals. So so we have to preach
- 18:40Algorithms that can be applied to a new business for to help
- 18:45the start and avoid the cold stuff basically.
- 18:49But but again also clients we worked with the they add some
- 18:55existing data somewhere even if sometimes it didn't know that
- 18:59because it's out there we'll just doing analytics.
- 19:01But actually what is used for analytics can be used, absurd
- 19:05training, the iPad from. So, so sometimes I already tell
- 19:08them we can do more with. And what the South research data
- 19:13behind case? I think it's worthwhile by just
- 19:15on that pre-trained algorithms, you know, that's kind of a warm
- 19:18standby, but now being a pain you need to have, we need to
- 19:23have had kind of that training that data to be able to provide
- 19:27those warm standby. So you know beat be Nathan.
- 19:31If you're talking fashion electronics home where DIY
- 19:36garden kind of our call verticals then That pre-trained
- 19:41is that if you came to us and asked about, you know, I'm a
- 19:45mobile phone operator and I want to sell Mobile's, we've got less
- 19:49data. Therefore, that pre-training
- 19:51isn't not to save it, it you won't get there but it means,
- 19:55but you are kind of semi starting from scratch because we
- 19:58don't have that that capability because we've not been building
- 20:02it over a period of time where we can offer it to our core
- 20:05vertical. Say you know just Spirit of
- 20:08openness very conscious that You know there are certain verticals
- 20:12where you probably would need to start from scratch and it's
- 20:14going to take a bit of time to help with that learning, sorry,
- 20:17pork. No, that's fine.
- 20:19I was just going to say I had an experience last night.
- 20:22Actually, there is kind of relevant to so I am, I recently
- 20:25lost my suitcase. She's very frustrating and I had
- 20:29a number of new items of clothing and that, and I decided
- 20:32and I just got some money back. So, I decided to go and look for
- 20:34some of the items that I went on and clothing.
- 20:36And I just checked to make sure they're like, is new guys and
- 20:38I'm not, but I was using the search to basically look some of
- 20:42the items and as typing in Brand Plus gender.
- 20:46And then I was looking for like the cheap Offside Brand Plus
- 20:49style. And I was getting such a myth
- 20:52representative set of product back.
- 20:54So I didn't realize to start off with and then afterwards, I was
- 20:57like, they must have more of this brand but basically by
- 21:01adding the gender, it would get tired free products, a PC men's.
- 21:06And then when I went through to the the category that was
- 21:08actually like a much broader range, but they just weren't
- 21:11serving those queries. And that's like how I shop and
- 21:13obviously, I'm not necessarily the average consumer.
- 21:15But yeah, I can this stuff's definitely obviously.
- 21:18Lee super important, but getting more and more important with the
- 21:21way, like, the average user is starting to operate it.
- 21:25I think it's getting even more complicated.
- 21:28When you have too much choice. They, you know, poor, I'm a guy
- 21:32probably shop in a very basic way, to be honest.
- 21:35But, you know, let's say you go on any of the big retail sites
- 21:39and you go type in men's white T-shirt, you know, which is
- 21:44quite specific in some ways and not too.
- 21:48But equally not too complex. I mean, actually, I'm going to
- 21:53throw it out to you. James going to pick on you Joe's
- 21:55if you did type in men's white T-shirt into a, you know, major
- 21:59retailers, e-commerce site. You know, how many do you reckon
- 22:02you get back? I know I didn't prep you for
- 22:07this one, but we go three stone yet thousands because I would
- 22:10expect there to be started with with multi colors and random
- 22:15products, where the word white is appeared when it's not a
- 22:18descriptor of color because a lot of people don't figure the
- 22:22tools properly. When they put the, we were just
- 22:24we're talking about this before the episode is sometimes.
- 22:27The issue is not tall. It's the business effort to set
- 22:29up and use the tool that correctly.
- 22:31So yeah, I depending on the size and how Well, they've actually
- 22:34prioritize resources research, I would say the big site probably
- 22:38thousands. It's interesting.
- 22:41So we, you know, Ruby starting to talk about the the Paradox of
- 22:45choice. You know, we all think more
- 22:48Choice gives us more freedom but actually it creates more
- 22:52complexity in our mind. So actually some of the
- 22:55websites, you go type in men's white T-shirt because it's not
- 22:58super complex. Actually, some of the sites are
- 23:01quite quite well optimized for it.
- 23:03But I did it on one side the other day and there's 1700
- 23:07t-shirts and actually, when you were scrolling through them,
- 23:10they're all white. They were just plain white
- 23:12t-shirts though. I think, you know, we do have
- 23:15this issue. Not only relevance dealing with
- 23:20that long tail. But also, how are we personalize
- 23:23and our search results? Because we know that people kind
- 23:26of any buy something when they see it when they met when they
- 23:29see it is when they know they do kind of want it.
- 23:31And I think in some ways I think this This is where
- 23:34personalization of search starts to kick in because there is just
- 23:37too much Choice there. We need to kind of help guide
- 23:39people, the other big ones, long sleeved, black dress, you go
- 23:43type that into many retailers and see what results you get
- 23:46back that enzyme. So in the rest you have an
- 23:49Eclectic dress taste in run. So white T-shirt in the week,
- 23:53black dress on the weekend Jay's I did so many years.
- 23:57Dare I say Arcadia and selfridges these are the typical
- 24:01things you know. Yep interview.
- 24:04So that takes me very nicely on to my next question, which is
- 24:09essentially, if you're implementing like a
- 24:11best-of-breed AI + ml solution, what do you think the team needs
- 24:15to look like to get the most out of the product?
- 24:20So, actually, the AI Market is getting more and more maturely,
- 24:24we're talking about just before about GPT, for example, any all
- 24:27of us tried to play with it. That needs very accessible and
- 24:30you can really play with the with different AI software right
- 24:33now. So so the market is quite
- 24:35mature, which means it doesn't require a any particular skills
- 24:39to use the the I platform when you are talking about
- 24:44e-commerce, of course it requires some merchandising and
- 24:47business skills because That's what you will need that.
- 24:50I will Leverage The but the power of AI and so if you rely
- 24:54on a button arrow on the solution, like attract, you
- 24:57don't need any that testing. Killer seems like that to to run
- 25:01software. That being said, we have to, we
- 25:04see two types of companies Appliance to be.
- 25:07So when we think AI is critical for their business and they want
- 25:11to massively invest in it. And the one we think it's a
- 25:15messed up, but it's more another one to observe too.
- 25:18Business strategy. The first type, we usually
- 25:21create an AI team with data scientists and the engineers,
- 25:26they will develop their own algorithm.
- 25:28So they will massively invest in AI because they think that would
- 25:32will make their company different tomorrow.
- 25:35The second type will mainly rely on best-in-class, back Partners,
- 25:38like attract 12 them to leverage a ions are critical Janet.
- 25:43So at attract, we can work with both kind of clients.
- 25:47We have different features for Zeus to kind of Appliance we we
- 25:51built, what we call the need for an algorithm orchestrator.
- 25:54So we can bust and activate some algorithms that are created by
- 25:58our more advanced science. So we all know thing to add on
- 26:03the value of human control on top of the algorithm, which is
- 26:06quite interesting because usually when they create data
- 26:08science team. So just for the data scientist
- 26:11love to work on the most the best model that we generate the
- 26:15more error I anything like that. But it sometimes forget The the
- 26:19business partner is in the expertise of the merchandiser in
- 26:24it and so they don't give a lot of power to the merchandisers.
- 26:28We will manually curate them product sometimes but also
- 26:32Define what is the strategy of the business?
- 26:35And so we know that sin to add that layer of human control on
- 26:41top of what is automated by AI. So that's how the first kind of
- 26:45planes which are very Advanced and the other kind of clients.
- 26:48Liable, not going to library with some probability
- 26:52algorithms. You just need to pick the one,
- 26:54the world and play with it. There's them excetera, but there
- 26:57is no nursing technical nursing mathematical, and they'll
- 27:00totally, it's accessible to anyone.
- 27:04Hypersonics helps e-commerce companies make more profit every
- 27:07day. This AI driven platform delivers
- 27:11recommendations for pricing and inventory that lead to bigger
- 27:14profits. Visit hypersonics, dot AI
- 27:18forward, slash podcast to get a free trial, that makes sense.
- 27:24And I guess the next one is quite an interesting one because
- 27:27I definitely wouldn't put attract in this bracket.
- 27:29You obviously wrote with a lot of the biggest brands in the
- 27:31world, but I've seen - you know, Implement more of a generic AI
- 27:37solution and then you'll see the kind of way that they attribute
- 27:41success. And they'll just essentially
- 27:43take anything that's gone through that, you know, layer of
- 27:47Automation and the tribute of Revenue directly to it, but I
- 27:50guess the key is looking at incrementality and the impact
- 27:54the automation or AI has had on. Kind of all the uplift has been
- 28:00seen as a result of that. How do you hold on?
- 28:03Merchants typically, look at this and how to attract help
- 28:06users analyze sous chef So that's a very good question in.
- 28:11And to be honest, the market with many, many providers,
- 28:16sometimes use attribution as you said, so they will attribute
- 28:20everything that goes through recommendation or through search
- 28:23to the platform and of course, it is that for marketing reason,
- 28:26because it's a great boost to it.
- 28:28So I drill bit the maximum number of purchases, but we all
- 28:32know that. It's not really true, it's a
- 28:34real epic, it's just an attribution system.
- 28:37Attribution, systems are good. But for me, More tool to monitor
- 28:41the performance, when you see a drug or a sedan grow in your
- 28:46attribution, it means something changed.
- 28:48And it's interesting to have such kind of overwhelming, but
- 28:51if you really want to measure the how it performs, the only
- 28:56way to do that is through an a/b test attribution, doesn't help
- 29:00to read to measure the real F polished personalization with
- 29:05it. So everything is about A/B test
- 29:08but a Be testing is not in you have different way to use a/b
- 29:14testing and proper AP testing needs some methodology, you
- 29:19can't test everything. At the same time you can test
- 29:22all the audiologist use case at the same time and so sometimes
- 29:25playing sour little lost when we when they see what they want to
- 29:32do some I detest so so I think it's important to provide a good
- 29:36methodology and the platform that helps the Giants to do
- 29:39things in Purple way and in the right order so that what we are
- 29:42trying to to do in to serve and important to start with to go
- 29:48back to what we were discussing at the beginning of this
- 29:51podcast, start with use cases because it's something you can
- 29:54really understand there is a strong perimeter so you know
- 29:59exactly on what you are working on.
- 30:00So start on your strategy, use cases a t-test suits, use cases
- 30:05calculate the the uplifted will generate and then you can Create
- 30:10some more advanced use cases to give diving the details of each
- 30:13use case but it's really a test in nam approach.
- 30:16Keep it simple at the beginning, work on what we generate, the
- 30:19more are arrived and and that you will be able to justify
- 30:23internally. Also to create the confidence
- 30:25stress test and in render said before, and then you can run in
- 30:30more and more AP tests to, to Really configure things in a
- 30:35very precise way. Doing that is basically what we
- 30:40create the difference between you and your competitors.
- 30:43As an e-commerce platform, everyone has access to a I right
- 30:47now. But the way you implement it,
- 30:49the way you test it, the way you deploy it, make it a bitter user
- 30:55or a live could, you know? And that's what will make the
- 30:57difference between your competitors in you.
- 30:59So that's very keen in the process.
- 31:03And of course our goal is kind of link to the conversation.
- 31:06We've had about algorithms is, what advice do you have for
- 31:10tuning? Our wounds?
- 31:11I know any class you said that people can engineering says, can
- 31:14create their own and then they can use your torment on top of
- 31:17it. All, they can take your default
- 31:19ones or they can you customize. So, what it can?
- 31:23I've seen this rep, you can get lost in algorithms and try to
- 31:26overcomplicate things rather than starting simple build up
- 31:29and learning from it. What is your advice on?
- 31:31How do you pray? How How do you chunio?
- 31:34What and what also? What are the limitations of?
- 31:36Just Ryan on one single algorithm go right.
- 31:38There we go, we've done it. Yes, so s economically, ginning.
- 31:44It's a there is different. Sounds different games of
- 31:46tuning. You have.
- 31:47If you were out talking about that their scientists, they will
- 31:50tune met him at. Akkad pyramidal seems very
- 31:52complex. We we don't really understand
- 31:54and then yet become complex, but it's always a sort of tuning
- 31:59based on business strategies. So, using an algorithm, but
- 32:04configure ET computer it in a way that makes sense for your
- 32:08business. Like, for example, if you are on
- 32:10a Doctor you are Amici brand e-commerce website.
- 32:13You are under the category page room that sells t-shirts, for
- 32:17example as a business expelled would I wanted to see to show
- 32:24the different brands I have in my shop?
- 32:26That's the kind of tuning you can do.
- 32:28So, the algorithm by itself is just push the product that have
- 32:32the most chance to be solved for this particular user.
- 32:35But as a business manager, I want to tune some some things
- 32:39because I know I know Nobody experience.
- 32:41It will create some long-term value, for example, that.
- 32:44So kind of tuning, you could expect from next business.
- 32:50Expel to merchandiser on on their website.
- 32:53I think, what we what we conscious of is different people
- 32:57different Merchants at different stages of have their maturity
- 33:00around this. They get part of our business
- 33:04has actually kind of guiding. So, given we've been doing this
- 33:07for a long time, we have a good feel.
- 33:10Sound what's the Baseline? And what good looks like and
- 33:14what kind of typical use cases and business strategies in a
- 33:18specific vertical, people are trying to tackle or kind of
- 33:21overcome resolved, Etc. So we can guide merchants in
- 33:26terms of AK you in this vertical.
- 33:28You're trying to achieve this kind of strategy or certain
- 33:32objective. Will we know just from history
- 33:35that you want to go and test this a these two algorithms
- 33:38we've got 14 but you know, go test.
- 33:40These two ones because we know that those two ones typically
- 33:43have the better result for you though, it does depend on
- 33:46maturity. But equally what we try to do is
- 33:50be there to help guide people rather than just selling the
- 33:52tool is kind of trying to guide them and guide them with the
- 33:55experiences we've had in the vertical.
- 33:57So it would look after And so I want to get back to the when I
- 34:02go reason to rule them, all you need to keep in mind that
- 34:05algorithms are created to solve the use case.
- 34:08So when algorithms that sort of, also use case it does not exist
- 34:12at least today, perhaps it will come when the Earth.
- 34:15But today, that doesn't. So the first thing to think of
- 34:18is the use case and start by using algorithms that have been
- 34:22created for social space and, and in our library, is very
- 34:26easy, you have access to you, you can access the Repeal you
- 34:30skate, for example. So you say what you want to
- 34:32sound, and then you will reduce the list of algorithms that can
- 34:35solve that particular issue. And then you will have, of
- 34:38course, different algorithms, and some algorithms will, for
- 34:42example, be based more on product data.
- 34:44We were talking about that on subject before some algorithms
- 34:47will be more focused on product data.
- 34:48Some algorithms will be more focused on user behaviors.
- 34:52So, depending on the on the, on your website, you will not
- 34:56choose exactly. The same kind of algorithms.
- 34:57For example, if you are a small business - with not a lot of
- 35:00traffic, you will probably use more product data, then user
- 35:04behavior. And if you are a large business
- 35:06International with a lot of traffic, you will probably get
- 35:09better results for and then with an algorithm that relies a lot
- 35:12on behavior. And in our platform you can
- 35:15blend that so you can use the hell.
- 35:18I go is mix algorithm that uses traffic data with algorithm that
- 35:21uses product data. So you can really get to
- 35:24something more advanced and that's part of the tuning, you
- 35:27can do as a male chauvinism, So it's a way to configure The best
- 35:33scenario for your website and and to do that properly you will
- 35:37run different A/B test and you will go step by step to the, to
- 35:40the best configuration. But again, there is nothing
- 35:42complex in that. It's a, it can be complex is a
- 35:46platform if the platform is complex.
- 35:47But if it's well organized with some examples, anything like
- 35:50that, it's greater quite straightforward.
- 35:53It's a new maid set. So merchandisers need to to
- 35:57learn in detail at the beginning, of course, but But
- 35:59it's not more complex than creating and rather from annual
- 36:02reports to run your own website. I do you think we need to carve
- 36:06as an industry starts get away from thinking you know, they
- 36:09spend their that's the best algorithm that you can go and
- 36:12choose. I think the tourniquet is point,
- 36:15is based on the use case, but also the flexibility to be able
- 36:18to go, they be test all these different algorithms or said we
- 36:22we've got a library of algorithms continue to develop
- 36:24them bed. We've always been of the my
- 36:27frame that this kind of needs to be open.
- 36:29So if you want to bring Sensei line Stein into our platform and
- 36:34perform an a/b test or you have a data science team and, you
- 36:38know, they're building their own algorithms, but they need
- 36:41somewhere to kind of manage their more test them, or even
- 36:44blend them with other algorithms, that's kind of the
- 36:46mindset. We've always had wanted to
- 36:48always be open rather than saying, these are our
- 36:51algorithms. This is your box of Tricks.
- 36:53There you go. We did believe that and it kind
- 36:56of goes against what we kind of look to do in terms of.
- 36:59Optimization experimentation? Yeah, I think that's a really
- 37:02nice point because otherwise, I mean, a lot, a lot of challenge
- 37:05to people in the past has been constrained by Black Box
- 37:08algorithms, where they're not really sure why things are being
- 37:11successful if they are. But also the the, you do get
- 37:15certain brands where they are very, very focused on their own
- 37:18manual visual merchandising, and they have a very set way of how
- 37:22they think the business should be visually presented online and
- 37:26in might not be optimal from a commercial outcome.
- 37:28Point of view in the Anyway, you can ever get a compelling
- 37:31business case, has to run that proper scientific A/B test of an
- 37:35AI driven. Algorithm is optimized of
- 37:38outcomes versus a manual brand driven, way of merchandising the
- 37:42page, and, and to demonstrate because otherwise those
- 37:44arguments become completely circular and never end.
- 37:48Yeah, yeah. And that kind of comes back to
- 37:52that trust and confidence piece because it can be a blend of as
- 37:56having Nick a cement. Should you know, you're you,
- 37:59Using maybe some elements of AI to do, do the give the term, but
- 38:04the grunt work, but you are using your merchandisers about
- 38:09half that tacit knowledge and understanding how to curate to
- 38:12been going kind of direct and guide.
- 38:15The direction of what is being created from the algorithm as it
- 38:18were. So we kind of see it as and it's
- 38:22another kind of industry thing of.
- 38:24It's one or the other. It's even wall manual or it's
- 38:26all been handed off to a. I we kind If date really agree
- 38:30with that, we think there's actually a place for kind of,
- 38:34there is an element of manual that's needed in some instances,
- 38:37then whereby business as usual. Actually, probably a, I could do
- 38:40it in an automated way, but there's kind of this, this
- 38:43Middle Ground where they come together.
- 38:45And I think that's kind of what we see more often is the control
- 38:48of it. They from the merchants.
- 38:51Yeah, there's a realize point to make actually.
- 38:52Is that AI is inhuman independent, because the ability
- 38:57to give it parameters to make smarter decisions at scale, but
- 39:00still maintaining the ability to there, will always be that need
- 39:04to curate stories and ensure the especially if you're a multi
- 39:06brand, you might have some commercial contractual
- 39:08obligations to not constantly push this brand.
- 39:11That seems to get stronger conversion rate, which an
- 39:14algorithm could do if you target purely on the conversion rate.
- 39:17So yeah, interesting question. I've got on tracking actually, I
- 39:20think this is really important. People, listen is the, I think
- 39:24the last few years have been lowered to talk about the
- 39:26evolution to Cookie list world. And the issue with party could
- 39:29Cookies and challenges reset up. First-party Cookie tracking ID,
- 39:33you talk people through how you handle, the ability to track
- 39:37these user Journeys, and to maximize that data capture
- 39:40piece, So the first attractive are really relied only on
- 39:45cookies. Sorry, different 32 tracks users
- 39:49went. When personalization, of course
- 39:51requires with regularization, I don't identify yourself,
- 39:54something that will help us to recognize a user without the
- 39:57different page. You will see will Pros.
- 40:00Most of it the time each cookie base because that's the simple
- 40:03way to do. It will become less and less be
- 40:05very good device. With the as of today.
- 40:07See, it's still working on many, many device.
- 40:11And but it can also be leveraged.
- 40:13It can also leverage other technical features like local
- 40:16storage fingerprinting. We can also rely on the add-on
- 40:20ideas that I provided by ad Network, like Google Facebook
- 40:24site provides our own ideas to to track users, we can rely on
- 40:27that. And of course, when the user is
- 40:30logged in it way easier, we have a access to their account, IDs a
- 40:34loyalty card. We can, we can also use the
- 40:37credit card fingerprint. So there is, there are different
- 40:39ways to to, to be able to understand that this user
- 40:43object, this specific thing on the website.
- 40:46Sometime, you don't have access to any user egg because the user
- 40:50refused all the cookie. You don't want to share any
- 40:53data. So you can't really do 1:01 to
- 40:56and personalization because it's not what the I won't end up
- 40:59rolling, we said about the fur is a trust that we build with
- 41:03the end user. So what you can do in that when
- 41:06we need abandoned, we can still personalized.
- 41:08But you know, when to believe is so we get, we will understand
- 41:11some, we will understand some Treads about this shoe sir.
- 41:17So we will personalize to Zeus thread but not freely to these
- 41:20shoes are in particular and so that's a full back.
- 41:25When we cannot win the user, don't want to share it.
- 41:28Enough data so. So we can person that I think.
- 41:31Also, I want to highlight that it's really what is really
- 41:34important is to create a relationship between the user
- 41:39Shopper and the brand. And so explaining why you need
- 41:44to track sus data absolute. And usually what we see what we
- 41:48see is that users are ready to to share data for
- 41:51personalization. They are less ready to share
- 41:53data for ads. So explaining really, what is
- 41:57the benefit for them? Them to share the data.
- 41:59And we can do that in the cookie policy because they can really
- 42:02check and uncheck exactly what they want to to to, to use and
- 42:07why they are ready to share with their data.
- 42:09If it's well, explained, we can, we can get some, some, very
- 42:12interesting data from the user. This is the, like
- 42:15personalization, like they are using Netflix. 35.
- 42:17Exactly like to reach our personalized.
- 42:21Yeah, register like constant adds to that.
- 42:23Yeah, exactly. I'm great and then move it onto
- 42:29the day kind of implementation of performance side.
- 42:33So a lot of people, I think the user tracked would go down the
- 42:36client-side, kind of, you know, JavaScript front-end
- 42:40implementation route, you know which can have an impact on
- 42:43performance. And then equally the server side
- 42:45route can be a lot more complex and you know achieving some of
- 42:49the personalization side is a lot more complex.
- 42:52What's your advice? The the best implementation
- 42:55Russian route. Yeah.
- 42:57And how did how did your typical implement?
- 43:00So, sh inside the receiver. Basically, you have a when
- 43:04implementation that these Quick N Easy, but but less Bravo mon
- 43:08so, is the guy inside implementation.
- 43:10And then you have the one, which is more future-proof.
- 43:13So, our advice is to start with the plain side implementation,
- 43:17because it's very easy to run, a t-test on the server site owner,
- 43:21glance and implementation. It's quick to deploy.
- 43:24So you can start validating the array and, and justifying so
- 43:29investment, it will require To go to several side
- 43:32implementation. But at the end, what we see with
- 43:35all our clients, the one who moved from a client-side
- 43:38implementation to a server-side implementation.
- 43:41So they are right, increased basically, many 4p of
- 43:45performance reasons. And as you know, this is a speed
- 43:49display speed on internet is key.
- 43:51You can really improve your area just by accelerating the, so the
- 43:56display time of the beige and so that something you can do with
- 43:59the, we service. Siren blaring.
- 44:01So from a technical point of view server-side rendering is
- 44:04definitely bitter, more future-proof more performant but
- 44:08from a business point of view, sometimes you want to validate
- 44:12things with a with a quick and dirty but we get the further
- 44:16Greek way of interpreting things and then move to something more
- 44:20and more technical Absolutely. And and then the last question
- 44:25so when it comes to be to be, I guess on paper, it should be
- 44:30easier when it comes to product Discovery because, you know,
- 44:33most users are logged in and you have a lot more day here on the
- 44:36customer. And you know, there's more
- 44:38chance of them being a return customer etcetera.
- 44:42But equally you know, there's there's usually a lot more
- 44:46complexity around B2B and do you typically work with B2B
- 44:50merchants and kind of what advice?
- 44:52You have for B2B. So if you need from Port is
- 44:58historically, not being a key, vertical course.
- 45:01Our growth is generally being in in fashion and the vertical
- 45:05spoke of before but we have got a number of merchants that are
- 45:10B2B It is a different challenge. Yeah, as you say so you
- 45:17typically have less a custom pricing custom catalogs.
- 45:21Custom buying approvals that all has an effect on what products
- 45:25should be surface to certain individuals in a buyer
- 45:28organization. So that obviously has an impact
- 45:31to product Discovery. Said it, you know, it is
- 45:35possible, it just needs a bit more effort to understand kind
- 45:39of their text at their processes and then cover the right
- 45:41approach as to how How we go and deliver.
- 45:44But what's interesting is that be to be, you know, those
- 45:48catalogs are quite quite large quite wide.
- 45:51Deep, complex buyers typically will use Park, codes, those Park
- 45:58codes, also change. So old part gets part codes, get
- 46:02replaced by new ones. And it's kind of, well, how do
- 46:04you make that connection? Because someone might be only
- 46:07familiar with the old part code, rather than the new one.
- 46:10We also see that a little bit like Grace, Real, where you've
- 46:13got these major catalogs. That search is kind of the
- 46:17primary activity to go, go find, or discover something in B2B
- 46:23it's usually based specific terms.
- 46:25So we see a bit more of an elongated kind of long tail.
- 46:30So, you know, if you think about say someone, but it's like a
- 46:33screw fix, but in the trade world, you know, waterproof,
- 46:37electric cable connectors, you know, it's pretty specific.
- 46:40You know, and pretty complex through is to constantly.
- 46:43Optimizing the search results off the back of that and that's
- 46:46kind of where we do. See a I could could help help
- 46:51those folks you know, because it's even more complex than some
- 46:55of the other verticals. It's bigger.
- 46:58It's more complex is more volume and that's kind of where I
- 47:01really could help. Help help with some of those
- 47:03tasks for sure. Nick, I did it.
- 47:06Did I cover that? Okay.
- 47:08Or is there anything else you wanted to add on that?
- 47:10I know you've had a few engagements and B2B as well.
- 47:12Yeah, so though I agree with everything you said.
- 47:15I think it which is important is when you are a B2B business,
- 47:19there are some features you you need to under personalization
- 47:23and we work with B2B business since for 45 years.
- 47:28And there are there are things like pricing catalog.
- 47:31As Imran said you will you can create different pricing catalog
- 47:35for each. And that's the kind of things we
- 47:37undone naturally. And so, so there are some basic
- 47:40features you can see, a not, not basic technically, but the like
- 47:45you did by order of business. And, and that's the kind of
- 47:49things you get, you, we have an actively and that's the
- 47:54challenge we have. So, the in term of
- 47:57merchandising, the way, you will approach, merchandising for B2B,
- 48:00and b2c is very different on a B2B business.
- 48:03Usually, you will really match. Sunday's your website based on
- 48:08what you want to display for this specific client.
- 48:10So you will do a kind of manual personalization because you will
- 48:14you personalize it for 100 different clients or more but
- 48:18definitely try to, to automate things that you have.
- 48:21Some key clients, where you really want to configure
- 48:24manually house, they will see the, their each PLP based on
- 48:28what, you know, they usually buy or something like that.
- 48:31So that's the kind of stuff that is a, that can be done quite
- 48:34easily. Now in operation.
- 48:35It from to. So, as someone said, it's not
- 48:38all men vertical. Definitely that not real, you
- 48:40will find the more planes in our innovation in our paint base.
- 48:44But but we have a couple of B2B clients and and we build some
- 48:48features dedicated to them Thanks fella.
- 48:53We've covered a lot of really interesting topics today.
- 48:56If inevitably some people will have questions on some of the
- 49:00things we covered or might want to probe in a bit heart but hard
- 49:04about how they would really integrate that and what some of
- 49:06the challenges are implementations, how do they
- 49:08reach out? Yeah, so I mean more than happy
- 49:15for them to reach out to me directly, obviously we've got
- 49:19contact forms kind of on the website.
- 49:21So, you know, more than welcome to reach out, maybe just put in
- 49:25the title reference this podcast and then that will kind of get
- 49:29get sent to me. But yeah, I think just just
- 49:33reach out through the usual channels, you know, Nikolas and
- 49:35I are on need then Etc. But you if you put reference
- 49:40this podcast, then contact us for more.
- 49:42On our website, you'll get to a straight.
- 49:44And hopefully, from memory our marketing team of create a few
- 49:48ctOS with some of the reports I mentioned earlier as well.
- 49:51Joan. So hopefully that helps with
- 49:53people. So yeah, in my necklace, really
- 49:56appreciate you taking the time really enjoyed it.
- 49:58I love products Discovery. I think it's such an important
- 50:01area and no site gets 100%. Right.
- 50:04We will know it. As well as opportunities to
- 50:06improve whether it's merchandising search, will never
- 50:09say that appreciate you taking the time and thanks very much
- 50:11for listening. And do reach out to demand
- 50:14because if you've got more questions, keep an ear open for
- 50:17next episode, we drop them every week.
- 50:18And do give us that rating on Apple Spotify or YouTube, or
- 50:21wherever you. Listen.
- 50:22Thanks very much. Everybody more information on
- 50:31this topic, head over to re-platform dot.
- 50:33F m, for our audio podcasts to discuss a project or if you'd
- 50:37like to chat about any of the topics covered in this episode
- 50:39in more detail. Please reach out to myself,
- 50:41James guard and My co-host, pull Rogers, via LinkedIn and
- 50:44Twitter. Thanks again for listening and
- 50:45keep your ears peeled for the next episode.