Powerful machine learning data strategies from The Very Group

we're really delighted to have steve plett, who is the chief data officer from the very group who will be sharing his own story about the incredible success and revolutionary breakthroughs that the very group is experiencing by leveraging data and by really looking at modern technologies, including things like a i.

I'll talk about this topic for a few moments and get it over to uh to steve and then we'll have some conversation. I may go a little bit quick. uh but uh please make sure if you have any follow up that you want to reach out to me, we'll be giving you contact information at the end.

This is a critical moment but a very exciting moment in the retail industry because this is the inflection of three important big trends. Two of them as aws uh partners and friends, you know, well, which is the massive increase in compute capability that the cloud offers as well as the explosion. the fire hose of data structured unstructured data from an incredibly infinite number of sources.

These two ongoing trends that we've known for a number of years are now intersecting with a new one which is the sophistication of models including ll ms leading to things like generative a i. This is a very exciting time.

As you're probably aware, amazon has always been a leader as it relates to a i and retail, we sell literally thousands of products every minute packages flying around everywhere and don't forget about alexa, the language models used in alexa. you know, i once heard someone say that you don't hear with your ears, you hear with your brain, alexa is really an application, an early application of advanced a i. and it's really the kind of thing that we've focused on over the last few years.

So if you think about the cascade and the evolution of a i, we started off with the notions of artificial intelligence and sort of the turing test type of things. but ultimately, we got to machine learning which focused on the search of patterns, search for patterns, something very familiar for anyone who's had to deal with thousands and thousands of iterations of consumers every minute. as all of you do about a decade ago, amazon started really getting into what we call deep learning, which is a further subset of ml, which uses layered neural networks and our accuracy just went off the charts.

But the next horizon, the next frontier is actually generative a i. and we're going to talk a bit about this, which is the idea that we use these large models against those giant volumes of data to really change the world with what we call foundational models.

There's a whole bunch of use cases and this is like, you know, just scratching the surface. but just remember that anything that requires customer insight, operational insight, large amounts of data, there's a wide variety of generative a i use cases that you can really apply. and uh in this short period, we're going to talk about a few use cases, steve scott that uh the very group has been experimenting with, but just rest assured that this is the beginning of the application of gen a i really pervasively across retail.

For those of you who have not experienced our just walk out technology, the amazon contactless uh technology that allows you to basically take something and just walk out there happens to be one at the las vegas airport at the hudson nonstop as you're boarding your plane, please try it. it's an incredible experience. and what we'll find is that these kinds of experiences will draw us to things like uh generative a i because our customers will demand these kind of experiences, autonomous retail.

I mean, you think it's about computer vision and sensors and all this stuff, but really those are just inputs, inputs into a really sophisticated a i based uh sort of model that allows us to deal with the complexity of people picking up items from all over the store, maybe putting them back and, and getting the right price and the right item count.

If you go over to that hudson's uh nonstop, you'll find that amazon one is there. that's the palm reader, the biometric palm reader that allows you to actually have the entire checkout experience without even reaching for your wallet. customers love this experience and i think it's a key takeaway that it's not going to be a bright shiny object that gets you to gen a i and other advanced technologies. it's gonna be your customers who are gonna demand that you use these uh kind of technologies to really enrich their shopping experience.

We're using a lot of video for example. and people have used video in stores for as long as there's been video. the difference is the application of intelligence. what can i learn from this video? can i learn dwell times or pads to purchase? can i understand what kind of shopper marketing tools worked or when an associate talked to a consumer? you know what happened again, the inputs of video have been here forever. but the intelligence to be able to make sense of it is what's changing now, how do you, how can you use this kind of information?

Well, please, you know, join us uh you know, in the various places you can meet aws retail or if you happen to be in new york, jan in january for the national retail federation store show, we can actually share with you dashboards and uh kits that allow you to use aws solutions to act aws services to actually build your own behavioral data and analytics.

Uh and also we have a wide variety of partners. so at the end of the day, we are a very technical group at re invent, but we also focus on the fact that it all has to lead to great business outcomes. that's why we're here is to really make our businesses grow, but more fundamentally to serve our customers better. and some of the things that you can do better with these technologies, things like omni channel personalization to actually acquire and retain customers, um better engagement being proactive thinking, what that next best activity would be to really delight your customer. and then on the operational side, everyone knows that, you know, retail is rife with inefficiency and where are the places that we can improve store operations and improve, improve supply chain.

And finally, you know, this brave new world of discoverability, you know, we've been so used to search being what it is for the last, you know, 2025 years. but as you get to these generative a i models, for example, that start to really understand and can search and delve with a much richer experience. consumers will ask us for that. and so you want to be ready.

Personalization is a very specific area that we focus at amazon a lot, the system itself can be large and costly. we've spent a lot on it in amazon retail, but going forward with uh generative a i, the opportunity is here uh for you to actually leverage what amounts to be a partner and intellectually uh capable partner that happens to be a computer to help you build these new solutions. and i think that uh when steve and i have a chance to talk, we'll talk about how you can actually do that.

Of course, as you know, personalization is very near and dear to amazon. uh the recommendation engine was one of the most important first applications of a i that we had at amazon, we continue to invest in it and we recommend that our partners and customers do the same because it's really about where customers want us to go. they want the market of one, they want us to know them intimately to predict the next best activity and to really get to a world where we're anticipating their asks and there's no way to do this without personalization.

So at aws retail, we engage with thousands of retailers and brands worldwide, some of the largest brands in the world. and we've been compiling a list of those questions about uh use cases that uh our customers have been asking about the home page experience in digital is the entry point to their experience, the customer's experience with you. so that's getting a lot of attention. some of the obvious things like basket size and order value and cross sell and upsell these are things that the business needs to focus on the other points that are on this chart. you may say, well, we do this every day. so why are we calling it out? the key thing is that amazon, if you're here, you've heard this phrase before. we always work backwards, work backwards from the outcome we're trying to get to. and this is the opportunity i think for us to discuss what are the outcomes and for you to tell us where should amazon and aws invest in technologies and capabilities to help you better achieve outcomes like this.

We talked about this sort of evolution of forecasting and we've lived this, we like to say aws was born in retail and built for retailers and we have the scars to prove it because, you know, there was a time, you know, in a garage in jeff bezos' house when we were doing this stuff on spreadsheets the investment in data and ultimately uh artificial intelligence, deep learning. and uh at the end in generative a i has been very uh critical foundational to our success.

So we have a couple of key platforms or solutions that uh services that you should be aware of uh forecast. if you haven't seen, it gives the ability to take models and capabilities for time series forecasting for a wide variety of use cases without having to have an army of data scientists. similarly sage maker, which is our ability, sage maker canvas, which is our ability to leverage our underlying foundational model models in a way that doesn't require deep data science. we really recommend you take a look at.

So with that, uh i do want to introduce steve uh for those of you who aren't aware, very group is one of the leading uh commerce players in uh europe based in the uk. they've been a great aws partner for many years and steve spoken at our events before i really want to uh give him the time to share with you their success and then we'll have a few moments to chat after.

So with that, let me turn over to uh steve from the very group. this v thanks, rob. good afternoon, everybody.

Yes. So my name is steve ple. I'm chief data officer at the very group and the lights are bright, but I was going to show our hand anybody from the uk in, we've got few, a few people in.

So, um well, before I explain the very group, um I'm just going to tell you a few things about myself. Um my favorite book, something called super crunches written in 2007. Um you know, all about how data is going to change the world. Um my favorite film moneyball. Yeah. How data change baseball forever. Not just because of brad pitt. My wife does love brad pitt though. Um my favorite pastime is actually gambling, but not in vegas on a sports book called bet fair in the uk. So very much the mathematics of gambling really interests me and i even met my wife on match.com. So I'm all about being data driven um in my life and in work.

Um so the very group was really a unique opportunity to take my passion about data and apply it in business. So who are the very group? Um well, today, um we are the largest pure play online digital retailer uk. Um so we offer and we sell everything other than grocery. So all your leading brands. So 2000 brands from electrics, fashion, sports, beauty, home, garden, furniture, and everything we stock and we sell it purely online as a pure digital player. And we also couple that with really convenient ways to pay. So we call that very pay um allowing our customers to spread the cost to buy. Now pay later to take three and to really manage um that as a way of conveniently paying for all the items that they'll, you know, pay through um the very um websites.

Um but that's not where we started. So we've got a very long history. So the very group actually dates back to 1896. So it was one of the first uh uh catalogs in europe and it was called keys and co um and if you think of um really back to that time, it was a printed catalog that was sent out once a year, maybe seasonal as we got into the 19 hundreds. Um but it was a really flat static assortment. Um and you really only communicated with the customer and the consumer once a year. So static, static consumers, static um communication um and sort of one size fits all.

Um so that was back in 1896. Um let's fast forward to very today, but actually what happened over the, you know, 100 plus years, um, was really transformation. So in about 1930 cases became little woods and anywhere that lived in, in the uk, probably in the nineties will have known the little woods catalog, which was one of the biggest in the uk. Um it also became a stores business so very much in many major high streets in the uk. There was a little wards or an index where people would go and order in store and collecting store. Um and very actually is the new proposition was about 2009 and now all the stores are gone, that would still exists as an online brand, but everything is consolidated into the online proposition called very so very.co.uk and very dot ie for our irish business.

So let's fast forward to today and think about how dynamic our customer base is compared to back in 1896. So the customer base of today, um, is really diverse. Uh they're always on our assortment is now hundreds of thousands of products. They um unbelievably the spa and actually we're in our peak period at the moment, 90% of all sales are done through a mobile, be it mobile web or on or on native mobile app

So, you know, you think of how many consumers in the UK - about 45 million of them - using the mobile and apps every day. Um you know that they're time poor, lots of juggling and you know, really interfacing with us as a brand via that mobile device. So I guess to make sure we give them an amazing experience, it's really about offering them ease - the ability to find the products that they want, um choice - so hundreds of thousands of brands, you got to have a great search for that. Uh and the ability to be able to pay um really easily and conveniently all on a mobile device and of course get it distributed.

So um we're really proud that we can, you can place an order on very uh 10pm and have it delivered next day. Um so really, you know, you think of the dynamic nature of our business now, um you know, it's unparalleled in terms of the consumer expectations on that small device. Um really want and you know, a great retail experience.

So how do we think about it and frame it? So we very much know that in order to compete, we say we want to understand the shopper's DNA. Um because we know we really need to understand them to make everything more relevant, more timely and more personalized. And I'm going to bring that to life for you in a second. And really how we think about that through the power of data and analytics.

So a few more facts and stats, so you can see the, the playground of data that we've got to leverage. So we get a couple of million daily visits to our website and mobile apps. Um we've got about 4.5 million active paying customers. They can pay by cash or they can pay through the, the very credit facility. Um we've got every digital media, touch point, imaginable off site media, emails, s ms push notifications.

Um we've got a call center where um we'll both leverage chatbot capability and also be speaking to the customers. Um in terms of our credit offering, we're actually outside of the leading banks in the UK, one of the biggest lenders in the UK. So we're a bank as well. So we're fully regulated FCA compliant. Um and we've got a full supply chain. So we deliver almost 50 million um products annually um through a fully automated warehouse and supply chain um from all over the world.

So we take that data and we really think about how do we think about the customer with all those touch points, bring it together and enable us to um really innovate around the consumer based upon what we know about them, their DNA if you will.

So in order to leverage that we very much um leverage our data strategy, which is actually four pillars, three of which are on the screen. So data insight action is how we really frame the opportunity to leverage that amazing data asset, push it through our uh descriptive and predictive analytics engines and really drive action across a multitude of touch points and, and business lines within the Very group.

So it's all about being customer centric, it's all about being relevant, timely and personalized. Um and there's some really good groupings there for how we think about our action systems and the point where we make decisions that you know, can influence and support the customer.

Um so the first place is um is marketing and digital customer experience. So I've already said, you know, uh from a marketing perspective, you know, um to attract brand new customers and keep our current customers engaged. You have to be present on all the, all the necessary channels. So you've got to be in TV, press radio, digital touch point, Facebook, um um and all the various uh digital media spaces, Googles everywhere. You have to make sure that you're showing up where the customer wants you and compete in terms of um every Google search. I want a new TV. I want a Dyson, I want a new Apple um pad, whatever it may be, we have to show up in the right place. So we very much use our data and insight for customer centric marketing.

Um we also leverage our uh our data and insight for what we call a digital customer experience. So those millions of visits every single day to our um uh websites, we take our customer understanding and really, you know, drive relevance through better search through better recommendations through better product placements.

Next on the list, in terms of action systems is what we call customer centric decision in. So, um how do we make sure we support our customers through, you know, check, checking and offering them credit, reducing fraud and, and stopping bad actors. Um right through to retail, um things like demand forecasting, price elasticity, promotional effectiveness. So we very much again, take our data and insight and power those action systems and last, but not least all things that are really virtual assistance, um customer facing bots and really offering, you know, true service to the customer.

Um you know, we get a lot of questions about financials, we get a lot of questions about um delivery returns. Um so we really need to have, you know, a great customer service and we pride ourselves on really understanding net promoter score and driving that.

So let's talk more about how we're structured to leverage data and insight. And then i'm gonna talk about the technology that underpins it.

So in terms of um our business model, so as, as a chief data officer, the operating model is very much a hub and spoke model, the hubs are sense of centers of excellence that master their skills and master their crafts. And also um really think about the platforms that we're going to innovate with.

Um that's um and the spokes are really the business units, the performance unit, the vertical businesses. So in terms of hubs, i operate um five different hubs. So I've got a data platform and engineering capability very much, you know, hugely technical, think about how we scale our infrastructure, how we scale our warehouses or lake houses and, and really think about that, you know, the data foundations, um b i and analytics where we're thinking about, you know, descriptive analytics, dashboarding and i i have analysts embedded across the organization to support them with both ad hoc and, and really um um advanced analytics.

I also have what we call the customer forensics team where we've brought together um both quant and qual understanding of our customer, single customer view attitude and all research. Um and everything we know about the customer in a single hub.

Um we've got a data science team which is now a iml deep learning that um really think about how we're going to use, you know, real leading edge technologies and approaches to solve business problems. And last but not least, i also run data governance and operations thinking about privacy, security, quality governance, risk ethics.

So that's my, that's my hubs and in terms of the spokes, those are the vertical business units that we offer are really um platform services and products in support of that business unit.

So I already covered some examples. So let's help marketing with econometrics. So they know how to spend, you know, their next marketing dollar. Let's help retail with product price promotion. Let's help our financial services with fraud, credit checking. Let's help finance with better forecasting and cost management.

So we take our hubs, we really partner with the businesses and we think of ourselves as really like an internal agency deploying capability um platforms, products and services to allow the business units to innovate and scale with data and analytics.

So let's bring that to life further and really frame it around the customer. Um so like many b to c businesses, we're very customer centric. We really want to think about the customer across the entire life cycle. So these are just a few examples with every bubble really being some form of data and analytics capability that we've deployed.

So, um you know, how do we reach our customers through to, you know, acquire them, landing them on the website, converting them into a buying customer, engaging them through crm, keeping them active, understanding, churn and you know, hopefully, um keeping as many of them in that um loyal funnel as we can.

So just a few examples of very basic things, but drive tremendous value. Um so we've got one on there which will be a data process. So a price drop. So we know how many people visited the website yesterday, how many of them were looking at this fantastic shirt um that nobody actually wanted to buy. So we discounted it the day after we know that as a price drop and we send an email out that's really triggered because we know who's shopping, the look and we can, you know, do a very, very simple um data process that drives value for the business and the and the customer. So that's, you know, a great example.

Um we might have a statistical modeling capability such as churn prediction. So, you know, what, what, what are our customers doing? How long are they going to stay? Have they been engaged? When's the last time they engage with our communication or website? When's the last time they logged in and try to predict um when somebody might maybe looking to lapse. So we can, you know, hopefully, um get give them an offer, give them a better service and, and keep them, you know, um and reactivate them within the base so that they might be a good example.

And then more and more we're moving into more machine learning and deep learning algorithms. So um email recommendations, you know, for, for any retail or offering, you know, next best product, next, best action via your, you know, many channels such as email.

So we've got a good example. Uh what i'm actually going to do is double click and zoom in on demand forecasting. So that's just one of the bubbles to to drill down.

Um and the next two or three slides are definitely my vegas show slides to get ready to be wild.

Um so, um i'm just going to try to really bring to life forecasting. So if you think of Very being this multi category retailer, we have at any one time, you know, hundreds of thousands of skews, so stocking keeping units, um we've got, i've already mentioned, you know, 2000 brands, we offer every product imaginable that you can, you know, purchase for your house, your pets, your kids, your clothes, your makeup, your furniture tv. s anything with a plug.

So a huge assortment and a huge range and we've got 4 million customers. So one of the biggest challenges we've got as a company is really understanding the demand for every individual unit, color, fit and size and being able to buy the right stock for our customers.

Um so forecasting that demand. So how do we do it? So if you think of really each branch on here being something that we got to predict i'm gonna bring it to life with just, you know, um an example on here that was watches.

So, um we've got many brands. This one is Calvin Klein watches, you know, and we've got different time pieces here. So each stock and keeping unit, we need to know how many to purchase, what color, what strap, what face, et cetera. And that's one product within our, you know, 172 100,000 product lines.

So what we did was very much started to think about a forecasting engine. So using many of the technologies Ravi mentioned before, so AWS offer a forecast surface which is really good for, you know, uh predicting um and leveraging that at maybe a category level or a range level. And Amazon SageMaker who can build our own models and, and do a champion challenger and, and really, you know, predict every line.

So a few of the facts and stats that we've used um and we're on a continuous journey really to harness um machine learning technologies and techniques.

So, um we have got about 90 buyers or end users that really are the ones that we're trying to support with a recommendation of how much stock to buy and or replenish. So we run um twice weekly, a brand new forecast for every individual skew and product that roll rolls up into a category forecast.

Um we really look at the, you know, key metrics such as the model accuracy, weep um the availability, how much we're going to put in front of the cook, how much we've got in stock? Uh and also one of the big things we think about is actually the end user adoption. Ie the 90 people. Do they understand the model? Is it explainable? Are they gonna um use it when it comes to pressing the buy button or are they going to override it and trust their gut?

Um we also make sure that we're leveraging more and more data sources. So we're looking at um demand, we're looking at stock, we're looking at marketing spend, we're looking at weather data, we're looking at seasonality. We're looking at, you know, different curves based upon the category type.

We're looking at market intelligence, we're looking at competitor information to really think about the data sources that can help us be, you know, sharper and more accurate and really get a competitive advantage. Um if we get it right, that means for our customers, we're going to be in stock, it's going to be on the shelf. There's nothing worse than going to a retailer searching for a product and it's saying out of stock. So that's if we get it right, that's what we get. So we get, you know, customer um really relevance, timeliness and a great experience for them.

If we get it wrong, we've got a couple of problems. Problem number one is it wasn't in stock. Um in that example, problem number two is actually we might have bought too much stock. So that becomes a real big cash problem, um, for any business that's bought too much stock and now you can't sell it. So, you know, it's a fine line to make sure you've got that accuracy and really, you know, think about being customer centric but understanding, you know, cash flow and how you're going to run your business against that scale.

And if you think about it, we've got, you know, 230,000 skews that we're trying to match against 4.5 million customers and make sure making the right prediction, we can get an amazing customer experience, great net promoter score and really a business model that drives, you know, great revenues and great margin. So that's the challenge.

Um and we, you know, we're making great strides on that. How did we do it? Um so we very much uh partnered with, with AWS. Um so we actually set ourselves a pilot initially. So we um and we very much use the hub and spoke model. So we took um my hub teams. So, you know, a cross functional team of engineers, of analysts and data science from my hub. So that's the DNA team.

Um we partnered with the retail spoke, we brought them into the project. So the buyers, the merchandisers and the planners um and we bought AWS in with there at the time, Forecast experience, Sage Maker experience. And we co created a pilot from end to end over a 12 week period.

Um with the ambition to show the business how much more accurate we could be with our predictions. Um you know, an individual uh product line, not only did we want to prove that we want to prove um the ability to tailor it and tune it and for it to get better. So that was another dynamic of the project.

Um what did we move from? We moved from, you know, some sophistication because we had some statistical modeling but it wasn't machine learning. So a bit of R and some time series analysis and now we've moved much more forward into a lot more algorithms and a champion challenger environment as well where we're able um for a given category or a given range or actually a given product to use different models on different behaviors and deploy them into the environment to get, you know, ever optimized uh forecasts of demand.

Um where does that show up? It shows up for those 90 buyers in a replenishment app where ultimately they're making the buy decision um to, you know, order the stock from our suppliers. Um more and more sharing our forecasts with our suppliers, those brands. So they can understand the performance on very and you know how to interrelate with um you know, a supply chain timing.

Um and sharing more and more information. So, you know, a huge success end to end and very much grounded in that data insight and action framework and very much grounded in really leveraging the capability within um Amazon um across um AWS Forecast and AWS Age Maker where, where we build our own, um it doesn't stop there because we're always looking to improve the accuracy.

Um so, uh again, anyone that works in retail, you know, will know new products without any history. So if you're in fast fashion, that's really difficult. Um so if you've never sold the product before, how are you going to forecast demands? That's, you know, problem.

What about really seasonal um um product ranges and shapes. What about intermittent demand? What about erratic and noisy? There's so many different problems to solve, especially across different categories. So, um you know, swimwear, seasonal, um Christmas sales versus a pair of jeans that you're going to sell all year. So, you know, lots of different dynamics that we can engineer into our forecasting to improve the accuracy, you know, to really get a better customer and business outcome in terms of um as it worked.

Um well, it has. So the very group has seen um growth in their sales uh in a really tough market um and an increase in market share. Um we've also reduced our operating costs um as we, you know, better leverage and better stock and better supply chain excellence, um probably we're most proud of our net promoter score, which is getting better year on year and not just because of this one example, but we also are looking at, you know, prices and price elasticity.

We're looking at um um uh promotional effectiveness and which promotions to offer, which offer we're looking at delivery charges. And we're looking at loyalty programs where so, you know, we really doubled down on understanding the customer driving that promote a score and leverage the understanding through our data, insight and action framework.

And finally, um you know, i haven't talked much about the team. So my team, we branded ourselves DNA. So year on year we do a AAA survey and engagement survey. And year on year, my team say they're, you know, really engaged at very why we, we're really working on, you know, advanced problems.

So it's great being a technical team to be able to get your hands on, you know, new technologies, but very much, i do believe the hub and spoke model where we core with the business. And in this example, with AWS, when we're all engaged, working on the same problem, common goals, common outcomes and really innovating and you know, learning as we go failing fast and trial and has been a big part of that engagement score and also our innovation as a team is up.

Um and anybody that lives in the UK might have heard a Data IQ which is one of the biggest data networks in the UK. And we were shortlisted for six awards um in, in the last year um for a lot of the work that we're doing not just in the topic of today, which was really advanced analytics, but what we of our descriptive analytics, what we're doing with our lake house technologies, how we are, you know, bringing self serve and to the whole company, how we're thinking about data products, how we're thinking about ESG you know, so many different problems that we could solve with data and analytics in a, in a retail company.

So it's, you know, really exciting um which leads on to really um look into the future. So um we very much um are a customer centric um organization, really, you know, looking to help UK families um get more out of life playing a bigger role in that. And also thinking about our assortment or range or offer and you know, how we can support the, the, the families, especially um when, um you know, there's a um challenges in terms of everybody's got, you know, petrol prices going up, you know, rental going up, how can we support the customer through that?

Um we really think there's an opportunity um you know, as Ravi slide showed um to, to really think about how we scale AI and the use cases across the whole organization. And we've, you know, set ourselves a target of being one of the, you know, leading AI retailers in the UK um to leverage that technology, but framed in our business and business purpose, you know, to delight customers and really offer a more relevant timely experiences and empower our colleagues, you know, to have a great job um supporting those customers.

So as we think about the future, um we've actually this week just announced a partnership um again that we're doing with AWS. So we've launched um an incubator. So really where we're starting to take um our most advanced, you know, next gen. So JAI deep learning use cases. And again, we've done the same thing. We've partnered with AWS and we're starting to see those ideas.

Um so the IDC have just done a really great analyst report on, you know, our, our, our DNA framework and some of the other use cases um that we are leveraging a lot of AWS technology, um especially as we think about JAI use cases.

Um so i think i'm going to invite Ravi back up on the stage and we're gonna probably have a little bit of chat about, you know, some, some of the newer use cases. So, thank you very much.

So you know, SEO plays a big part in our, you know, search optimization. So having a full feature rich page that you know, um can be understood and and scrollable and scrapable is, you know, a big part of our digital strategy.

So another way, I mean, if a tree fell in the forest, nobody there to hear it is in your inventory, but nobody can find it. It may as well not even be there, right?

So, I mean, as you can well imagine at Amazon being a digital retailer, we think about these things all the time, you know, we make a big investment in products and what if people can find them. So using Bedrock and Gen AI for product description seems like a really, really great place to start, but it is just the beginning, isn't it?

Where, where are you seeing some of the use cases that, you know, look interesting as you're looking into 2024 and beyond, what are the kind of areas that kind of have your wheels turning?

Yeah. So um well, I mentioned before that um we've got a, you know, a call center with agent support in um customers from everything from the financial services to delivery to processes. So we see obviously a big opportunity to leverage, you know, a better customer facing chat capability and also better knowledge base information for our agents.

So, you know, you know, advancement in that space, I'm sure is going to be a big step change for us. Um we also think when it comes to um well, the search but let me call it conversational um retail.

So, um so, you know, going past search on to, I think I'm going to a wedding this weekend with my two kids and I want, you know, an outfit or, you know, so conversational retail and how that starts to sit on top of your data assets um is, is really interesting.

Um and then really a big one when it was, it was in a great to look at amazon q this morning. Um is how do we empower our colleagues? So, um you know, understanding, you know, every category wants to self serve some information. So, you know, how do we take, you know, conversational b i, right? Um to, to empower, you know, um many of our spokes to make better decisions.

How do we um you know, um i've got 150 people in my team, many of them generating code. So put code whisper on the top of that, you know, get better performance from a team. Um so there's a, you know, huge performance productivity and, and um uh yeah, sort of colleague empowerment at play.

So it's interesting how that works when people first heard about, you know, gen a i and code whisper and ways of actually accelerating productivity of developers. The first knee jerk thought was does the d developers like this? Because it's like, oh well, you know, is this doing my job? But it turns out that it liberates them to do the really high value interesting work, you know, and we can actually use this as a way to get the scale, which is great.

And I wanted to ask you about conversational commerce because I think that's a really interesting area. Obviously, I have a 14 year old daughter. And so she's exactly at the age that Alexa started about the same time that she was starting to sort of talk. And now, I mean, that's the only way she thinks about it, right.

So I, I as a digital player, where do you see voice and interaction that way going? And how does Gen AI start really lending itself to that intimacy?

Yeah. Well, um I think it said before about um if 90% of our sales are on a mobile device, um there's, you know, you're gonna hand free in car, you can sort of see, you know, so many different channels to explore and that exploration, we see boys playing a big role as we move forward.

Um so, um we also see um um again, new propositions. So um leverage in augmented reality leveraging um you know, lots of front end capability as well. So, um I think it's just different channels and different propositions calling your API s right is how it and bringing it together, bring it together.

Yeah, you know, the thing about voice too is that it is both the sort of the source of information about your customer, they're talking to you, but it's also the target of your response because it's not a monologue, it's a dialogue. So it's really this opportunity, isn't it to get much more intimate and much more sticky with our customer relationships.

Yeah. And you know, if we think again back to being customer centric relevance, timely personalized. It fits very much that mantra.

It's funny though because when we talk to your peers in the industry frequently they'll fixate on this sort of, you know, natural language processing. It's like, well, that's been around for a while, but it's like what I said earlier, you don't really hear with your ears, you here with your brain, right? It's, it's the comprehension that matters, isn't it?

Definitely that, that 90% mobile thing really strike me though because boy, that's a fire hose of new data that we never had, right? Things like location and all sorts of things. But I like to say sometimes security is where innovation can go to die.

What kind of implications does this new fire hose of data? Married with new processing capabilities like ja, i, how do you think as a cdo about protecting and being a good steward of customer of customer information?

Well, I think I thought about data inside action is all three value creation pillars. We've got 1/4 which we call trust. So our trust pillar is very, very much, you know, thinking about privacy, thinking about security, risk, ethics, clients governance.

So I've got a in the same way of hub and spoke, I've got a hub team that obsess about that. So and support the business. So we roll out data stewardship models, data cataloging, we manage obviously gdpr compliance, eu cookie directives being a digital retailer.

So consent management, that's all wrapped into everything we do. Um I've taught a lot today about value creation. Um but, you know, that's called data as an asset. There was a risk and a liability is really that trust pillar.

So, um and you know, designing it in from, from the the target really. Um and you know, a big part of it is framing it in the customer again. So keeping our customers they are safe, you know, making sure it's secure and making sure we treat it with the respect that you would want your own data to be treated and the use cases that we're going to leverage it for.

Yeah, I think that's great. I mean, for many years, we thought about this area as being about compliance and about, you know, being and saying no, you can't do that. But now it's really about, you know, holding that trust, isn't it?

Absolutely. Yeah. And again, we, we were also as good as a bank because we're a compliance. So we've got that whole side of our business and consumer duty. And so, you know, it's a big, big part. Um, I can come back next year and present that if you want. Oh, please do, please do.

Um you know, I was curious about uh when you think about uh the various different ways that you're going to unify channels, right? Uh and I don't mean channels because I know you're primarily digital but channels of information, whether it be uh customer behavior or past history, uh information that you may have had before.

And I'd like to actually ask you about demand forecasting for a moment. You made an observation that sometimes you have to act in the absence of a historical record. What advice do you give? Especially with the emergence of new technologies like gen a i or uh uh you know, deep learning. Is it possible to do more accurate predictions in the absence of historical data?

Yeah, probably then comes back to classification of your products and you know how you are engineering your models to think about different features. So actually that um uh the the trend of the product itself can be a future for your model to, to be built on.

So um and you can also think about complementary products, you can think about products that have got similar traits that have got history. You can also think about that category that's got, you know, certain shape and a certain history.

So there's definitely lots of techniques that are that, you know, some of the more deep learning models allow you to leverage to get predictions without history. So yeah, we're definitely leveraging that.

Yeah, it used to be that we were really reliant on really having a strong taxonomy and knowing what things like other things. But now with, you know, machine learning, right, the models themselves can detect counterintuitive patterns between things and say this is like that in a way that we may not have been able to do it for you.

It's interesting that, you know, we worry about the black box, but to some extent, the machine can actually tell us something we wouldn't have seen with our own eyes.

So as you're thinking about the journey that you've gone on and you know, the i dc report is a great sort of uh benchmark that very groups kind of out, you know, ahead of most of your peer group, but nobody gets it perfect, right?

What are a couple of uh gotchas or things that if you could go back and roll back a little bit, you might have done a little bit differently a year ago or as you were designing the dna framework. What are the lessons learned that you might share with our group?

Yeah. Don't strive for perfection from the get go. So, really, you know, agile really release fast, fail fast, you know, get, get feedback loops.

So, you know, we probably originally very early in some of our, you know, advanced um approaches would let the data scientist probably get a bit too intellectual with it. Probably, you know, one perfection and, you know, trying to get the forecast accuracy really high before they show it and share it.

Then they did share it, they couldn't explain it and we were rolling it back to something simpler. Anyway, that's an interesting one. But yeah, release early pilot, early small to start with bought design with scale.

Um otherwise you'll end up in a, you know, a poc only environment. So those are the two things, you know, with these fast, but design would scale in mind, which if you choose the right platform, you'll be able to do again a bit of a cultural thing too because a lot of times we reward people for not screwing up.

And now we're sort of saying, don't worry warts and all we want to see what the possible is. So get it out there fast and let us see it. Right.

Yeah. Yeah. Well, if you're going to, you know, create value that if you've got the best model in the world, but no one's using it and you've got, you know, one regression analysis that improves the value by point point 1%. And so using it, then there's only one that creates value.

I think the best way to show people is practically, I've got a person on my team that is so excited by using, you know, sort of the foundational models that he's just churning out a whole bunch of demos just to show people in like 10% of the time that we used to take just to show it, you know, the, the practicality of it.

Yeah. Well, that's really great. Um we're getting close on time. Um let me just ask you one last sort of summing up question as you think about, uh you know, sort of the practical implications of the confluence of these three major trends, massive computing virtually at no cost, you know, in credible new disparate source of data.

Now, these language models are any predictions or anything that you'd want to throw out to the, to the audience about what they might anticipate what they could see next.

Oh, gosh, I guess disruption innovation. I think if you haven't started your journey of understanding, you might get left behind. Um that, that's definitely a view that we've got, you know, you think of retail supply chain, you know, product price promotion, conversational retail, there's, there's gonna be a lot of innovation coming.

So, you know, you want to get ahead of it. That's tremendous advice. Just get started, right. Don't wait, just get started.

Well with that. Uh I want to thank Steve uh and the Very group for presenting uh a really compelling case of how you can bring data and advanced technologies to serve a customer and improve operations. Uh and thank you all for joining us today.