Jeff Wilke | Inside Amazon’s AI-Powered eCommerce Growth | TransformX 2022

[Music] Thank you I’m thrilled to introduce our next Speaker Jeff Wilkie Jeff is Chairman and co-founder of Rebuilt Manufacturing A private company helping to bolster America’s industrial competitiveness He retired as Amazon’s CEO worldwide Consumer in February 2021 after more Than 21 years as a corporate officer Jeff’s career includes leadership Positions at applied signal Honeywell And software development at Accenture He holds an MBA and a SM from mit’s lgo Program and the BSE in chemical Engineering Sama kamalade from Princeton Jeff is an active philanthropist and a Mentor of startup Founders who identify As female and or people of color He is vice chairman of code.org and the Chairman of the advisory committee on Supply chain competitiveness Jeff also serves on the board for fix The first U.S company led by two female Co-founders to trade on the New York Stock Exchange Jeff is joined by Alexander Wen CEO and Founder at scale Alex over to you Jeff uh super excited to be here Chatting with you today Hey Alex Um awesome well uh we uh we had you at

Transfer Max last year and there we Talked about you know a myriad of topics Around Ai and uh and broader industry Which is a lot of fun but uh for this Time we wanted to talk through uh E-commerce in particular so Um you obviously ran uh Amazon consumer As CEO for many many many years and Helped build you know the largest uh E-commerce uh offering uh there is so I You know I figured to just start out With uh you know we’ll talk a lot about Ai and e-commerce broadly speaking but Um and zoom into a bunch of pieces of it But maybe first we’ll start with the Question how fundamental do you think AI Is to doing e-commerce effectively Well I Alex I think AI is fundamental to Doing anything at scale effectively I Don’t mean a scale AI I mean at you know At very large uh scale Um there are a bunch of techniques that That many of us who did things that were Operationally intensive or had achieved Some scale Um had used in the past including Operations research and a bunch of Applied math techniques but the Performance that you get from Modern AI Algorithms is just better in all in Every way it’s more efficient with Computing it uses data more effectively You get conclusions faster you know you Can grow your understanding of whatever

It is you’re trying to optimize faster So you get multiple cycles of learning It’s just it’s better in every way and I’d encourage everybody who’s you know Tentative about it to dive in Well you know what so when thinking About AI uh and and the sort of like Different maybe let’s talk through the Different sort of slots in the overall E-commerce kind of architecture where Um you know AI found value or AI found Have found itself to be useful Sure Well you know if you think about The and actually I should I should step Out we can talk about e-commerce but I Think this is true for for retail Commerce generally including in physical Stores so you you have uh you know you Have ">marketing and efforts that are About Bringing people to a retail experience Whether it’s online or in a store You have discovery of stuff that happens Sometimes outside of the store online or It happens while you’re in a session You have the the process of of you know Agreeing to buy something and and Remitting uh payment and Um and then any other information that’s Required and then you have the the Process of getting the stuff Home or to a business wherever it’s Going to be used and uh really each of

Those I found to be highly susceptible To improvement from from AI you know one Of the things that we’ve talked about Before is AI in its current form didn’t exist when You started Amazon of course those you Know won’t uh that was uh more than 20 Years ago so no AI Um and uh and and really a lot of modern Deep learning only started in Earnest You know in the early 2010s and then by Mid 2010s it was like 2015 2016 it was Starting to seem very promising and then You know all of a sudden fast forward to Now it’s it’s Um the big thing but how what was the You know so clearly you had to you had To operate and build Amazon without all That stuff for quite a bit of time Before it sort of became the the thing Maybe tell us about that Journey what You know Um what when did it become sort of a toy Versus a tool versus a sort of Foundation Yeah well let’s talk about each of those Pieces because I think they’re they’re Interesting Um start with ">marketing Um you know in the and you could go back To the sort of very beginning of uh Bidded advertising where you know you Had to decide what keywords basically to Bid on for for sponsored search and then

Where to to show display ads Um and then if you were operating in the Physical world you might have to decide Where to put Billboards and all of that Was done with kind of rule-based Heuristics in the early days that you Know took some basic understanding of When consumers are uh available more Likely to be available you know what are The basic economics of your business so That you can decide how much you’re Willing to bid and you had to respond You know pretty fast so you didn’t have Time to do that much computation with The Computing that was available at the Time and Um and so that’s a that’s sort of a an Applied math beginning And um and I’m going to go through each Of the pieces quickly because they all Kind of have that in common where there Was some algorithm that was that was Employed and that later we found could Be dramatically enhanced with uh with AI So if you think about Discovery so You’re in a store actually you’re on a Website and you’re searching for stuff You know again we had some pretty Interesting early algorithms for for Search even before pagerank Um and then for recommendations there Was actually a very early machine Learning approach called collaborative Filtering which was employed in the

Recommendations algorithm you know People who basically bought uh this also About these other things and and that Proved to be actually pretty interesting For customers even though it was a Pretty simple algorithm it made Recommendations especially in things Like books where you had a very long Tail to the catalog uh it made Interesting enough recommendations that People were attracted to the site Um you know in the in the process of Getting stuff from wherever the retail Store was or from a warehouse to a home Or business again we used a lot of Operations research so you know linear And non-linear programming techniques Optimization approaches you know trying To decide first where to put inventory My store warehouse and then the optimal Way to get it to the home and you sort Of TR you you gave me this Arc of of AI And it was it was it matches my Experience uh pretty well although I Would argue the time from 2010 to maybe 2013 was was really compressed and very Rapid advance in what I saw so we had The benefit of 15 years of using these Non-ai-based approaches that were pretty Data intensive and algorithmically Intensive and so we had the backbone to Be both interested and capable of Employing different techniques and when Some of the you know the early work on

Random forest for example and uh and Some of the other uh a few folks have Called them sort of um well groups of Different AI approaches that you know Now seem to be collapsing with Transformers as you and I have discussed But before that when there were these Sort of competing groups we would find Applications for those different Approaches across all of these different Types of algorithms and it would make Them better so by 2011 I was already to A point where I could say for ">marketing I’d rather have the team that’s trying To figure out what to bid on keywords Where to show display ads you know maybe Even in the physical world uh how much Money to spend especially on TV for Example switching from the old Algorithms to ml Um we were seeing in the Discovery Space Dramatic improvements that we could make To to search and Discovery so in Addition to you know classic search Browse where you’re guiding people Through a set of experiences that allows Them to narrow their focus onto things They might be willing to or interested In buying we found the same kind of Thing uh happening too where we had some You know heuristic based things that Were okay we flipped to using all the Data that we had available on these Products and using AI or early versions

Of of AI and ml we were able to make Dramatic Improvement in the ability of Customers to find what they wanted and Then same thing was true on the you know Getting stuff to customers really the The thing that amazed me there was uh it Took so long to solve these optimization Problems the old way in the logistics Space And so you were limited to sort of Uh getting close and then kind of Guessing or using averages to decide What was the optimal Warehouse what’s The optimal cost how best to Route Things because doing that doing that all In the space of you know click one you Know click to place your order and then We respond with uh you know a very Precise delivery time and and assign the Methods so you have a very small number Of of uh seconds if that uh proved to be Really hard with the old computational Techniques with AI you could make all Kinds of real-time optimization Decisions that’s that lets you go all The way back into where to place Inventory in a real-time way that made The operations much more efficient so uh And then by as you pointed out by the Middle of the 2000 teens We were firmly in the campus AI is Basically winning almost every time we Try it against a traditional algorithm And so then it’s like can you get enough

AI scientists right next to the computer Scientists and employ these techniques As fast as possible yep and uh and I Imagine that’s persisted obviously and Now that’s you know there’s a lot of uh Enterprises that are trying to trying to Get there as well but uh but that’s yeah Uh that’s obviously been a source of Probably a source of massive growth in The in the intervening period You know one of the things that at scale Um we we think a lot about or or uh one Of the things that we’re known for Frankly is is being focused on the data That fuels these algorithms more so than Even the algorithms themselves and in uh The e-commerce context uh your your Data’s your data by your products I.E Your catalog and so you know we’ve Talked a lot about this but maybe Um start out with by just telling Um talking through the history of of Amazon and its catalog which is sort of Known in the industry as being the sort Of most pristine uh most most uh Sublime Quality catalog but but tells the story About how how it got there Well it started as a Bookseller of Course and that gave the company the Advantage of building a very big catalog So it knew it had to hold millions and Millions of Records in the catalog and So it wasn’t as constrained by some of The typical you know computer science

Problems that people might have if they Optimize for a much smaller set but as You can imagine the schema the you know The the column headings for the what was Important for these items was very Different for books and then as we got Into music and movies it you know you Could see it starting to change so the Way you describe a book Um you know it has uh uh it has a Particular uh identifier you know on Every book that’s different from a SKU That you might find on a on another Object uh the the form factor is very Consistent you know size and weight Basically in a very tight band versus All the things you could sell in the Physical world World Um you know color for example color size These things style these things don’t Really matter for you know for books in The way that they do for fashion for Example Um and so as we started to get into the Non-media businesses Uh at first the size of the catalog Wasn’t the constraint it was we didn’t Have the right Uh fields to describe adequately the Products that were being displayed on The site and it looks a little goofy if You you know if you describe things with The same detail page that you have for a Book and it’s a kitchen item or a

Consumer electronics item or a piece of Fashion So the the first step was to get out of The book schema and into something more Generalized and and the team came up With this thing they called ptds or Product type definitions where for each New product type they would actually Create a list of all the fields that you Could imagine you’d ever want to use to Describe this thing Um not with the help of any kind of AI Algorithm or really any statistics but Just you know just Brute Force kind of Surveying the all of the data that was Available Um online and And ended up with what became hundreds Of these ptds and the problem is and I I Could see this as we started to expand Rapidly into especially the hardline’s Business as as you know the more ptds The more of a maintenance problem you Have the more challenge you have when There’s inconsistency among them and You’re trying to display search results For a multi-category uh search term Um it just they got more and more clumsy And and then the defects would reveal Themselves in sometimes humorous ways I Mean you would have Um fields or product description bullet Points that were that had head headings That didn’t make any sense for the kind

Of product that you that you were Talking about so for example you know When when we moved in a consumer Electronics people want to know if They’re buying a uh you know a a Complicated piece of uh of componentry For a an AV system or something they Might want to know some interesting Technical specs well we didn’t have Technical specs on books and we didn’t Have technical specs on videos so There’s there’s stuff that you start Learning As you go that was super slow and I had This point of view before I saw the AI Algorithms really working but I had this Point of view that these attributes of The date of the products that we’re Selling should be emergent and not Something that we had to figure out kind Of a priori before we would bring on a New product or a new category of Products and I didn’t have I tried to Find some statistical methods for Accomplishing this again like at the Very beginning of the ml approaches that We were employing I couldn’t it was Clunky but again in that same time frame Between 2010 and 2015 just saw a Dramatic Improvement in the ability for AI algorithms to take very large data Sets on physical products and and Actually tell you so it’s the emerging Tell you what the attributes that matter

Are and in fact and I’m not sure this is Done yet at Amazon but I always imagine That you would be a able to dynamically Change the browse node structure Dynamically change the configuration of Detail pages and the specifications Product type specifications that are on Those detail Pages as you learned more Emergent properties from the items that Humans are adding so in the long run What you do with the marketplace is you Just add stuff And the AI can figure out where to best Slot it to make sense for the customer And to maximize sales Um you know and we were we weren’t there Yet when I when I left but I could kind Of see where where that was going what Was really clear though as we started to Do this Was that the data The accuracy of the data about the Products mattered a lot So the algorithms were only in the AI Output was only as good as the Information that we were feeding it and The truth is which really surprised me But many manufacturers in their you know Early 2010s had pretty bad catalogs so The the the data they were storing about Their products you know was not very Accurate um they often would reuse Um SKU numbers they would uh often Introduce a new product like consumer

Products does this all the time they’ll Say you know new and improved and Same product new packaging kind of thing Well those would often be conflated in a Catalog entry and that leads to a very Imprecise e-commerce offering if you’re Not sure which one it is is it the old Packaging or the new packaging and it Matters because customers if they come To your site and they’re not sure They’re going to get what you’re telling Them they’re going to get they lose Confidence you’ll lose conversion So uh I just I became really focused on Making sure our data was as accurate as It could be and then employing ML and AI To have these uh properties be Discovered emergently and that to me Yields the highest quality catalog and Again as long as you’ve set up the Architecture so that you can scale with The size of the catalog you aspire to Build which is kind of job one You’ll fill that database with much more Accurate information Um I think you can focus on those two Things and on a scale of like you know This is one of you know many things Obviously that that Amazon worked on Through the years on a scale of one to Ten in terms of sort of the long-term Impacts for uh for you know Amazon it’s E-commerce success how important do you Think the the sort of the catalog

Qualities the quality of that catalog Data as well as the sort of ability to Use AI to enrich and fulfill all all These attributes on on the catalog items How important were those things in the In sort of the ultimate successfully Amazon e-commerce I mean I think it’s Among the most important uh attributes Of success for the company if I if I Think about Um the you know kind of this this Transition that occurred and it began Frankly in the 2000 to 2001 era Um with an idea that was simply we Should have item Authority and that’s That’s what we would call it so we Should be able to know for sure that an Item we’re presenting to you as a Customer that we know what it is and we Can describe it accurately and of course It took all the years that I described To get to a place where it was more and More accurate but organizing the whole Website around item Authority Eliminated the problem of customers Having to to do the work to sift through A bunch of results that are kind of Cacophonous to find and almost random to Find the thing that they want you know Which is what some Marketplace offerings Do they sort of you would dump all the Things into a big holding tank and and You get close with search and then let Customers sift through Page after page

After page of results that’s a Prescription for lower conversion the Beauty of having item Authority is that You could then say there’s one detail Page for every item And you can let multiple Sellers List Against that single item but that’s an Authoritative detail page that you can Then Market directly you can have an Auction based Um you know sponsored links which of Course Amazon launched with its Sponsored advertising uh programs so That people who wanted to boost the Success of their brands could bid on Their detail Pages or the brands detail Pages and know for sure that those are The brands that were going to show up it Helped when we got to worrying about Counterfeit so of course Something else that was happening in the Early 2010s is that there are a huge Number of uh Asian uh e-commerce Competitors typically principally who Were starting to attract uh customers in The US and Europe And Um you know as they started to uh to Come in uh it it just Highlighted for me even more the Importance of having the item Authority That would allow us because some of These would list as sellers too on on Amazon and to be able to identify the

Counterfeits from the uh the authentic Merchandise especially using all the Catalog data that we had was another Element of the customer experience that Was pretty vital in the you know and Aided dramatically by AI Sounds like uh you know there were a Number of things that you you mentioned There which is one obviously it was sort Of core to creating the best political User experience which results in a Higher conversion which sort of had all This business impact it was also core to To actually enabling Amazon to build the Uh the ads business that is built the Sponsored sponsored ads and sponsored Listings uh business that Um has been core to the in search ads And all the stuff that has been core to The sort of success of Amazon and then Also Accord to the sort of this problem Of counterfeits which is critical to Maintaining user Trust Yeah yeah and the ads business is Totally you know you you won’t have the The same conversion on an ad if you if You send somebody to a whole list of you Know of items that they have that the Customer has to sort through if you if You can let let a seller or brand bid on A thing and know with high probability That when the customer searches in a Specific way it’s that thing they’re Looking for you’re going to get much

Higher ad conversion so yes it led to a Better ad business as well as to better Conversion for the Totally Um and uh And so the another piece of this that You talked about was this sort of Concept of item Authority and the sort Of um the unification of all the Different Um sellers on a on a uh on a per product Basis and we’ve talked about this before There’s there’s other marketplaces that Have not yet implemented item Authority Or maybe Midway through the process of Implementing item Authority Um you know do you uh I’m curious you Know a lot of folks try to approach item Authority in a very in a very manual way And this is obviously all about we know We’re talking about AI how do you think AI actually enables you to do things That to build item Authority Well look in a world where People had spreadsheets and you know ad Hoc systems for maintaining data about Their products AI might be you know several steps away But in a world where most companies now Either manufacturers or sellers have an Authoritative at least an authoritative Set of data uh it may or may not be Accurate but at least they’ve got you Know a row for each thing that they sell

You you’re well positioned to employ AI On top of it to make uh significant Progress at at any scale so so what that Lets you do is it just lets you move Faster deploy less less human resources To clean up your catalog along the way And get the benefits in terms of Conversion and customer trust faster so For me for me actually speed is the is The ultimate Advantage maybe some Element of scale but really it’s at any Scale whether you have 5 000 items that Change enough that you worry that your Catalog could fall out of sync or you Have 5 million or you know 50 million Items you’re going to be able to move Faster And earn more trust with customers if You’re using the most modern tools I Think Um and and this is this is an Interesting point especially right now We’re obviously in a point of uh you Know a fair amount of economic Uncertainty and so there’s sort of Questions around hey what are the you Know Um is this the right moment you invest In in technology that might be expensive Um especially if you know you’re not Sure what the ROI might be you know I’m Sure you had to navigate this a number Of times through throughout the sort of Uh 20 plus year Arc at at Amazon how do

You think about that you know in how do You think about investing in moments of Of economic uncertainty and and how do You think about navigating that Well there’s this it’s so interesting What difference a year makes you know Um last year we wouldn’t have had this Conversation Um you know at the conference and you Know the macroeconomic uh environment is Very different I I think there’s there’s An adage that I’ve heard that is really Apt that you know in times that are Tough macroeconomically Good companies survive and great Companies thrive And And then you have to ask well what are The key tools for thriving And you know I just I just think in if You’re operating a company in in retail That is more than a boutique you know More than a boutique that a single great Store manager can manage with a Spreadsheet Then now is the time to lean in and make Sure that you’ve got a great Architecture for data capture and Storage and usage and the right Algorithms operating on top of it which I think you know are increasingly almost Always uh AI based or at least you know A hybrid of classic techniques and an AI Um and my feeling is every month you let

Go by where you’re tentative about it Is more months later that you’re going To have to deploy more capital and a More frantic Pace to catch up because There’s just you know there was a time When Amazon was kind of the only one in Retail that was that was deploying these Techniques this way that’s not true Anymore I mean in the two years that I’ve been out of Amazon I’ve run into Plenty of Founders Um and people that are operating Tech Organizations in some large retailers Who are very sophisticated about the use Of AI you know they may have Catalog data that suffers from more Legacy problems than I had because we Started cleaning it up much earlier they May have a smaller catalog their Problems may be different But there are so many people doing this Now I I kind of think if it’s not table Stakes yet within the next couple years It’s going to be table stakes and you Risk the entire franchise that you built Uh if you don’t at least keep up right Um you know one of the one of the Product decisions that you’d also Mentioned sort of in the in the mix There was um was the buy box Um and uh and this is actually something That we’ve we’ve talked about before as As a uh as one of those project sessions That’s sort of like you know it took a

Lot of iteration to get to the level of Sort of like Beauty and simplicity that You ultimately got to Um maybe maybe just talk through that What were the obviously uh we talked About a little bit but what was the Original hypothesis behind the sort of Like authoritative buy box concept and And what were some of the the product Decisions that you you know you and team Got really right Um in that process Well the I would say that buy box is kind of At least on Amazon is synonymous with us With a single authoritative detail page So you you basically present An item that you’re confident you know What it is you’ve described it properly It has images that will make sense to Customers and that will help them decide If they want to buy it or not it has Reviews and ratings all that all that Stuff’s on the detail page But if you’re going to open your store Up to multiple sellers you have to you Have to make it simple for customers to Identify the offer that they want To select if they’ve decided that’s the Thing they want to buy and the you know The evolution of the buy box was just a Way to take this space that before you Had item Authority was just how you buy On each detail page but the detail pages

May be duplicated To using the buy box and some of the Other features that are in it or just Below it As a way to drill into all the offers That might be available At first we just had a single offer Shown in the buy box and that persisted Actually for more than a decade after we Began to do the item Authority work and You know worked for a long time you Basically had you know an offer that was Free shipping uh it was typically the Lowest price as long as it was available Immediately and from a highly rated Seller super simple but what happened Over time is you start we started to Have a proliferation of different kinds Of offers so for example You might have something that is Shipping from China And would take two or three weeks to get To the U.S but if the customer is Willing to wait the cost was way lower Than something that could be you know Shipped immediately and delivered Tomorrow and we didn’t have a great way To give customers those trade-offs for Highly rated offers so we started to do We went back and spent actually Three or four years re-architecting the Way the buy box worked the way offers Worked the way the schema worked so that You could present multiple offers at the

Same time in the buy box Um you know this is important for for Regional offers so for example one of The things you’ll see on Amazon is a Persistent address there’s a location Widget that’s very simple it’s in the Nav but that was really important Because before that was launched we were Guessing where people were and in fact You’d often default to like the center Of the country there was some city in Kansas I think that’s actually the Geographic center and that would be like Literally all the algorithms would use That as the Assumption of where the Thing was going to be shipped it’s way Better to just ask the customer or just Say here’s where we think you are you Know change it if you like and then have The items that appear or the offers that Appear in the buy box adjust Uh and be relevant based on your Understanding of where the customer is And so we had to do a lot of work to get All that right Um but the the it began with having the Buy box be a way to once you had item Authority to get to multiple offers that Were taking advantage of that yep Um you know you talked about how this Was one of the most important you know Product decisions or product features or Underlying uh functionalities that uh Enabled the success of Amazon I’m

Curious what were what were some of the Others Well look I think the logistics Capability that Amazon has is is pretty Remarkable Um you know I’m partial to it because I Helped to to build it in the early days But um a team just accomplished some Really amazing things over uh 20 years And Um you know it allowed Amazon to Continue to scale in terms of the Breadth of the catalog that they uh Could support with Direct Delivery the Number of sellers who could just send Stuff I mean if you think about the FBA Business uh it’s not Amazon ordering Things from a vendor so that they can Decide what to have in their store It’s Just Trucks show up with fulfillment by Amazon stuff and you you have to kind of Anticipate the amount of warehouse space You’re going to need and you know you Many of the folks uh that are watching Will know that Amazon spent tens of Billions of dollars in capital Uh over the last couple years to build Warehouses just in case Now the the challenge with this of Course is that if you hit a Slowdown Which happened you know over the last Year you may be over built Um but my thinking was always that you’d Rather have the insurance policy to

Protect the customer experience and make Sure you could accept the inbound from Sellers than to be cutting it too close To the The Razor’s Edge Um and you know and then you just assume That as you grow you’ll be able to grow Into the additional capacity to get yep So we talked about a bunch of different Interesting uh interesting sort of Advancements that you all came up with At Amazon one is sort of the the item Authority catalog one is this sort of Rich rich uh product information that Enabled better search and Discovery and Then there’s then there’s Logistics of Course what what sort of the thread that You think ties those all together or at Least uh you know these are the seem Like very disparate Innovations Um so how did how did Amazon sort of Come up with all them You know I I think the threat is is Competence in computer science and a Deeply analytical Um culture and uh you know even you know If you’re a retailer often you’ll think About the merchants are the most Important and kind of the brand and you Know the look and feel and all of that Matters but I would argue in today’s World you have to be competent in Tech Or you just are going to fall further And further behind those who are Um of course there are going to be

Luxury boutiques and small experiences That are highly tuned that don’t need to Have all of that technical Sophistication but I would argue that by And large those are going to be the the Minority of experiences I think Consumers have come to expect the kind Of precision that you only get when You’re sophisticated with technology yep 100 I’m curious if these uh going back To the catalog piece they’re really you Know now that I’m thinking about it There’s sort of two components that we Mentioned one is the the Um item Authority catalog and the other Piece was what we talked about which is Using machine learning to to pull out a Lot of the rich attributes from from an Individual page you know Um comparatively What was sort of the relative value of Those two things like how how did how Did that play out over time Well I think they’re coupled I think Deciding to have an authoritative Catalog is kind of the item Authority Piece and then how you go about making Sure that it’s authoritative when it Comes out and then it adapts to the Changes that are coming in It goes to what tool tools do you employ To to grant that assurance and I found That AI is you know Far and Away the Best tool to ensure that an

Authoritative catalog remains so right What are some of the biggest trends that You foresee in the world of of sort of E-commerce in retail even just generally Uh like what what what does the future Hold Well to me the most important thing is That and this has been around for a Little while with names like omnichannel Uh retail but uh I think this notion that e-commerce is One thing and physical stores are Another thing is antiquated I don’t know Any customers who wake up in the morning And go what am I going to buy online Today They’re customers that I met and all my Friends and the way my family behaves is You wake up and you say what do I need And where can I go to get that most Efficiently and sometimes you’re in a Mood where you want to browse and play And sometimes that’s okay online but Often you’re going to want to be in a Physical location for that Um sometimes you’re away from your home And you’re you know you you want to Order as efficiently as possible and you Don’t want to browse through a bunch of Stuff you just want to search and find It that in that case a mobile app that’s Super good with great item authority to Let you get in and out really fast might Call it a spearfishing kind of uh

Shopping session is really important Um and behind that customer experience Or the front end experience is the Question of where is the inventory that Fulfills that experience and my argument For a while has been that inventory is Going to be serving both stores and Direct to Consumer delivery uh in an Increasingly fluid and dynamic way over Time And to me that’s the that’s the most Interesting trend is how as you if You’re a larger retailer Um and you’ve thought of your world as a Physical store world and reluctantly did Some e-commerce stuff because it seemed To be growing fast over time I think you You want to optimize how the back end uh Of inventory storage Logistics stocking Your stores and getting things to Customers can be optimized to be one Great system with the help of of you Know great computer science and AI Um you know to me that’s that’s the Biggest trend is this sort of Um kind of consolidation of these Various models into a system that simply Moves items from manufacturing locations Factories in the world to the homes and Businesses that want those items for Their use And and in that transition you know I Can make guesses but like in that sort Of in this transition to you know

Whether it’s omnichannel or just sort of Like a more unified buying experience What are the key leverage points that You think AI will will play uh where Where AI will play a huge role I think it’s all the ones we talked About it’s it’s ">marketing Discovery Um the the transactional experience and Then the the logistics and I can see More and more opportunities to uh you Know to do some some strategy I would Call it you know kind of Um topology strategy work Uh with the help of AI so I think Increasingly we’re going to have the Ability to optimize the uh networks that We operate the placement of inventory And and that sort of planning stuff that Happens before a uh a sale uh in more Sophisticated ways so there’ll be less Guesswork which means return on invested Capital should go up the amount of Inventory in the system should go down Inventory turns in retail in my opinion Are are way too low awful Um in in the world on average they turn I don’t know somewhere between two and Four times which makes no sense to me Um there’s 24 trillion or so of Inventory of retail sales every year in The world and if inventory turns four Times that means there’s six trillion Dollars of cash and stuff that’s already Made sitting in warehouses basically in

Mattresses around the world if you could Go from four turns to eight turns you Would take three trillion dollars of Inventory out of the worldwide system And free that Capital up for deployment And all kinds of other more productive Things and you know that’s actually to Me the this will be the monitor that I Would I would watch is overall not Because people are doing crazy things to Get inventory off their balance sheet Not because they’re forcing all the work Onto their suppliers all the short term Stuff that you know many of us did over 30 years but real fundamental change to The efficiency of Supply chains will Lead I think to dramatic Improvement in Inventory turns and then again Consequently freeing up all that capital For further investment yeah totally I Mean that and that all comes from being Able to better estimate you know the the Sort of Um projected demand and matching level Supply and then having confidence in Those in those estimates you know There’s a great analogy that another Speaker had As we were filming this with these which Is like we were talking about climate Change and sustainability is a broad Problem and it was like you know do you Think that every HVAC system in every Building around the world as well uh is

Well set up obviously not uh is Incredibly wasteful and then another Example that you know one in one in Every three fish that get caught and Getting thrown away or going bad or in Some way and so the sort of the the this This metric that you mentioned the sort Of like inventory turns is is a great Example sort of This Global Issue around how do we how do we better Utilize our resources Um and and uh not have to have so much Sort of slack in the system to uh deal With you know uh variability in demand Yeah and that’s that cash that’s freed Up for society can go to Wellness and Hunger and education and you know Environmental sustainability I mean There’s just there’s so much that we can Do uh that’s uh better for the long-term Future of humanity than put cash into Mattresses Totally you know what um as we think About the AI ecosystem you actually know You know scale and what we do uh pretty Pretty darn well uh better than almost Anyone Um what are the ways in which you think Scale can work with e-commerce players To transform their businesses Well I think we’ve hit on a number of Those things I I think as you pointed Out a couple of times uh scale has this Great focus on helping companies to

Improve the quality of their data and The usability of their data and I think Um all of the stuff we’ve talked about Starts with that so you know and because Scale has seen so many different Applications of AI across so many Different kinds of companies I think It’s really well positioned to help a Company assess kind of where it is What’s the state of its data what’s the State of its systems what’s its kind of Capability and and preparedness to use Sophisticated algorithms and then as I Know you’re building a whole Suite of Capabilities and tools that will let Companies that are that range from much Less sophisticated to really Sophisticated and wanting to be you know World-class to employ AI uh really Effectively and efficiently and you and You have for some of the human in the Loop processes that are necessary to Clean data to improve the use of that Data in the early algorithms to modify The algorithms to make them more Effective you you all can sort of Provide that labor Um at the time just the time that the Company is going to need that support Um so that they don’t have to build up That capability and learn how to do it All themselves Um you know one at a time So I I think the suite of capabilities

That you’re offering is is incredibly Valuable for for this world the retail World Um and I’m I’m hoping that more ctOS and CEOs from that world talk to you yeah Um you know maybe to close uh we we’ve We’ve had lots of open-ended Conversation about artificial Intelligence and and uh I’ve uh I’ve Bored you with my short story ideas Among among many other things and Um uh I I’m kind of curious you know you You’re always a very optimistic person But uh you know artificial intelligence Obviously a Hot Topic technology work Because the capabilities are advancing So quickly you know Um are you an optimist about how the Technology will affect humanity and and If so why Yeah I am and I think All of us have to be because it’s what’s Been clear to me in the last 10 years is That the Genie’s out of the bottle in Other words enough people know about These techniques know their power know The kinds of things that they can Accomplish that we’re not gonna Force ourselves to forget So if you proceed under the assumption That we will use machines to complement Our work and our thinking as humans more And more over time Then of course we should organize how we

Use them to Produce results that we’ll be proud of And I I think it’s in our best interest And I think that typically when humans Have something that ends up being good In the long run And that’s in their self-interest to Pursue It usually works out well for us because Generally I think humans are Self-interested companies certainly are Self-interested and in this case the Self-interest aligns with a technology That can be hugely enabling for Improving our uh you know our condition As humans and improving the efficiency Of the things that we do so we can free Up more resources to do things we want To do instead of the things that we have To do just to survive So for all these reasons I I remain an Optimist Um and uh and I hope you are too Everybody who’s watching of course Jeff uh this has been a blast I think uh It’s always great to talk to a legend in In any field who is uh built uh you know Incredible you know one of the wonders Of the world and um and learn about how That happened and uh hear about how AI Was the critical part of that Journey so This has been such a gift thank you so Much Chef Um thanks for coming back to transform X

Alex great to see you as always and Thanks for including me Thank you [Music]