Revolutionizing Retail: How Adidas Utilizes AI for Success in B2B

Introduction and Speaker Backgrounds

Siddhi Saraiya: I lead the group at the product management group at Adidas that's responsible for B2B. So something that people don't know about Adidas is most of our revenue really comes from wholesale which means selling to other companies. Companies like in the US this would be a Footlocker or a Dick's Sporting Goods. In the, in Germany or in, in Europe this could be a Zalando or also a Footlocker.

So many, many different companies I saw somebody from Colombia on so there would be like a Mercado Libre. Right. So globally from, you know, to other companies and, and my team is the product managers that handles all interactions between buyers.

So when they're looking to buy the Adidas product and then also making sure that we perform in marketplaces. So content syndication out to those marketplaces, text as well as images to make sure we perform on digital marketplaces. Yeah, I love that.

Samitha: And I'm Samitha, I lead enterprise marketing here at Jasper. So my role similar to Siddhi is all about figuring out how businesses can make the most out of generative AI and really figure out how to marketing co pilot fits into their enterprise technology stack. So the, you know, prepping for this webinar, it was so interesting talking to you about all of the different lines of business Adidas has. You know, typically when we think of Adidas we think of it as a consumer product, something that we go through, go through and see on a shelf and just buy.

And it's interesting hearing the entire supply chain that goes behind that. Oh, we have Megan joining us. Welcome Megan.

Adidas B2B and AI Implementation Journey

Meghan Keaney Anderson: Can you tell us a little bit more about Adidas decision to implement Jasper and what sort of things you were thinking about in this B2B buying journey that you know, thought hey, I want to think about AI and this is the next step for us.

Siddhi Saraiya: Yeah, definitely. So I think if we put ourselves back a year ago right into kind of Q1 2023, I think a lot of us came back from that holiday period really excited by the potential of AI and, and our goal was to think about efficient ways to bring AI into our processes and our ways of selling to B2B in thinking about these existing processes and how to improve them. A lot of the initial work was also really around experimentation and exploration.

So it began quite wide like what types of areas could be impacted and improved by AI. And from there we felt that one of the most mature and accessible areas in AI was, was around these large language models and really it's application and marketing copy and it's applications in all things that were language related. So that's where we began to hone in and began exploring more specifically into that, into that area.

So it wasn't really always about, you know, Jasper or copy or AI or Jasper or copy. It was really about AI is so interesting, like how can we apply it? What is the business problem that we can solve with this?

Experimenting with AI and Initial Tests

Meghan Keaney Anderson: So I would imagine I'm going to hop in there on a question because that is interesting. I would imagine that the interest in AI didn't hit evenly across your organization. Can you tell me a little bit more or tell everybody a little bit more about how the role you played in sort of championing that use, championing that use inside Adidas and kind of what tactics you use to encourage a company as large and impactful as Adidas to start to experiment?

Siddhi Saraiya: Yeah, definitely. So there were definitely some pre existing interest in the organization and also some really like strong and concrete experiments that were happening in different parts of the org. Right. Adidas being such a big company, we have a very well developed different areas of brand sales, etc. In the sales org. I think that one of the biggest or the most impactful tactics was really getting executive buy in generally and just excitement.

So the first step was creating an educational set of slides and not talking about necessarily the technology, but talking about the business problems it could solve. Because not everyone is interested necessarily in the technology, but depending on your audience they're interested in, how can it help them. So really a wide educational sort of deck. A second tactic was really finding a wide group.

So it wasn't just me going alone, but it was really bringing in a partner from the brand org that builds the actual product, bringing in somebody from the tech org from enterprise architecture and also reaching all the way across to get legal involved very, very early on. Right. Because when you're thinking about copy and communicating about what you're building, you want to make sure it's accurate. A third tactic, if I can keep going, was all around defining small tests. Right.

So I think that when it comes to trying something new in a big organization, it's really helpful if you can define a small and safe test to prove to yourself as well as to your stakeholders that there's kind of these legs there. Right. What are some small inexpensive ways that you kind of control that you can test and show the, the impact and then from there grow.

Meghan Keaney Anderson: Can I ask what were some of those tests for you all?

Siddhi Saraiya: Yeah. So for example, when we thought that product copy was going to be like the area that we're going to work on. Instead of going with our whole catalog, we first tried with 150 models. What happens if we send it directly to ChatGPT? What happens if we send it to Jasper? How does that work? It helps you test the different parts. Not only does it help you test, like, the technology, but also a bit of the process. Right. How do we actually send data back and forth?

Then you get deeper. Like, how do you train a model? What does that mean? Right.

And again, like, using the same 150 models that we. And I'm talking about physical product models, not language models. I got you. Yep. Actually, like, using 150 models. How do we train them? Train the language model so the output for those models is better. Right.

So finding, like, small subsets of the catalog we could test an improvement with was very impactful and helpful for us.

Meghan Keaney Anderson: Yeah. And that matters a lot because the complaint that you hear about AI most often is the outputs are pretty generic. They're. You know, it's amazing that it can do it, but it's not anything I could actually use in my marketing because it's so absent of my brand.

So what I hear you saying is the work that you sort of put in, even as part of this smaller experiment to infuse the models you were using with that kind of brand voice and product details really went a long way to make those outputs better.

Siddhi Saraiya: Yeah, yeah, especially. And also it helped us understand, like, you can get different levels of quality and. And the first thing that you get back is not the most you can get back. Right.

And I think that by experimenting a bit, we were able to understand, okay, if we put in a little more and we help to train the model a bit, then we get things that are usable. Right. So I think also going through a few iterations with it was helpful and not saying, okay, the first thing is too generic, but going a little bit deeper and trying.

Meghan Keaney Anderson: How long did that first experiment take? How many weeks did you dedicate to that?

Siddhi Saraiya: So I think that we went back and forth for. Yeah, for, I want to say, maybe six to eight weeks. So it took a little while of kind of going back and forth. We ended up actually using it more widely in our second season.

So in the first season, we felt like we're not really ready for prime time. And then we kind of waited, but then we had a bit more information. And then with our second season, we were able to kind of take it more widely.

Meghan Keaney Anderson: Would you say you were more testing the different models? And in this case, I actually do mean the underlying models or were you more testing which use cases were sort of the best for you?

Siddhi Saraiya: We knew pretty early on that it would be around this product copy use case for a variety of reasons. I think it also, there can be so many applications, so I think it helps when you kind of define and say we're going to try to solve this problem and kind of get to some depth on it as opposed to being really broad in the applications. At least that's the way that it worked well for us at Adidas.

So we really defined one problem space would be this specific copy topic and then from there we would, you know, we stayed with it for a bit before thinking about other things.

Training Models and Brand Voice

Meghan Keaney Anderson: I'm going to stay on this for a moment because there is some interest from the audience on digging a little bit deeper here. So let me ask some questions that are coming in on this topic and then we'll move on to some of the other challenges and experiences you had rolling this out. So Laura wants to know, you know, when we talk about sort of training the model especially I think in using Jasper, let's just talk within our own product here. Were you using Jasper's brand voice functionality or was it the Knowledge center within Jasper? What aspects of Jasper were using to kind of train for better results?

Siddhi Saraiya: So we were working pretty closely with the Jasper team so we were sending over like kind of big samples of copy and then from there using that to train. So I'm not sure which product feature it would fit into within Jasper because we were working so closely with our expert account to help make that voice accurate.

Meghan Keaney Anderson: So I can, I can probably ad lib on that a little bit because just knowing the product so well, typically when you're thinking about something like, like product specifics around, you know, sizes and specs and details around the product, that is something you would want to add to the Knowledge Center. And when you're talking about, hey, when we communicate about our product, we use this style guide. We use these kinds of tones and voices overall kind of the brand characteristics. That's what would go in brand voice.

Siddhi Saraiya: So we definitely use both. We use things that were product specific within the season. Then we use things that are like technologies that we have that are evergreen and then we use things that are like training the voice over all of our kind of brand guidelines of how we speak.

Product Copy Use Case in B2B

Meghan Keaney Anderson: Yeah, we've had a couple questions around the actual like product copy use case itself, so I'll kind of kind of combine them here. Tell us more about when you, when you talk about product copy, are you Talking about product pages at scale, what's the actual problem you're trying to solve there?

Siddhi Saraiya: Yeah, so in, in when you think about selling to B2B in a product company, you're actually selling quite far in advance of when the end user or like the, the end consumer sees the product. Right. The B2B sales cycle takes place ahead and then from there you're thinking about the brand creating that product. Right.

So the brand is creating the product and the sales team is beginning to sell. And in that time you have this creation of marketing materials that happens. One component of that is the product copy which goes along in that process of the brand selling to other companies.

So for example, we would be, right now we're in 2024. So you're think we're thinking about, you know, 2025, that we would be selling the future. Yep. Yeah, exactly.

So you're waiting for that material to that like content to mature from the brand perspective and you're going into sales. So that, that building that copy is something that used to take a lot of time. Right. We would use agencies, we'd use other, you know, kind of ways to do it.

And you would just have a lot of back and forth trying to, as the data matures, trying to build that copy as a part of your kind of package to go out to the, to, to selling to B2B customers. And then by using AI, we were able to get that essentially this could be short and long form kind of copy to describe the product. So you're having inputs that are all the different product attributes, things like gender or use or, you know, descriptions of the product and then from using it to build an initial description that helps people make buying decisions.

Meghan Keaney Anderson: Got it. That's really, really helpful to get into those details. Thanks so much for that.

So we've talked a bit about the use case. I want to shift gears and talk a bit about the challenges that you encountered along the way as you were rolling out AI internally. It sounds like it was very thoughtful in that you started with this small experiment and then based on the results of that experiment, you then have to get out to a broader audience. What were any of the roadblocks that you encountered or the things that gave you pause as you started to roll this out.

Challenges in Rolling Out AI

Siddhi Saraiya: I think that, you know, first of all, there is an existing process. Right. So you're not starting from ground zero.

So I think it's like, how do you create room for experimentation within an existing process without creating too much disruption? Right. And you need people to be your Allies as well.

So I think creating that, that space was definitely one of those topics. Another piece was just structuring the initial test. Like in retrospect I can say this is what we were trying to do, this is how we did it. As you're going through, you're feeling around in a dark room, you're saying, okay, we want to solve product copy, so we need to train a model. How do you train a model? What do we use to do that?

Now after it comes back, we need to spot check the copy to make sure it's right. Who's going to do that? How do we do that? Right.

So it's like really like figuring out all the different details. In the beginning it was also thinking about do we want to use AI to cover 100% of our catalog or 50 or 25%. Right.

So like all these different specifics I think were one thing where you're really figuring out the next problem as you go. Right. You don't know everything from the beginning, but kind of working in an agile way to set up the process. I think the other thing too, another challenge was trying to figure out how to be lean with it. Where do we need to invest and where can we use workarounds. For example, in the beginning we didn't exchange data over an API. We said, okay, we're not trying to test an API connection, we're trying to test the quality of the content.

So let's just use spreadsheets in a secure way to trade, to test the quality of the output. Right? So that you're thinking smartly about where you spend resources to make sure that you're getting the answers you need out of the exchanges.

Meghan Keaney Anderson: That's really smart. And I want to just kind of underscore a couple of things that you said there. Because it's not just technology.

So much of the AI conversation has been around what tools you use and the technology. And that is an important piece of it. But what I'm hearing you say is side by side with that. It's the technology decisions, it's the process decisions, it's the people decisions of to your point, who's going to do the fact checking, who's going to do quality assurance.

And it's sort of the full end to end program of how are we going to work differently in light of this technology entering our folds. And I think seeing holistically like that is an important thing to just underscore because there's so much experimentation happening. And that often is the difference in my mind between Something that is just splashing around an experiment and something that actually makes it into longevity within a company.

Siddhi Saraiya: And I think another part to highlight there too is also the people involved. Right. Because as you, as you mentioned, it's a process that's changing and there's many different functions in a big company that have their place in that process.

So how do you get people and representatives from those functions and bring them along as you're solving the problems? Right. Because when you present people a finished product, it's hard. Right. There's a lot of reasons why it won't work.

But when they go along with you a bit in the journey to say, okay, we're doing our best, they're really good intent. We don't know all the answers either. I think it helps create this spirit and it gets people on board. Right. Say, come on, let's solve this together. I don't know, what do you think?

So that cross functional group was very helpful.

Meghan Keaney Anderson: We've seen that a lot with some of our other customers around. This idea of bringing together this cross functional committee or AI council that no single person in the company owns AI, but this is a council that can help shepherd it in and expand until everybody's comfortable with it.

Measuring Impact and ROI

Meghan Keaney Anderson: Okay, so just to keep things moving because we have a lot of questions to get to, I'd love to know how you measured the impact of incorporating Jasper into your workflow. You talked a little bit about the experiment, but I do wonder how you think about the broader, how you measure quality, how you measure the return on investment. So you sort of can decide where you go from here with it.

Siddhi Saraiya: Yeah. So from a cost perspective, when you go from a model, sorry, a cost per model kind of service agency fee to a technology license fee, that whole, that process is, I mean the cost is just, it's more scalable. Right. It's more, it's obviously more cost effective.

So I think that that's one aspect. Another aspect of cost is the operational part of it. Right.

So how much time does it take from the different parties involved? We found that this was also like positive. Right. It seemed to be faster with like lighter touch.

So we found that was also, you know, kind of a positive part, you know. Yeah. Essentially in, in that cost and investment back and forth. Another thing that, that we thought about a lot was, you know, in the B2B space, it's not.

So it's not quite like B2C or E Com where you're able to measure conversions in the same way because there's often also A sales rep that's involved. There's many other touch points as well. So.

So it's not as singular from one perspective. Right. But we were able to bring up our coverage and we didn't see any negative impact.

So we felt that was positive. Right. When you have costs going down, you have coverage going up, you don't see a negative impact. That's quite positive we felt.

And the other thing around the ROI conversation was like the learning again it's not as, it's not as numbers oriented but just the fact that we got to play with an emerging technology in a really practical use case and learn I think was also quite valuable from an ROI perspective. We're still early in the journey so I'm sure that will come. Right. We will have those comparisons on quality, on conversion and especially as we start exploring more on the B2C side where that is so much more numbers driven. Like the E Comm site is one of the the top ecom sites in the world. There's a ton of traffic. Sure.

So if, if, when we start to use these types of tools there, if again my world is a B2B world so then that I think that those answers will be like more driving the conversation.

Meghan Keaney Anderson: Yeah, yeah. I think that's an important delineation there. There's lots of different types of ROI and they mean different things for different businesses.

So I feel like if we're giving, you know, advice as part of this for retailers or other companies that watching to think through, you know, what is the hierarchy of needs in terms of ROI for your venture into AI? Is it? I want to get to market faster.

So it's a speed and time thing, is it? I want to save costs on, you know, the number of, you know, freelancers I need to bring into the mix or the, the amount of content that I have to create. Is it about higher quality and optimization or is it about, you know, enabling your company to sort of learn and develop and innovate on technology as soon as possible. Right. Or any other spectrum.

But knowing you know, what the return is you're trying to get seems like an early part of the decision making process that will help then inform how you structure its use internally.

Siddhi Saraiya: Yeah, great.

Choosing Jasper and Partnerships

Meghan Keaney Anderson: So I'm of course, so we're incredibly honored that as massive fans that you all decided to run your experiment with us and to use Jasper. This is definitely a biased question or like self serving question but I would love to hear from you like why us? What was it that put Jasper over the top when you were Considering other tools that you could use.

Siddhi Saraiya: I think there is two aspects. One of it was a technology aspect and the other one was a more cultural aspect from a technology perspective. We thought it was really nice that you're essentially an aggregator layer on top of different models for us to have to integrate once and create this partnership once. I remember a few weeks after we went through our initial test, we saw there's another great model that's come out for translations. Okay, well I have to worry about that because Jasper is going to work on it for us. I think that that's great that knowing by using this product we will access the latest technology.

And our differentiator is shoes and shirts. We are a product company and your differentiator is technology. So it's great to work with vendors who are going to stay at the top of that game so we can focus on our differentiators. That's one.

And the other part was the cultural part. Like it was this right balance of maturity but also experimentation and flexibility that we liked. There's a lot of very, very young AI startups and then there's like a lot of, you know, a lot of vendors that are, it's, it's maybe harder.

So I think that was that right kind of size combination and I think it was really a partnership. And back and forth, there's a lot of calls and a lot of like, hey, we're seeing this, what do you think about this? So I think that consultancy for us was also helpful.

Meghan Keaney Anderson: Yeah, that's great to hear. We get a ton out of that, the value of that relationship as well. And that makes our product better. The first piece you were talking about we often refer to internally as interoperability, which is a $10 word for exactly what you described, which is being able to pull the best from a huge range of models and then also improve your reliability.

So if one of those models is having a bad day, we can switch over to the others and make sure that there's consistency and a kind of continuous updating of, to make sure you have the most recent, the most innovative of models that are available on the market. So yeah, that's awesome to hear. We don't get to talk about that that much because it's definitely techie and like under, under the hood of it, but it does, it makes a big difference to our customers. It's awesome to hear you say that.

Balancing AI and Human Creativity

Meghan Keaney Anderson: So one of the things that we talk about a lot is finding the right balance between AI driven processes and human creativity. We are marketers after all. We care about creativity. We want to make sure that AI is additive and a benefit to that creativity and not a detriment.

And so how have you gone about thinking through what processes and aspects of the work are great for AI and which ones require more or entirely kind of human driven creativity?

Siddhi Saraiya: Yeah. So again, staying to my domain because Adidas being such a big company, you know, I think that in the B2B sales world there are. When we think about creating content and making it ready for that world and working closely with brand on that, I think that there'll always be kind of like this pyramid of products, right. Like there may be the ones that are at the top that will always need that incredible storytelling and there may be sort of like you're more constantly in, in stock, you know, products that you can kind of use something like, like Jasper on, at least in the beginning.

And as it evolves over time, maybe you're able to use it for like the deeper the entire range. Right. That's one, one piece of it. I think that there's always going to be this partnership. It will never be like handing over. Even now we have people who are building the data and the data includes things like the consumer intent or the key attributes that also then are built into that copy. It's not completely built without that.

Also the softer inputs. Then I think there's also a different type of creativity. For example, again, I'm a technologist, so it's like, how do we train the model in the right way?

But that's also this kind of creativity to say this voice needs to sound different to reach this audience that's buying this type of product. Like at Adidas we have great fashion products, right. That are selling to key influencers in the fashion space.

Then we have like sports specialists. Right. Maybe the way you need to talk to different audiences is different.

And that's kind of this fun place we're in where we can think about those questions even as technology to say, huh, you know, the output is not what you need. Right. And then we get our colleagues involved from the right places and say, how would we better capture this in order to create outputs at scale that represent the need for that segment? The creativity changes a little bit, but I think it will for the foreseeable future. We will have close partnerships with people who are experts here as well. Right. It becomes not either or, but kind of a combination.

Meghan Keaney Anderson: Yeah, I love that way of looking at things. It's meant to be an amplifier and always a combination and not, you know, not a one for one replacement in that sense. Great.

Advice for Retailers and Next Steps for AI

Meghan Keaney Anderson: I'm going to ask just probably I'll jump ahead and ask maybe one more question from. From our list and then this is also a prompt for others to throw your questions into the question pane so that we're able to see them and maybe ask a few from that. And we can follow up after this discussion with answers to all the questions as so none are sort of left behind.

So I guess I'll keep my last question fairly broad, which is, I imagine we have a number of people on the call today who are within the retail industry or the CPG industry or kind of related consumer products as well. What advice would you give to others in your position when it comes to starting to, you know, begin that journey with AI and explore what the right use cases are for them, what the right way to introduce it is to their companies?

Siddhi Saraiya: Yeah, I think one piece of advice would definitely be to be bold, like go out there and make the pitch or have that conversation. Right. I come from a tech background and what I found when entering a physical products company is that sometimes you need to speak a bit of a different language. Right. Instead of being excited about the technology for technology's sake, you need to also understand your users.

And don't pitch the technology, but pitch the business problem. But if you see that problem, pitch it. Like, go talk to people.

And it needs a face and it needs a voice and an advocate. So be that person in your company. Also create a small group of people like we talked about earlier, that can join forces, sales people who represent the accounts, brands, technology, legal to come together, to do that together. Right.

Because I think in a big company it's what's really needed. We talked a lot about tests, structuring small tests with concrete results that show to you and your stakeholder group that it can be effective and also to hopefully create some room for innovation. If you are a budget holder, then, you know, create some space for these type of small tests as much as you can, because it's not only is it great for top line and bottom line, it's also a great driver of employee nps and it makes folks within the company also think in a little bit of a different way, which is always fun.

Meghan Keaney Anderson: Where should AI go next for your. I mean, write our product roadmap based on that coalition of people internally, based on the lessons that you've learned. What would get you excited about what AI could do for you beyond this?

Siddhi Saraiya: So I think my mind goes in many different directions at once, but I think that there's one part which is there's a lot of the easy stuff that needs automation. It may not be very sexy and cool, but I think that there's a lot of stuff where, you know, just. I'm thinking about more automation using AI.

So I don't know, again, like with Jasper, how that would work, but I think there's an area definitely there that's like, you know, making it easier to do interactions with our. With our sales and our accounts. Right. Whether that's contractual stuff, whether that's trade, whatever that might be, just the work of that, you know. The second part is diversifying content at lower cost.

So we have many different types of marketplaces we show up in. How could we create content at a lower cost that speaks to those different audiences? For example, again, a fashion influencer has very different needs than a running specialist. How could we begin to create this diverse content for those different needs? There's some basic stuff that still takes a lot of work. Translations in a global company is really, really hard. Right. How can we create a streamlined way to do translations at scale, those types of things.

So not always the biggest problems, but also really small but very painful problems. Things like translations and. Yeah, that would be a start.

Audience Questions and Closing

Meghan Keaney Anderson: That's great examples. I've got two questions from the audience that I'll hit really quickly and they're very like, tactical. Right.

So Pablo, first of all wants to know, what marketing channels are you utilizing AI in? Is it email, is it social, is it internal digital ads? Talk a little bit about the channels that you've seen internally Adidas use.

Siddhi Saraiya: Yeah, so there are different kind of experiments and also some adoption in different areas, some of those that you've mentioned here. So for example, there are some things that are happening, like in the email space. There are some things that are happening. I know that one for sure. The other ones I would have to check again with colleagues and others. Yeah, in our space, again, there's the copy topic. There's a lot of exploration around customer service. Right. How can we better use those tools in that space?

So those are some also in, for example, in B2B sales and in an E comm site for B2B. How do you allow people to. How do you allow people to search more efficiently?

So instead of just using a search bar, filters, can you allow people to kind of use AI? Whether that's a chat functionality, whether that's connecting with what we know about that account to help them find things more effectively. That's another space. Like again, experimentation and a Lot of that is just that right now.

Meghan Keaney Anderson: And to your point, let the problems and the needs dictate the channels that you use it on so that it's, it's grounded in outcomes. Last question and then I will let everyone, including you, go with my gratitude. Nevin wanted to know how has Jasper helped with sales enablement and competitive positioning of Adidas products? If that is an area that you explored and if not, I can speak to other companies have right now.

Siddhi Saraiya: It's not something that we've explored yet. It's really kind of limited to this case but. But yeah, it's maybe in the future, I don't know.

Meghan Keaney Anderson: I, well, I'll share this. We are not retail but I will share kind of how we use it internally for those use cases. It's true. It's truly about the enablement part. Right.

So we as a company will figure out where are we differentiated, how do we compare to alternatives in the field? And, and we will store that information inside our knowledge hub, inside Jasper. But the real magic happens when a sales executive or sales rep is in their email and gets a request from a prospective buyer around, hey, how does Jasper compare to alternative A, B or C?

And they can pull Jasper in right there within their email to answer that question without having to do a bunch of legwork and go to talk to a bunch of people across the company to get that positioning. So for us it's really around getting the entire company aligned around the most current messaging and then giving every individual the tools to access that central nervous system. Okay, great. That has brought us to the end of our time. Siddi, I am so very grateful for your insights in this. I think you have been, you know, a unique and kind of pioneering role within Adidas to, you know, thoughtfully bring AI into your workflows in ways that clearly make a difference.

So we're so grateful that you are using Jasper. We're so grateful for you sharing those early lessons with the rest of us today and thank you all for coming. We will follow up with more answers to some of the questions that we missed with more information. We will likely take a transcript of this call and turn it into use Jasper to turn into all sorts of other supporting content.

So stay tuned for that and thank you all. Have a great rest of your day.