How do you successfully adopt generative AI when it’s evolving at breakneck speed? How do you deliver relevant recommendations for shelves that span 18 million products? And how will you work with consumers’ personal generative AI assistants outside of your ecosystem while continuing to optimize your AI-powered search and shopping tools?

These are the monumental questions that leaders at the Otto Group, an international retail, logistics, and financial services company, are grappling with as they scale generative AI. Dr. Sebastian Walter, vice president of Digital and Consulting at Otto Group, which includes brands like Crate & Barrel, OTTO, Bonprix, and Hermes logistics, and OTTO Management Board member Dr. Boris Ewenstein, who oversees Retail and Marketplace for the largest German online retailer, are forming a response to this disruptive technology. They recently sat down with Miltiadis Athanassiou, head of Bain’s Retail practice in Europe, the Middle East, and Africa (EMEA), to discuss generative AI’s potential to rewrite the online shopping experience and how the e-commerce player can seize the opportunities ahead.  

Q: When it comes to AI and generative AI, what use cases has the Otto Group developed and started scaling? What are you most excited about?

Dr. Sebastian Walter: The Otto Group sees generative AI as core to the business. AI is part of our three strategic pillars—performance, innovation, and sustainability. The innovation pillar is strongly focused on generative AI.

This is demonstrated by the fact that we were among the first Germany-based companies to launch an internal Otto Group GPT. For almost 30,000 employees, we offer low-threshold, secure, and data protection-compliant access to generative AI for text and image generation, knowledge and compliance management, analytics, and more.

I am most proud about the momentum we gained by launching so many generative AI-based initiatives: We’re launching shopping advisers, both for customers online or for our instore staff. Our customer care bots are generative AI-driven. We have image generation for product backgrounds and we’re tackling product attributes via image recognition. We’re developing our campaign headlines as well as personalized newsletter subject lines with generative AI. We have established major commercial agreements with both Boston Dynamics and Covariant, using AI and robotics in order to automate our warehouse operations. We’re testing how we can leverage generative AI to manage the complexity of reporting on environmental, social, and governmental requirements. And these are just a few examples.

Dr. Boris Ewenstein: There are also more use cases where we get real value from generative AI already—things like scanning customer reviews for fraudulent reviews and partner content, especially images, for noncompliant content.

We’re also experimenting with deaveraging content creation to capture some cost efficiencies through synthetic content while repurposing budget toward elevated content, where it really adds value, especially profitable articles or articles that are bestsellers, to drive sell-through.

The things that really rock are customer facing—things that improve the customer experience. The way this plays out in OTTO is through semantic search. Think about buying something like running shoes. You might say, “I’m looking for a pair of running shoes. I have bad knees.” Something needs to translate a spec of materials, including “polyurethane foam” or “carbon midsoles,” into “is this a good product for someone who has knee problems?” So, semantic search is a big one.

Q: How do you measure success?

Walter: It depends on the use case. Key performance indicators are set by the business topic we aim to solve. It works easily when you have a direct comparison or A/B tests. A great example is chatbots: You measure cost per response, final resolution rate, customer satisfaction, and have a pretty good understanding. These save us a double-digit million-dollar amount, every year. Likewise, ads with or without AI-generated backgrounds are relatively easy to measure.

There are other use cases which are more challenging. For example, with the Otto Group GPT, obviously, there are qualitative questions we ask the user group. And over 80% say, “Yes, it increases my productivity meaningfully.”

However, generative AI solutions will come fast, no doubt. I’d rather have us working on edge cases where, as of today, it’s barely working, and success is hard to measure, but we generate early learnings and put them into place. In a year and a half, the tech will have advanced so much that there’s a high likelihood it’s working well. Then you’re already there—instead of just starting and being outpaced by the market.

Q: What was hard about getting generative AI to become part of your DNA? What were the most critical unlocks?

Walter: We said there are four important topics we need to nail. First, creating energy: top-down encouragement, conferences, experience-sharing events, trainings, prompting workshops, and so on. Just making the whole company aware a massive wave is coming.

The second is creating and promoting individual benefits—for example, providing a GPT that allows employees to upload confidential internal data and get analytics. It’s all to make individuals test and try, lose the fear, and see and realize the benefits.

The third and big pillar comprises the strategic use cases. We have derived a total of six focus areas, ranging from marketing and product development to logistics and customer care. Our Group companies identified many opportunities, and for some, the Otto Group Holding offered workshops to do so. Group companies have put generative AI targets in midterm plans. Then, the Group companies initiated remarkable efforts to put the use cases in place. In parallel, we developed some of these directly in the Holding, when we saw potential synergies.

From a Holding perspective, it’s that blend which worked well: providing ideas and some concrete developments or SaaS solutions centrally, while handing the baton to the Group companies to develop and deploy individual use cases on their own.

The fourth is foundational: What is our AI strategy? What are our ethical rules? What about compliance? What are the legal rules? To give you an example: Together with experts across the Group, the Tech Strategy developed a Responsible AI Guide, which serves as a value-based compass for decision making for new AI projects.

Ewenstein: The root cause of why adoption is so hard is the speed of development. If you try to get the whole company familiar with using something like ChatGPT 3.5, by the time everybody’s had a chance to prompt, we’re already working with totally different models that can produce much more compelling results.

Trying this out at scale, you have outcomes in big buckets. You have results that are phenomenally helpful to totally wrong, or borderline risky, or so generic it actually adds zero value. This is a really hard concept to embrace. You have to work with that and cherry pick the use cases where it is really helpful.

Q: Five years down the road, what role will generative AI be playing? What are the most critical things to get right?

Walter: We’ve been talking about this for the last 10 to 15 years, but now it has the chance to properly get into play: the shopping adviser or the personal adviser. For high involvement products, people will continue to search on their own. But if it’s standard products, they may not. In a few years, on the way home, you order a pizza with your personal generative AI. It knows your history, what kind of pizza you want, and the two pizza shops by your home. It does the ordering for you and speaks, on the other side, to a generative AI machine.

If you look at e-commerce and logistics, there’s a scenario where a shopping adviser is just one step away from being a fully personalized assistant. If that comes into play, it changes the value chain: Consumers will shop via their assistants, not via search, and then on our app or webpages. So, the question is, to what extent can we be the ones who remain the most relevant when it comes to e-commerce?

Ewenstein: Exactly. The thing that has value is the thing that knows you best. And the thing that knows you best is probably not going to live within the ecosystem of one particular retailer, even if that retailer is a generalist. The thing that knows you best probably lives in the stack on your personal operating system, very close to your day-to-day use. So then, how do you become agent readable? How do you establish APIs that allow a deep link to your catalog, to find you and retrieve, say, a product detail page? Then, the deeper question is, how do you become part of that conversation that individuals have with their agents in their personal ecosystem?

Related

The Gift of Personalization: AI for Holiday Marketing Magic

Winning retailers are using AI-powered personalized marketing to delight shoppers this holiday season and beyond.

Walter: As so many people say, this is the most disruptive technology, potentially of all our lifetimes. We’re at the very early beginning, like where computers were before the internet. There’s so much more to come. Relevance can’t be underestimated for our internal processes and how we do things, for the interactions with our customers and what we offer them, but also, for customer expectations. There are things which are now less smooth in the customer journey, and which I could imagine will not be accepted anymore a few years from now. Not speaking, of course, of the global macroeconomic impact.

Ewenstein: The one-billion- or one-trillion-dollar question is “where is the quality bar?” For this to be useful, to really, truly be embraced by customers, it has to be intelligent, perfect, bordering on artificial general intelligence. Well, that’s one perspective. Or is the threshold at which this unlocks impact much lower than we think? Because most problems are actually quite solvable, just not for the customer from where they sit at that moment in time. So, the question, ultimately, is “what is the tipping point?”

How do you believe generative AI will create competitive advantage in retail in the future?

Ewenstein: I think two things are important. One is relevance. The word “personalization” almost feels outdated. But imagine this infinite shelf of about 18 million or so product variations on OTTO alone. The customer doesn’t want to sift through thousands of product detail pages until they land on something that might be vaguely relevant. They want relevance right away.

The customer shares their personal interests through activities on the site. To respond with relevance to these signals, to master that art, is a big deal. Don’t forget, this doesn’t happen in some sort of theoretical ideal state. You’ve got other competing business models and interests that are crowding the space.

For example, you want to monetize advertising opportunities. Now, advertising, if relevant, isn’t such a pain point in the customer experience. It can even be helpful. Customers tend to welcome advertising that is relevant. They don’t experience it as advertising, in some cases. But if you get it a little bit wrong, the experience can be extremely detrimental, and you lose customer interest, and ultimately, they bounce. So, relevance is one thing to nail in the next few years.

The second one is that a company like OTTO has to nail expertise. If you think about entering a shop, and you’re looking for a new mattress or a washing machine, what you really need is expert advice. You’re trying to make a decision: What’s the best value for money I can get? In a physical environment, it’s very easy. A person will listen to your needs, synthesize what they heard, and then advise you toward a particular product. If they do it well, they will uptrade you. The customer will walk out feeling particularly satisfied because their unique value-for-money equation has been optimized. If you can get that to happen in an online environment, it’s a major competitive advantage.