AI Commerce: The store remains, but it takes on a new role

Customers are no longer just searching for products. They describe problems, situations, and expectations. AI systems translate these queries into recommendations, thereby changing where purchasing decisions are made. This raises a new question for retailers and brands: How do you stay visible when it’s no longer just people reading content, but machines deriving recommendations from it as well?

Burkhard Richter, a digital consultant at Diconium, has been working in the fields of commerce, digital transformation, and the integration of strategy and technology for many years. In this interview, he explains why AI commerce is more than just a new tool, why online stores aren’t going away, and why data quality, trust, and internal coordination are becoming key success factors.

The real bottleneck isn’t the technology

So is this what “Agentic Commerce” means now: My customer no longer shops themselves, but lets ChatGPT do the shopping?

Burkhard Richter: That’s exactly what the e-commerce industry is currently debating intensely: To what extent will AI agents and tools be integrated into our purchasing processes? And will they eventually take over these processes entirely for us, if we want them to? That would change quite a bit for users as well as for retailers and brands.

But before we reach the point where we can legally allow AI agents to shop independently on our behalf in Europe, several regulatory hurdles still need to be cleared. The real bottleneck isn’t the technology. The bottleneck is the lack of a legally sound framework for delegating decision-making, action, and liability to machines.

The six key regulatory gaps for agentic commerce in Europe are:

1. No clear legal representation
   AI agents currently cannot unambiguously enter into binding contracts in their own name.
2. Unclear liability for autonomous decisions
   Responsibility is currently spread across a fragmented mix of the Product Liability Directive, the AI Act, and national law.
3. Lack of payment delegation
   PSD2 and PSD3 do not allow for truly autonomous execution of payments by agents.
4. No standardized agent identity
   eIDAS covers individuals and organizations, but not AI agents acting autonomously.
5. Consumer protection is not yet AI-ready
   Disclosure requirements and right-of-withdrawal provisions do not function properly when an agent acts independently during the purchase process.
6. The AI Act addresses risks, but not commerce use cases
   The focus is more on security than on the specific requirements of autonomous commerce processes.

What does work well today, however, is the use of AI for product research and purchasing decisions. Many people are already using tools like ChatGPT to prepare for more complex purchasing decisions. Once people have tried it, they often don’t want to go back to a simple Google search.

For example, I was looking for a new grill over the weekend. My requirements were: The grill should be very durable, rust-resistant, easy to clean, suitable for four people, and cost no more than a certain amount. The AI explained the differences between cast-iron and stainless-steel grates and put together a comparison chart of three grills that met my criteria. That’s pretty much how it would have gone in a specialty store. In a traditional online store, it would have taken significantly longer.

This means that, in the future, customers will be more likely to end up at a retailer or with a brand when their purchase decision is already largely prepared. They no longer arrive at the very beginning of the journey, but with a specific shortlist that they want to confirm and then finalize. This fundamentally changes the customer journey.

In addition: Initial figures already show that while the percentage of visitors coming to online stores via AI tools has risen dramatically, the overall volume of traffic being redirected is declining significantly, since AI tools often answer users’ questions directly, eliminating the need to click further.

Brands remain relevant, but their promises must become more robust

What happens to brands when the purchase decision has already been made before anyone even sees the store?

Burkhard Richter: For brands, this means, first and foremost, that customer access is changing. The initial point of contact for a specific purchase is shifting more toward AI tools. But that doesn’t mean brands are becoming irrelevant.

A brand is a promise of a product experience. Ideally, this promise takes root in customers’ minds and is confirmed by their own experiences with products and touchpoints. AI doesn’t fundamentally change that.

Brands that fail to deliver on their product promise will have a harder time. And business models that rely on customers not having access to all information and comparison options will also face greater challenges. Those who profit from a lack of transparency will find it increasingly difficult in a world of AI-powered research. But to be honest: these providers didn’t have it easy before, either.

What is changing is the point in the purchasing process where brands can establish their promise. Online, it’s becoming harder to generate this preference only during the product search. Ideally, the preference has already been established beforehand. Then the user tells their AI agent: “I prefer Apple.” Or: “When it comes to athletic shoes, I like Adidas.”

So brand preference isn’t going to disappear. However, it’s increasingly forming outside the immediate purchasing process. In the medium term, this will also have an impact on marketing activities. Perhaps traditional marketing will even become more important again in some areas, because it’s no longer just about generating attention at the moment of search. It’s more about credibly establishing the product promise beforehand and then delivering on it.

The grill example illustrates this well: For the AI agent, what matters is the vendor’s justification based on my criteria. Why is this grill more durable? Why does it rust less? Why does it better suit my needs? Answers like these come from reviews, test results, and product data—not just because there’s a logo on the product. But if a product from a brand I trust is suggested, I’m more likely to follow that recommendation than one from an unknown brand.

An important point is this: In the future, content—including brand content—will no longer be read only by humans. It will be processed by AI. This has consequences. Brands and retailers must provide the data that AI agents need to generate recommendations for users. Brand preference itself continues to originate with people, but it plays a different role in the purchasing process.

Product data must reflect intentions, not just attributes.

What specific steps do I, as a company, need to take to even be considered by AI?

Burkhard Richter: It sounds simple: I have to give the AI the right data. This data must be up-to-date—and in this context, “up-to-date” means as close to real time as possible. It must be accurate, and it must be available on my platform. In practice, however, this is complex.

For one thing, companies should make their product data—which they already have—available via an API in a standardized format.

On the other hand, this existing data is usually not enough. What an AI agent also needs is up-to-date availability information, shipping costs, delivery times, and return costs.

Added to this is contextual data. Much of today’s product data is structured to work within an online store and to be well-organized for search engines. But we ask AI about products differently. We formulate a problem, an intent, or a usage scenario.

A simple example: I’m looking for a pair of jeans made in Europe. If this information isn’t included in the data, those jeans won’t be included in the AI’s selection. So it’s not enough to simply maintain traditional product attributes accurately. Retailers must understand their customers’ needs and the intentions they enter into AI tools—and ideally, they should tailor this understanding to specific target groups.

Companies thus face two challenges simultaneously: technical challenges in their IT architecture and content-related challenges regarding data quality, data logic, and data completeness.

The next step isn’t the next tool—it’s orchestration.

Everyone’s talking about AI tools. But you say the real issue is something else: systems that work together. Why?

Burkhard Richter: AI tools are important. But the next step is moving away from isolated internal solutions and toward integration, networking, and orchestration of internal AI agents. Only then can we truly rethink processes and deploy agents across the entire customer journey.

In the future, there will be specialized agents within the company. Some will optimize product data. Others will create customer segments. Still others will assist with fraud detection in payment transactions. The key is that these agents do not work in isolation from one another.

Ideally, this requires a unified database for customer data that is fed from various systems. In the future, we will increasingly see Customer Data Platforms that enable companies to build their own AI models based on knowledge about their customers.

Agents for Commerce means: Improving processes in the value chain

If I run an online store today, do I need to start building my own fleet of agents right now?

Burkhard Richter: We referred to this in our report “Agents for Commerce.” It refers to the internal use of AI to optimize processes in the value chain. This can be useful for becoming more efficient or for doing things that would have been too expensive without AI.

Yes, it can make sense to build your own fleet of agents that perform specialized tasks and make decisions independently. A simple example is an agent that automatically translates product texts. Or an agent that creates product descriptions that were previously missing because creating them was too time-consuming or too expensive.

Another example is an agent that recognizes when certain items are about to sell out and therefore makes alternatives more visible in the store.

In principle, there are opportunities at every step of the value chain to use agents to improve processes or make them more cost-effective. That doesn’t mean every agent is automatically a good investment. But companies should assess where their use might be beneficial.

Are customers already active in this area? Are there any initial successful examples of agents working together?

Burkhard Richter: Many customers are already quite active in developing agents for their own value chain. At the operational level—for example, when using Claude or Copilot in software development—many have already made significant progress.

However, when it comes to orchestrating agents, we’re still in the early stages. Often, a unified database is lacking. There are also frequently organizational hurdles. AI exposes internal silos that have arisen due to internal structures. It is precisely these silos that must be overcome if AI is to access data effectively and support processes across the board.

AI does not arise from the pressure to increase efficiency, but it does provide new tools to address it

To put it bluntly: Is the growth hype over, and are people focusing on money again?

Burkhard Richter: Yes , in a way, it has. But that didn’t start with AI—it’s been going on for a while. The COVID pandemic brought significant revenue growth to e-commerce. After that, things went back to normal. E-commerce revenues are tending to stagnate, and if anyone is growing in Germany, it’s mainly Amazon.

Retailers are increasingly complaining about margin pressure, additional competition from China, and rising operating costs. When consumer reluctance to spend is added to the mix, the call for efficiency and cost savings grows louder.

This is not a phenomenon caused by AI. However, AI is creating new tools to reduce potential process costs.

If you could name just three use cases where AI pays off immediately, what would they be?

Burkhard Richter: The three most obvious use cases would be:

1. Software development
   AI can support and accelerate operational development processes.
2. Product data such as text and images
   AI can assist with product descriptions, translations, image variations, and missing product information.
3. Optimizing the store search for conversational commerce
   Users will increasingly get used to articulating their problems and intentions rather than simply searching using filters, navigation, and lists.

AI as a “silent employee” means: receiving data, passing it on, and triggering tasks

You once referred to AI as a “silent employee.” What exactly does this employee do every day?

Burkhard Richter: What do AI agents do all day? Ideally, they exchange information with colleagues, receive data, pass data along, and do so without using too many expensive tokens.

AI agents are set up for specific tasks. What makes them special is that they can act independently and make decisions. For example, an agent responsible for translating product data receives a technical alert as soon as a new product is ready for translation in the PIM. The completed Italian translation can then trigger an image agent, which generates new images for the Italian target audience.

This clearly illustrates what it’s really all about: not just automating individual tasks, but connecting processes with one another.

AI can be a real opportunity, especially for small and medium-sized businesses

Is this more of an issue for large companies, or is it a realistic possibility for small and medium-sized businesses as well?

Burkhard Richter: It’s definitely a topic for small and medium-sized businesses as well. At a conference, I met the CEO and owner of a company that sells tools. He presented a truly innovative yet pragmatic AI use case. The company had seven employees, four of whom were family members.

I’d even go so far as to say that AI is a huge opportunity for small and medium-sized businesses to catch up, because the tools are available to everyone. Large corporations often face the challenge that resolving governance issues takes longer. In certain cases, smaller companies can experiment, learn, and implement solutions more quickly.

Deleting the online store would be the wrong conclusion

Now, here’s a provocative question: Will I actually be able to delete my online store soon?

Burkhard Richter: Quite clearly: No. You shouldn’t do that.

First, the transition to true Agent-driven Commerce—where agents shop on our behalf independently—still needs some time. It’s an evolution, not a revolution.

Second, this development won’t be equally relevant for all industries and product categories. Some sectors will be more affected, others less so. To name two extremes: Would you let an agent select and purchase your wedding suit for you? Maybe the black socks to go with it—that’s more likely.

Third, the functions of an online store will change. It will increasingly become a customer and service portal. It will also take on a greater role in communicating the brand promise.

Fourth, recommending products is easier than selling them. AI doesn’t simply handle taxes, shipping costs, returns, and liability as an afterthought. Even ChatGPT realized that an integrated checkout is complex and has since discontinued it. Having your own storefront will still be necessary, including for checkout and fulfillment.

What do online stores need to do differently today to remain relevant?

Burkhard Richter: The role of online stores will change once they are no longer the first port of call for purchasing decisions. Stores need to demonstrate more clearly that they deliver on their product promises. And the purchase doesn’t end with the order. After-sales support, customer service, and loyalty programs are becoming more important.

Until now, online stores have primarily been optimized for human conversion. Now, machine-driven conversion is coming into play.

Cross-selling and upselling may also need to be rethought. AI systems recommend individual products, but do not automatically suggest complementary products or accessories. Different mechanisms are needed here to continue increasing average order size and customer lifetime value.

Another key point is guiding customers to the right product within the store itself. AI can help here as well. Users will get used to simply describing a problem and then receiving recommendations. They’ll have the same expectation of an online store. Stores will therefore be served more through conversations and less through navigation, filters, and sorting lists.

And honestly: Is this a threat, or is it actually an opportunity?

Burkhard Richter: Both. It’s both a threat and an opportunity.

The threat manifests itself on various levels. On the one hand, provider websites will see a significant drop in traffic because many answers are already being provided by AI, and users no longer click through. This particularly affects business models that are financed by advertising. That’s a massive threat.

Another threat affects business models that thrive on a lack of transparency—that is, on the fact that customers cannot easily compare options. Brands that lack genuine differentiation and have primarily bought attention through large marketing budgets will also come under greater pressure.

But it’s also an opportunity. The traffic that an AI system directs to a store is of higher quality and converts more frequently. Brands with strong differentiation and a lot of positive customer feedback can benefit because these signals are relevant to AI systems and can influence recommendations.

Not every shopping situation should be automated

Listening to you right now, one might think: Soon everything will be automated. But where do you draw the line?

Burkhard Richter: No, I don't think everything will be automated anytime soon. We're currently experiencing a technological advancement that's bringing about incredible changes, both in our professional and personal lives. But here we're talking about AI Commerce and how AI can assist with shopping and selling.

When it comes to shopping, I think many people don’t actually want an AI agent to take care of everything for them. Shopping is often an emotional process. Of course, that depends heavily on what you’re buying. I’d much rather pick out my new bike myself than be surprised by AI at some point. With socks, though, it might be a different story.

On the one hand, I draw the line at the degree of automation. It doesn’t always have to go all the way to the point of purchase. The most valuable assistance often lies in product research and support with the purchasing decision. On the other hand, the line is drawn at the selection of use cases. Not all shopping situations are the same.

Where do you draw a clear line and say, “AI should definitely stay out of this”?

Burkhard Richter: There are critical processes where humans should be involved. This applies to medical issues, but also to certain service-related matters. Klarna, for example, has backtracked and rehired people in the service department following a wave of layoffs caused by AI, because customer acceptance had suffered.

Particularly problematic are poorly designed solutions in the service sector that ultimately fail to help. You describe your problem in detail, don’t get a solution, and end up on hold—often with a delay. In that case, AI isn’t a better experience—it’s just an additional source of frustration.

Trust is crucial on multiple levels

How important will trust and brand become when products are compared solely based on data?

Burkhard Richter: The issue of trust is extremely important—on several levels.

First, with AI, we gain an assistant to whom we can delegate tasks. Until now, that was only possible for a privileged few. But a secretary is only helpful if you trust that person. It’s similar with AI. For AI Commerce—and perhaps even true agentic commerce—to take hold, there needs to be trust in the AI. If I can’t assume that product recommendations are valid and based on a comprehensive dataset, I won’t use them.

Second, the AI must be able to trust the data it receives. If it has no indication that the data is reliable and up-to-date, it will rate it lower or not even consider it in its selections.

It’s similar with brands. Brands that have a clear, distinctive brand promise—and actually deliver on that promise—will continue to succeed. In the future, users will tell their AI agents what their brand preferences are.

Brands must therefore continue to build trust. Other communication channels may become more relevant again for this purpose. After all, one intriguing question remains: How does a brand create a need that the customer isn’t even aware of yet? That is precisely where brands will remain relevant, even in an AI-driven commerce world.

The first step is a simple reality check

If you’re in charge of e-commerce today: What would be your first concrete step first thing tomorrow morning?

Burkhard Richter: I would analyze whether and how my products are found and recommended in ChatGPT, Claude, and Gemini. That’s a very concrete starting point because it shows whether your own products even appear in AI-supported decision-making processes.

What’s the biggest mistake companies are making right now as they jump on the AI bandwagon?

Burkhard Richter: The biggest mistake by far is not engaging with it at all. Fortunately, very few companies are doing that anymore. The second mistake is blindly chasing every trend without a plan. The third mistake is failing to view the use of AI from a broad strategic perspective and instead allowing technical silos to proliferate unchecked.

Maybe we’ll soon be laughing at the old-school store navigation

In two or three years, what will we be laughing about today?

Burkhard Richter: Maybe we’ll laugh about how difficult it used to be to find our way around a large online store. And maybe also about how long some IT projects took.

AI Commerce is therefore not just about a single tool. It’s about visibility in new decision-making processes, better data, trust, more efficient workflows, and the question of what role the online store will play in the future. The online store is here to stay. But it will increasingly serve as a service portal, a source of trust, a transaction hub, and a machine-readable data source all at once.