AI Agents in E-Commerce: From Automation to Autonomous Decision-Making

Which tasks are AI agents most likely to take over at scale in e-commerce over the next 2–3 years? What will they already be ready for next year?

Pascal Beij, Chief Commercial Officer at Unzer: There’s a well-known quote often attributed to Bill Gates: “We always overestimate the change that will occur in the next two years and underestimate the change that will occur in the next ten.” It’s hard to predict just how profoundly AI will reshape e-commerce. Only three years ago, hardly anyone in Germany knew what large language models were. Terms like OpenAI, generative AI, or ChatGPT were largely unfamiliar. Today, there’s almost daily news about what AI can do.

So it’s an interesting question, but ultimately a bit like reading tea leaves. That said, I’m convinced AI agents will reliably take over standardized tasks such as inventory optimization, automated reordering, and recurring transactions. Within the next two to three years, they’ll increasingly become part of everyday online shopping.

How can we ensure AI agents don’t hallucinate or make bad decisions due to insufficient data? For example, overstocking warehouses or ordering useless products. The Wall Street Journal wrote: “AI agents can go rogue.”

Pascal Beij: Context is still everything. When AI agents draw the wrong conclusions, it’s almost always because the underlying data is incomplete or inaccurate. In those cases, the data foundation needs to be fixed. It will likely take time before binding regulations are in place across Europe. We’re still in an early phase. In the meantime, we’ll probably see many individual initiatives from retailers, payment providers, and manufacturers aimed at protecting themselves from hallucinating AI agents.

Some imagine AI agents shopping online on behalf of consumers. But given that tools like ChatGPT can still struggle with simple math or occasionally produce incorrect information, are digital shopping assistants overhyped?

I don’t think so. In the US, some of these agents are already available. Amazon is testing “Buy for Me,” the US AI company Perplexity AI launched a similar assistant late last year, and OpenAI has recently followed suit. For simple, repeat purchases, AI agents will soon be reliable. As the technology matures and gains experience, more complex purchasing decisions will also become automated.

What mid-term future scenarios do AI agents open up for e-commerce around 2030?

Pascal Beij: E-commerce will become significantly more customer-friendly, automated yet highly personalized, multilingual, and real-time. Agents will negotiate directly with each other, optimize prices dynamically, and make proactive purchases before customers even express a need. Hyper-personalized offers will become the norm, and comparison portals as we know them today may disappear.

How autonomous can AI agents become? Could you assign them tasks for several months without constant monitoring, like maintaining optimal inventory levels or adjusting product assortments?

I believe so. The well-known research and advisory firm Gartner estimates that by 2028, AI agents will make around 15 percent of all business decisions. Their use cases are as diverse as e-commerce itself, from pricing optimization and assortment management to competitive analysis. Even today, AI agents in customer support can handle 50 to 90 percent of inquiries on autopilot. That said, there are still risks, especially around governance. In practice, these agents will continue to be closely monitored to prevent issues like poor purchasing decisions or excess inventory.

Google’s AI mode is available in the US and UK. What impact will it have on e-commerce? Will there be a rush by vendors to appear in the AI box? How can companies position themselves there?

Pascal Beij: Unlike traditional search results, Google’s AI mode delivers curated shopping answers directly within search. At its core is a product database. Only retailers who list their products with Google will appear in those results. This requires structured, machine-readable product data and high-quality images. Retailers need to prepare for Google acting more like an intermediary or even a gatekeeper. If your data isn’t easily accessible to Google, you risk becoming invisible. At the same time, technical integrations will be necessary so AI systems can complete purchases seamlessly. As with many AI developments, we’re in a trial-and-error phase. A lot will be tested and refined quickly, so companies need to stay agile and keep up with changes.

In Google’s AI mode, brand names aren’t prominently displayed and results look more uniform, similar to Pinterest. How can brands still differentiate themselves? Or does branding become less important?

If AI agents start shopping on our behalf, brands will need to communicate their value more through objective quality signals like product specs, performance metrics, certifications, and verified standards. We already see this trend in reviews and ratings. In the future, differentiation will rely more on proven quality, transparency, and trust rather than pure image advertising. That said, I’m convinced brands will remain important. People look for orientation and identity, so influencers, testimonials, and strong brand storytelling will still matter. There’s also a possibility that in an increasingly AI-driven world, we’ll see a countertrend toward offline and authentic experiences. Brands can play a key role in building trust in that context.

Do we still need retailer webshops in an AI-driven world? Wouldn’t direct-to-consumer shops be enough? Should platforms like Amazon, Zalando, or Temu be worried?

The traditional webshop will evolve, but it won’t disappear. Manufacturers already sell directly, yet marketplaces have only grown in importance. They offer huge product ranges, fast delivery, easy returns, and more. I expect two parallel trends. On one hand, many retailers will build their own AI agents for their assortments. On the other, marketplaces like Amazon, Zalando, and Temu will remain attractive thanks to their scale, brand presence, and logistics.

Which business areas around AI agents could see a boom, and what new business models might emerge?

Let me answer with an anecdote: I recently saw a clip from Back to the Future Part II, which is set in 2015. Cars and skateboards can fly, yet fax machines are still everywhere. At the risk of being completely wrong, I believe anything that ensures trust, transparency, and data sovereignty in agent-based systems will grow. Customer service will also become even more central.

For AI agents to act on behalf of consumers, they need to know them well. Could we realistically see perfectly tailored shopping lists in ten years, generated without detailed prompts? How would AI access the necessary data?

Yes, that seems very likely. We’ve been talking for years about fridges that reorder milk automatically. But it goes further. For example, I need a new kitchen, and even today ChatGPT can suggest the necessary modules from IKEA. In a few years, agents might detect that we need a break by analyzing calendar data, sleep patterns, and activity levels, and then automatically book a spa hotel. Of course, data privacy and secure data sharing will be critical to building the trust needed for this to work.

About Pascal Beij

As Chief Commercial Officer (CCO), Pascal Beij plays a key role in driving Unzer’s growth strategy. Under his leadership, Unzer is building an integrated ecosystem to support merchants throughout the payment process and accelerate their digital transformation. The goal is seamless, easily integrated payment and software solutions to help businesses digitize their processes and meet changing customer expectations. Unzer focuses especially on industry-specific software and payment solutions for small and mid-sized companies. Before joining Unzer, Pascal was Senior Vice President Retail for Central Europe at payment service provider Planet. With over 20 years’ experience in payments and retail, he is widely recognized as an industry expert and thought leader.