Nets’ E-commerce Expert Patrik Müller shares the dos and don’ts when the AI chatbots are in charge of the consumer journey. Plus: Five retail trends that stood out at NRF in New York.
As part of the European payments company vNexi Group, Nets helps merchants get paid across online, in store, and in app, and provide the infrastructure around cards, wallets, and other payments.
– In the Nordics, we’re close to merchants and consumer behaviour, so we often spot shifts early in real transaction patterns as well in the surveys we do, says Patrik Müller, E-com Expert, who’s translating those shifts into practical guidance for ecommerce and retail.
What do your customers ask you about the most right now? And what do you answer?
– Two questions that are pretty common nowadays, ‘How do we not lose customers if discovery moves into AI?’ and ‘What’s happening on the payments side for agentic commerce?’
– For AI, accept the shift. Then, get serious about structure and product data so you’re readable for computer-to-computer commerce. Also, be honest about what you lose if the customer is not in your own UI: brand experience, loyalty mechanics, and direct relationship. Maybe you need to find other ways to do this?
– For payments in agentic commerce, there is real work going on, but it isn’t completely solved for Europe yet. On the standards side, the payment protocol AP2 is one important piece for agent friendly payments that we are supporting. And we’re building an MCP server as a hub to connect to different agents plus we also have committed together with Google to support UCP as well.
For agentic commerce, Müller continues, there are a couple of misconceptions which come up all the time.
– First, people talk about ‘agentic commerce’ like it’s one single thing. In reality, it’s a spectrum. At one end, GenAI helps you shortlist and compare, still very human driven. The end state is agent to agent, where your personal agent can negotiate with the merchant side, logistics and payments, end to end.
– Second, there is a huge difference in the rule systems between Europe and the US; trust, identity, permissions, and authorisation are very regulated in Europe, which means that we will stay behind the US, as far as I can judge.
We might stay behind other markets, but, given the regulations here and other obstacles, when is it realistic to expect ‘real’ agentic commerce also here?
– Europe has what I sometimes call ‘the European problem.’ We have tougher requirements around identity, risk minimisation and the Strong Customer Authentication, and that’s a good thing, but it means the path will be incremental.
– Near term, we have the assisted commerce. The agent helps you choose, configure and prepare the purchase, but you still do the actual purchase at the sellers website.
– Medium term, more autonomy will come. This grows as standards, such as UCP, mature, and the first step will be the ability to do the full purchase within the AI assistant. Agent-to-agent is probably a few years away for consumer purchases. There’s also some uncertainty when it comes to the human factor. Full automation sounds nice, but most people probably still want the last click. The experiment with Amazon’s ‘Dash buttons’ a couple of years ago is a good reminder of that.
Müller recently went to NRF 2026 along with around 40,000 other attendees, which he describes as a ‘proper reality check for retail.’
– There were about 200 speakers and 1,000 exhibitors at Javits Center in New York, so you get both strategy and what is already live in the market. In the sessions I attended, AI was mentioned 230 plus times, which tells you where the mindset is right now.
– If I boil it down, there were five big trends:
1. Discovery is shifting fast. One stat that stuck with me is that 58% of consumers say they have replaced traditional search with GenAI tools. And referrals from GenAI tools to retail sites were up 4,700% year over year in the data point I saw (July 2025). That pace matters, because it change fast where the customer journey starts.
2. Google clearly wants to sit in the middle of this. Sundar Pichai (pictured) attended for the first time and talked about how retail related AI usage has exploded, with a figure of 11,000% growth in retail related tokens processed from Q4 2024 to Q4 2025. He also introduced an important standard called UCP, Universal Commerce Protocol, basically giving AI clients structured commerce data computer to computer, instead of trying to interpret human web pages. This is a biggie!
3. AI monetisation is starting. OpenAI is testing ads in ChatGPT, and Shopify launched support for ‘ChatGPT checkout’ with an extra 4% fee on top of the standard 2.9%. You can almost call that an AI tax. It’s a signal that new toll booths will appear in the commerce flow, or at least be tested.
4. Physical retail gets an unexpected tailwind from digital fatigue. One stat I saw was that 81% of US Gen Z often wish they could be more disconnected from digital devices. That matches what you see on the ground, more focus on vibe, community, and sensory experience. The shoe store Golden Goose is a good example, with its in-store customisation where staff can spend around 40 minutes per pair to do individual decorations.
5. Logistics speed is becoming a behaviour driver. The drone delivery example was interesting. Google owned Wing is already at roughly 750,000 deliveries, and they forecast around 40 million US customers having access by 2027. Walmart is rolling it out, aiming for 150 stores in 2026, and some locations are already doing around 100 to 200 drone deliveries per day.
If you look at beauty brands, how can they prepare for this shift already now? Any dos and don’ts?
– Beauty has an advantage here because guidance and trust matter, but it also raises the bar on product information.
– Dos:
1. Make your product data structured and brutally clear: ingredients, allergens, claims, shades and undertones, finish, usage, routine step, certifications. If core facts are hidden in images, you’re basically invisible to AI clients.
2. Make availability and promises easy to read: stock, delivery options, price, bundles, and return rules. Agents will be biased toward what they can fulfil confidently.
3. Prepare for the interface not being yours. The customer will stay inside an AI client more of the time, thats a fact.
4. Build guided shopping that actually helps. Beauty is perfect for routines, skin type logic, and decision trees. Personalisation is also moving from “recommendations” to “creation”, I saw examples where a questionnaire can generate personalised products like custom fragrances.
– Don’ts:
1. Don’t assume an AI client will “figure it out” if your site structure is messy. I had a real example recently where a product was easy for me to find, but the AI assistant could not find it at all.
2. Don’t wait to investigate how you, or your e-commerce platform supplier, can start this journey already by now.
– One thing to add to this is that most brands don’t need a new trend. They need brutal focus on the fundamentals. The biggest lift often comes from fixing hygiene factors like product pages that answer real questions, fix site performance, fewer checkout errors, clear terms for delivery—which need to be very flexible!—and returns, and of course making sure that you are offering the right payment methods in your checkout.
