AI agents are bypassing the ecommerce funnel entirely — researching, comparing and completing purchases before consumers ever visit a website. Most product and ecommerce teams are still optimising for a model that is no longer the primary decision path for a growing share of purchases.
Someone asked me recently where I think ecommerce and online businesses are heading with AI. My honest first reaction was that I didn’t know. Not because the question is unanswerable, but because I’ve spent the last while so deep in delivery — firefighting, short-term fixes, shipping things under constraints — that I’d stopped asking it.
That’s worth naming. Sustained delivery pressure is good at one thing: it kills your peripheral vision. You get faster at executing on the current model without ever questioning whether the model itself is still the right one.
So I sat with the question. Here’s what I actually think.
The funnel you’re optimising doesn’t exist anymore
For the last decade, the mental model for ecommerce growth has been roughly the same: get people to your site through search or paid channels, guide them through a structured journey, convert them at checkout. The metrics are familiar — traffic, bounce rate, session depth, conversion rate, average order value. The optimisation levers are familiar too.
The strategies that drove that model are no longer sufficient — not because teams are executing badly, but because the funnel itself is no longer the primary decision path. Most teams are improving the wrong thing.
Consumers now use ChatGPT and Gemini to research products before they ever visit a store. Agents are completing purchases autonomously. The browsing session — that familiar arc of awareness, consideration, comparison, conversion — is being compressed or bypassed entirely.
ChatGPT launched transactional checkout capability in late 2025. Google followed with its own commerce protocol in early 2026, backed by a coalition including Walmart, Target, and Shopify. These aren’t pilots — they’re infrastructure bets by the two largest traffic sources on the internet.
The funnel didn’t disappear overnight. But for a growing category of purchases — routine, repeatable, price-comparable — it’s already structurally irrelevant.
Three things are changing at once
It’s tempting to pick one thread and pull it. But the honest picture requires holding three shifts together.
1. The interface is changing
Your website used to be the primary interaction surface. Increasingly, AI agents sit between the buyer and the brand — acting as procurement copilots, research assistants, and embedded recommendation engines before humans ever visit a website.
Brands are no longer marketing only to people. They’re being evaluated by systems.
2. The funnel logic is inverting
In the traditional model, you competed for attention at the top and converted at the bottom. In an agent-mediated world, the work that matters is not top-of-funnel content — it’s the infrastructure underneath: product data completeness, schema markup, feed hygiene, checkout API reliability, server-side tracking. The channel rewards operational rigour over editorial volume.
3. Personalisation is becoming structural, not tactical
Agentic AI helps shoppers by remembering preferences, maintaining continuity, and improving recommendations over time. The implication for product teams is significant: the relationship between brand and customer is increasingly mediated by a third party with memory. Your CRM doesn’t have that. Your personalisation engine probably doesn’t either.
The trap most teams are already in
Most product and ecommerce teams right now are doing something that feels productive but is quietly misaligned: they’re optimising for a world that’s ending.
They’re A/B testing checkout flows on websites that fewer high-intent customers will visit. They’re investing in SEO content strategies built for search engines that are rapidly becoming less relevant to purchase decisions. They’re building personalisation features on top of architectures that weren’t designed to be machine-readable.
If an AI agent has already decided your product doesn’t meet its criteria — incomplete data, weak trust signals, missing schema — no amount of checkout optimisation saves you. You were never in the conversation. The work that determines whether you win or lose is happening upstream of your website, in infrastructure most product teams haven’t touched yet.
And here’s the structural problem: while 88% of organisations now report using AI in at least one business function, most are still in the experimentation or pilot phase, with only about one-third scaling AI programs across the enterprise. There’s a lot of activity and not much transformation.
The teams that will be fine are not necessarily the ones with the biggest AI budgets. They’re the ones that have asked the uncomfortable question: are we optimising for the right model?
What product teams should actually do
I’m wary of listicles that offer easy answers to structural problems. But there are a few concrete reframes worth making.
Treat your product data as a product. If an AI agent can’t accurately interpret your catalogue — missing fields, inconsistent descriptions, no structured data — you’re invisible in the channels that are growing fastest. This is an infrastructure problem, not a content problem, and it needs to be on your roadmap with the same seriousness as your checkout experience.
Stop measuring what’s comfortable. Last-click attribution made sense when the journey was linear. Revenue operations teams must now adapt attribution and reporting to an environment where AI intermediates key interactions. If you’re not measuring assisted conversions and agent-driven referrals, you’re flying blind on the most important emerging channel.
Build trust signals that machines can read. Reviews, return policies, delivery reliability, price consistency — these were always important. Now they’re also the inputs that determine whether an AI agent recommends you or your competitor. Only 46% of shoppers fully trust AI recommendations today. The brands that earn agent trust early will have a durable structural advantage.
Why product teams stop thinking strategically — and how to start again
I started this post by admitting I’d stopped thinking strategically. That’s not a failure of intelligence — it’s what sustained delivery pressure does to product thinking. You optimise locally and stop asking whether the global model is still correct.
The question someone asked me was a useful interruption. The funnel isn’t dying slowly in some theoretical future. Consumer behaviour changes that took 10+ years during the rise of ecommerce are now transforming in 12–24 months just as dramatically, if not more.
The product teams that will navigate this best aren’t the ones that predicted it perfectly. They’re the ones that created enough space in their week to notice it was happening — and asked whether their roadmap reflected reality.
That’s harder than it sounds when you’re firefighting. But it’s the work.
TL;DR
- AI agents are increasingly completing purchases without visiting ecommerce websites — the traditional funnel is being bypassed, not just disrupted
- The work that determines whether you win in this environment happens in infrastructure: product data quality, schema markup, feed hygiene, checkout API reliability
- Most teams are optimising the wrong thing — A/B testing checkout flows while their product data is invisible to AI systems
- Trust signals (reviews, return policy, price consistency) are now machine-read inputs, not just human ones
- The teams that adapt fastest are not those with the biggest AI budgets — they’re the ones asking whether they’re optimising for the right model
Delphine Ragazzi is a Product Owner with 20 years of experience across digital analytics, CRO, and product delivery. She writes about product decisions, data, and AI at douli.com.
