Agentic Commerce: Your Next Customer Is an Agent. Is Your Store Ready?
A 2026 IBM study found that 45% of consumers already use AI for at least part of the buying journey. ChatGPT Shopping is live for every US user, with Etsy and more than a million Shopify merchants connected. Google and Shopify co-authored one commerce protocol, OpenAI shipped another, and McKinsey now sizes the agentic commerce opportunity at three to five trillion dollars by 2030. Underneath the numbers sits a plain shift that most merchants have not priced in yet. The customer arriving at your store is, increasingly, software acting for a person, and software shops differently.
We watched an early version of this from the inside. Botilia, the largest Greek wine retailer, is where our AI stack grew up, and e-commerce is the industry we know at the shelf level. So this piece is the merchant-side view of agentic commerce. What actually changes when an agent does the buying, which parts of your store it can and cannot see, and the work that decides whether agents pick you or your competitor.
What agentic commerce actually is
An agentic purchase starts with a goal, a person telling their assistant to find a gift, restock the pantry, or source a part. The agent plans the steps, queries merchants and marketplaces, compares options against the person's constraints, pays within limits the person pre-approved, and hands back an order confirmation. The distinction from a recommendation engine is the ending. A recommender suggests. An agent transacts, and money has moved by the time the human looks up.
Two standards are racing to become the rails. UCP, co-developed by Google and Shopify with a council that includes Amazon, Meta, Microsoft, Salesforce, Stripe, Target and Wayfair, covers the journey from discovery through checkout and post-purchase. ACP is OpenAI's protocol, launched with Etsy and now behind ChatGPT Shopping. Perplexity settled on PayPal, and Amazon runs its own assistant on proprietary rails. For a merchant the takeaway is less about picking a winner and more about the fact that every major discovery surface now has a transaction path attached. Where products get found is becoming where products get bought, with no click to your site in between.
The agentic purchase: a goal in, an order out, and the merchant chosen on machine-readable substance.
Software shops differently
Everything merchants learned about persuading a human weakens when the shopper is an agent. An agent does not feel the hero image. It does not linger on the brand film, respond to urgency banners, or get charmed by the copy your team polished for a month. It reads structure. Product attributes, prices, stock, shipping costs, delivery windows, return terms, reviews it can parse. Then it compares those across every store it can reach, ruthlessly and in seconds, and presents its human two or three options with the reasoning attached.
This is a demotion for the storefront and a promotion for the data underneath it. The product feed, the structured attributes, the machine-readable policies, the things e-commerce teams treated as plumbing for a decade, are becoming the merchandising. A store with beautiful pages and a thin feed is, to an agent, a thin store. A store with complete, accurate, structured data is a store the agent can confidently buy from, and confidence is what the agent is optimising when it chooses where its human's money goes.
Being citable is becoming being buyable
The discovery half of this is the same shift we have been writing about all year under a different name. Agents find merchants the way AI answer engines find sources: through structured data, entity clarity, and content that machines can verify. The work we described for generative engine optimization on your website and in MCP, clearly explained is the same muscle. A merchant whose products, policies and availability are machine-readable is simultaneously easier to cite and easier to buy from. One investment, both outcomes.
There is a verifiability point hiding here too, the same one from our test for what AI automates. Agents favour claims they can check. A shipping promise stated as a structured field with a track record beats a banner that says fast delivery. Over time, merchants build a reputation with the agent platforms themselves: order defect rates, fulfilment accuracy, return friction. That reputation compounds the way seller ratings did on marketplaces, except now it decides visibility across every agent surface at once.
What this means for your CRM and your margins
Two second-order effects deserve a CFO's attention. First, the relationship moves. When the agent intermediates, the customer relationship risks collapsing into a data relationship: the platform knows the person, you fulfil the order. The counterweight is service worth remembering, loyalty the agent is told to honour, and post-purchase experience good enough that the human tells their agent to prefer you. Predictive CRM work, like the behavioural segmentation we run, shifts toward earning that named preference. Second, comparison pressure rises. Agents surface the true cheapest-equivalent option more often than tired humans ever did, which squeezes lazy pricing and rewards genuine differentiation: exclusive products, bundled expertise, service levels an agent can verify and report back.
The merchant readiness list
The work is concrete, and most of it is unglamorous. Get the product data complete and truthful: attributes, GTINs, variants, stock, prices that match the page. Make policies machine-readable: shipping costs and windows, return terms, warranty, in structured form rather than a PDF. Adopt the rails as they reach your platform, ACP and UCP support is arriving through the major e-commerce platforms rather than requiring bespoke builds. Instrument agent traffic separately, because an agent that visits and does not buy is a signal about your data, not your design. And put your own autonomy limits on the selling side: what your systems may confirm, discount, or promise to an agent without a human looking.
The Greek-market angle
For Greek merchants this shift is bigger than it looks, because agents flatten geography. When a German consumer's assistant sources olive oil, wine, or ceramics, it compares a Cretan producer's feed against a Munich importer's on equal terms, in the buyer's language, with none of the discoverability disadvantage a small foreign merchant usually carries. Greek e-commerce has always exported harder than it should because being found abroad was expensive. Agents make being found a data problem, and data problems are cheap to fix relative to marketing budgets. The producers and retailers who structure their catalogues now inherit demand that used to require a marketplace's permission.
The same lesson applies defensively at home. Greek consumers are on the same assistants, and the local loyalty that protected established shops erodes when the agent quietly checks three competitors per purchase. The moat moves to fulfilment quality and product truth, which is where it always should have been.
Where to start
Run one honest test this month: ask a shopping-capable assistant for the products you sell, in English and in Greek, and see whether you appear, what data it shows about you, and whether it can complete a purchase. The gap between what it found and what you know about your store is your work list. It is the same exercise we run for answer-engine visibility, applied to the transaction layer, and it takes an afternoon.
We build the machine-readable layer that makes merchants visible and buyable to agents: structured catalogues and pricing intelligence through AI Wine Intelligence, grounded site answers through the AI Website Assistant, and the predictive CRM that keeps the customer relationship yours when the agent sits in the middle. All of it grew up inside the largest Greek wine retailer, where the agentic shopper is already in the logs. If you want to know what agents currently see when they look at your store, get in touch at inbusiness.gr and we will show you.