Assess Your Agentic Commerce Readiness

Summary

Somewhere between 10 and 40 percent of a typical retail site's traffic is already agents, and most companies have no way to see it.

In a DataCamp webinar, Richie Cotton spoke with Jim Cook, founder of the AI consultancy Agentic Landmark, who spent 35 years leading digital shifts at GE, Sprint, and Credit One Bank before turning his attention to agentic commerce. Cook's argument is that this traffic is invisible by design: agents crawling a site to evaluate products don't click through or generate page views the way a browser session does, so analytics tools file the activity under "direct" or leave it unattributed. Meanwhile, agents that do complete a purchase convert at a rate Cook puts 42 percent higher than average, because they arrive with a narrow, pre-filtered set of criteria rather than browsing.

Cook walked through a framework for scoring how exposed a business is to this shift, built around five factors he splits into "weather" (traffic dependency, structured comparability, agent substitutability) and "roof" (brand strength and structured data maturity). He also covered the four technical standards agents use to read a site, why the EU's new digital product passport law puts Europe ahead of the US and UK on this, and which product categories face the most immediate risk. His closing point: agentic commerce readiness isn't a project with a launch date. It's a position a business either holds or doesn't, and the first step is getting the people who own the product catalog, the analytics stack, and the checkout flow into one room to ask where the exposure actually sits.

Key Takeaways

• Agent crawl traffic already makes up an estimated 10 to 40 percent of ecommerce site visits, but it shows up in analytics as direct or unassigned traffic because agents don't browse the way humans do.
• Roughly 66 percent of US retail product pages are now machine-readable, a threshold that flipped within the past year as sites raced to catch up with crawler demand.
• Agentic commerce isn't a new sales channel. It's a machine making the same choices a person used to make, which means pages designed to persuade a human may never be seen by the buyer at all.
• Consumers currently sit on a delegation ladder with four rungs: researching alone and deciding for themselves, letting an agent narrow options to a shortlist, giving an agent full authority within defined rules, and letting agents negotiate directly with other agents. Most people are still on the first rung.
• The EU's digital product passport law took effect on August 2, 2026, requiring unique product identifiers, material composition, origin, and conformity declarations in structured, machine-readable form. The UK has no equivalent law yet but does enforce actively; the US remains largely unregulated until after the fact.
• Exposure varies sharply by category. Electronics, apparel, footwear, beauty, healthcare staples, and travel face high exposure now; luxury goods, bespoke or highly configurable products, and relationship-driven B2B sales have more runway.
• Chat GPT users can already connect their bank accounts through a Plaid integration launched in April 2026, letting the assistant answer questions like "can I afford this trip?" using real account data.

Deep Dives

The traffic already hitting your site that your dashboard can't see

Cook opened with a number he wanted the audience to sit with: 10 to 40 percent of ecommerce traffic today is already agent-driven, and almost none of it registers correctly in standard analytics. "It's just that you can't see it," he said. The distinction he drew was between agent crawl volume, machines retrieving product data to evaluate it, and AI referral traffic, the much smaller slice, still under 1 percent of visits, where a chatbot sends a person to a specific page with a click attached. Crawlers don't click. They read a page, extract structured data if it exists, and move on, which means Google Analytics or Adobe Analytics typically buckets that activity as direct or unassigned traffic rather than flagging it as machine-originated.

He illustrated the scale with a ratio from his own client work: one human visit for every 198 OpenAI crawls hitting the same site. Using the fictional running-shoe brand Meridian as his working example throughout the session, Cook said the company was "basically blind" to this shift, because its analytics stack "haven't been able to architect... in a way that allows you to see" agent activity, and "these agents don't necessarily click, move through page views, and those kind of things." That blindness has a direct cost. Cook cited data showing agents convert 42 percent better than average once they do transact, a function of arriving with a pre-filtered, narrow set of purchase criteria rather than browsing casually. A business that can't see agent traffic in its funnel is, by definition, unable to measure whether that funnel is working for the channel driving its best conversion rate.

The practical implication, Cook argued, is that marketing budgets are already starting to shift toward channels that get credit for conversions agents are actually driving, and away from the invisible crawl activity that set those conversions up. "That gap is really the whole problem we're gonna talk about," he said, framing the rest of the session as a response to that visibility gap rather than a forecast of something still to come.

Why agentic commerce isn't a channel, it's a machine making the choice

Cook was direct about a mistake he sees brands making already: treating agentic commerce as a new marketing channel to optimize, the way search engine optimization or answer engine optimization gets treated. "It's definitely not an ecommerce channel," he said. "It's not a new channel. It's a machine." The distinction matters because it changes what "optimizing" even means. Where SEO and answer-engine optimization aim to get a page ranked highly enough that a human clicks it, agentic commerce means the buying decision itself is being made by software, often without a human ever loading the page a marketing team built to persuade them. "Think of agentic commerce as a machine doing the choosing that a person used to do," Cook said. "If you treat it as a channel, you're gonna optimize a page that the buyer potentially never sees."

To show how much authority consumers are willing to hand over, Cook laid out a four-rung ladder he calls posture. At the bottom sits the Curator, where a person uses an agent purely for research and still makes the final call themselves. This is where most people are today. Above that is the Scoper, where a consumer sets parameters and the agent returns a shortlist for the person to pick from. Higher still is what Cook calls delegated authority, illustrated by his Meridian shoe example: a shopper hands the agent a defined scope, including preferred brands and substitution rules, and the agent completes the purchase within those bounds, including redeeming loyalty points, but checks back before finalizing. At the top, still in single digits of adoption, sits full orchestration, where an agent runs the entire process end to end and can transact directly with a brand's own agent with no human step in between. Cook expects that top rung to remain rare for another 12 to 18 months.

Third-party research he cited, refreshed in June 2026, put rough numbers on the middle of that ladder: 74 percent of people already delegate routine shopping tasks like deal comparisons to some kind of agent, 32 percent let an agent decide within set limits, and 9 percent hand over full delegation. The categories overlap rather than sum to 100, but the direction is consistent with Cook's framework: authority is moving from the human to the agent gradually, one narrow task at a time, not in a single jump.

The four standards agents use to read your site, and why the EU just made one of them law

Cook grouped the technical requirements for agent visibility into what he called an accessibility stack, four layers deep. The most familiar is robots.txt, a passive file that permits or blocks crawler access. Newer and less widely adopted is an llms.txt file, which goes a step further by pointing agents toward the specific pages and data a site wants surfaced, paired with schema.org markup that renders product data in a structured, machine-parseable format. The fourth and, in Cook's view, fastest-moving layer is WebMCP, a set of callable tools that let an agent not just read a page but act on it and initiate a transaction. It remains a W3C draft standard, but Cook noted that "a lot of the major players are getting on board to support" it, including Google.

He drew a sharp line between what these layers reward and what older SEO-era thinking rewarded. "Being unforgettable was the last era's advantage," Cook said. "Being machine-readable is this era's table stakes, to be honest. Clean data will outrank a brand an agent's never heard of." Where a human reader responds to atmosphere, phrases like "premium build quality" or "stunning views", an agent needs the underlying attribute: the metal the boiler is made from, the floor level and window orientation of a hotel room, the documented eligibility criteria behind a stated interest rate. Cook wasn't arguing that brand-facing copy disappears. He was arguing that a second, parallel layer of structured data now has to exist underneath it, or the brand simply doesn't get considered.

The regulatory backdrop makes this less optional than it might sound. Two days before the session, on August 2, 2026, an EU law establishing a digital product passport took effect, requiring products sold in the bloc to carry structured, machine-readable data covering unique identifiers, materials and composition, country of origin, the responsible economic operator, and a declaration of conformity. Cook compared the pattern directly to GDPR: the EU moves first and treats the requirement as binding, the UK sits in the middle with active enforcement but no dedicated AI law yet, and the US remains largely deregulated, addressing problems after they occur rather than setting requirements in advance.

Scoring readiness: the five-factor index, and why two of the factors matter more than the other three

To move the conversation from "this matters" to "here's where we stand," Cook shared a scoring framework he built and calls the Agentic Readiness Index. It splits into two groups. Three factors, worth 65 percent of the total score, he calls "weather": conditions largely outside a company's control. Traffic dependency measures how exposed revenue already is to search- and retail-led discovery that agents can intercept before a human ever reaches the site. Structured comparability, worth 20 percent on its own, measures how cleanly an agent can rank a product against competitors based on available data. Agent substitutability measures how easily an agent could swap the product for a rival's without the consumer noticing a meaningful difference.

The remaining 35 percent Cook calls the "roof," the part a business actually builds and controls. Brand moat, worth the larger share of that 35 percent, measures how much preference and loyalty survives an agent's side-by-side comparison regardless of the numbers. Structured data maturity, worth 20 percent, measures how complete and machine-legible the underlying product data is when an agent shows up looking for it. Cook was blunt about the second one: "if the data is sitting all in images and PDFs, it's not gonna work," because that data "can't be easily accessed and read by an agent" for inclusion in a decision set.

Running Meridian, his example running-shoe brand, through the index produced a score of 65.9 out of 100, with 45.65 of that coming from the weather side, meaning the bulk of Meridian's exposure sits in conditions it can't directly change, while only 20 points of upside remain in the roof factors it can actually build. Cook's advice for any company running the same exercise: don't try to fix everything at once. "You don't have to boil the ocean," he said. "You just need to understand your position," then close the gaps that are actually within reach before defending that position going forward.

Which categories are exposed now, and who needs to be in the room

Cook was specific about where the risk concentrates today versus where it's still further out. Highest exposure right now sits with categories that are frequently replenished or easily compared: electronics, appliances, footwear, apparel, beauty, healthcare staples, travel and hospitality, and commodity B2B supplies. Longer runway exists for luxury and craft goods, experiential purchases, anything bespoke or highly configurable, and relationship-led B2B sales, categories where a strong brand moat or a human sales relationship still resists an agent's comparison logic. Asked about high-consideration purchases like buying a car, Cook said agents are already doing the research work, "these are all the top consumer rated minivans," but consumers aren't yet handing over the final purchase decision itself for anything that large or infrequent.

On where the responsibility for fixing this sits inside a company, Cook pushed back on the idea that it's a marketing or ecommerce problem alone. "It's not just about digital, the digital organization, the ecommerce organization, and the marketing organization," he said. Data engineering and merchandising need a seat at the table too, since the underlying fix is a data-structure problem before it's a marketing one. He pointed to product information management, or PIM, systems as the most immediate lever most companies already own but underuse: "The PIM companies should be your next best friend." Asked directly how a company gets its product data into shape, Cook's answer was simple: check whether the details actually sit in reachable, schema.org-tagged fields, or whether they're locked inside a downloadable PDF and a paragraph of marketing copy an agent can't parse. Roughly 66 percent of US ecommerce product pages fail that test today.

His closing homework assignment for the audience wasn't a budget request. It was a meeting: get the people who own the product catalog, the analytics stack, and the checkout experience into one room and answer three questions together. Are we in an exposed category? Can we even see agent traffic in our numbers today? And what's the widest gap we actually control? "I'm not saying you need a budget or a vendor right now," Cook said, "but the people who own the pieces need to get together in one room" before deciding what to build.