Executive Summary
AI visibility is not AI readiness. An AI buyer’s journey stops at your weakest link, and an average score hides which one.
Assistants now shape how buyers research, and open protocols from the largest companies in commerce are carrying them into the transaction itself. For a machine acting on a buyer’s behalf, readiness is a chain: being found, being understood, being usable, and being transactable. This article walks through the four places the chain breaks, the commercial questions leadership teams must answer, and the one reading rule that makes the framework useful.
Key Takeaways
- The buyer’s interface is changing: Assistants mediate research today, and protocols from OpenAI, Stripe, Google, and Shopify are extending them into checkout, unevenly but unmistakably.
- Visibility is not readiness: Visibility is one link of four. An AI customer must find you, understand what you sell, work with your platform, and complete the purchase. We assess all four as DUCT.
- The weakest pillar sets your score: Read the lowest pillar, not the average. A business scoring well on three pillars and 1.5 on the fourth behaves operationally like a 1.5, because the journey stops at the break.
- The failures are silent: Most failures never show on the page. Gated specifications, unstructured compatibility, stock locked in the ERP, and checkouts that assume a person fail silently, and only machines see them.
- DUCT is a diagnostic, not a maturity ladder: The order describes the buyer’s journey. Your implementation priority comes from where your chain actually breaks and what the break costs.
For 25 years, businesses have optimized commerce platforms for one kind of customer: a human, on a screen, reading pages. The next customer, increasingly, is a machine reading on a human’s behalf. A buyer asks an assistant which product fits, which supplier to trust, and what the trade-offs are, then acts on the answer without ever seeing the pages you built. Behind that sits something stranger: software authorized to research, compare, and buy.
The largest companies in commerce are laying the rails for that second customer now. OpenAI and Stripe published the Agentic Commerce Protocol as an open standard in September 2025, and Google and Shopify answered with the Universal Commerce Protocol in January 2026. Adoption is still early and uneven. Mainstream autonomous checkout has not arrived, and nobody should panic-buy a chatbot. But both protocols carry an AI buyer past the product page and into the transaction itself, and that changes what readiness means.
It also changes who notices your platform’s weaknesses. Human buyers have learned to work around them: they download the datasheet, phone the trade counter, and log in to check the price. An AI agent does none of that. It exposes every gap your customers quietly absorb, and it never tells you which gap sent it to a competitor.
AI Visibility Is Only One Link in a Four-Link Chain
A machine acting for a buyer has to complete a sequence: find you, understand what you sell, get answers from your platform, and finish the job. That sequence is a chain, and chains have a property that averages hide. It breaks at its weakest link. A business can excel at three of the four links and still lose the sale at the fourth.
Almost all the market attention sits on the first link. Answer engine optimization (AEO), the work of getting cited by ChatGPT, Perplexity, and Google’s AI results, is a real discipline, and we practice it. But a citation only starts the journey. The failures that cost revenue happen further down, and they hit hardest where buying is complex: catalog-heavy, B2B, and compatibility-driven commerce, where the deciding information has always lived in datasheets, dealer portals, ERP systems, and the memory of experienced staff.
Why We Built DUCT to Test the Whole Chain
Most AI-readiness assessments stop at the first link and ask whether AI can see and cite you. We built DUCT to follow the whole journey. The name stands for:
- DDiscoverable
- UUnderstandable
- CCapable
- TTransactable
These four pillars trace the path an AI buyer takes, from discovery to transaction. Three things separate it from a visibility audit:
- Full journey: DUCT tests every link an AI buyer must pass, not only whether assistants mention you.
- Weakest-link reading: DUCT reads your result by the lowest pillar, not the average, because the journey stops at the break.
- Commercial policy: DUCT asks the one question a technical audit skips: whether, and on what terms, your business should sell through AI assistants at all.
The result answers a leadership question rather than a technical one: where should your next AI commerce investment go first? Four failure patterns, one per pillar, show the framework at work. They are illustrative composites, not named clients, assembled from the failures that show up when teams actually test readiness. The first pattern starts where most companies feel safest: their own website.
Discoverable: Your Website Is Visible, but Your Product Knowledge Is Not
Imagine a manufacturer with 10,000 product pages, all indexed, all fast, all ranking respectably. To an AI buyer, it barely exists, because the information that would let anyone choose its products sits somewhere else: load ratings and tolerances in PDF datasheets, compatibility tables in a dealer portal, and trade terms behind a registration form. The pages a machine can reach are brochureware. The substance sits where no assistant will fill in a form to follow it.
This is a Discoverable failure, and it is more common than a blocked crawler. The commercial cost is direct: if an assistant cannot reach the facts a buyer decides on, it shortlists someone else, and you never learn you were in the running.
Discoverability therefore asks more than whether AI can technically reach your site. It asks whether the information a buyer needs is open, machine-readable, and accurate wherever it appears, including places beyond your own site where assistants also read about you. It also raises an ownership question most companies cannot answer: if an assistant started describing you wrongly tomorrow, who would notice, and what could they change without going outside the business?
Getting found, however, only brings the machine to the product. The next link decides whether it can do anything with what it finds.
Understandable: AI Cannot Recommend What Your Systems Cannot Explain
The model found the product. It still cannot tell whether SKU A fits a 2021 Golf, whether product B is the premium alternative to product A or an unrelated line, or that product C is mandatory with product D and pointless without it. The attributes are half-filled, the taxonomy has shifted twice in 10 years, and the compatibility knowledge that actually closes sales lives in the heads of a trade counter team.
A human customer never sees this problem, because a human phones the counter. An assistant has no counter to call. It infers from incomplete data, gives a weaker answer, or recommends the supplier it can reason about more confidently, which is not necessarily the better one. Your best product can lose to a competitor’s better-described one.
What the human sees
Brake pad set, front axle
$64.90
Add to cartIn stock near you. Fits most popular models. Unsure? Call our trade counter.
What the agent reads
Why Schema Markup Alone Cannot Close the Gap
The Understandable pillar asks whether your product data, relationships, and content are structured well enough for a machine to reason over, not merely display. The answer runs deeper than markup. Schema can describe a product. It cannot compensate for a commerce model that never captured the relationships a buyer needs: compatibility, substitutes, bundles, variants, constraints, and which product is the premium version of which. If your systems express those relationships nowhere, no amount of structured data on the page can invent them.
Our research keeps finding that exact gap. Structured data markup is now widespread, while complete machine-readable product records remain rare. In catalog-heavy sectors, poor product data taxes the whole operation, not just the website.
Suppose your data can explain every product. The agent’s next questions go to systems your website has never had to open up.
Capable: Systems Built for Human Logins Leave Agents at the Website’s Edge
Picture a B2B distributor with a clean, well-structured catalog. The buyer’s assistant asks the two questions every trade purchase turns on: is it in stock, and what is my price? Live stock exists only inside the ERP. Trade pricing requires a logged-in session for someone with a browser and a password. The agent has reached the edge of what the website knows, and the website is the only part of the business it can talk to.
Capable is the pillar this breaks on, and it exposes a wider truth about integration architecture: most commerce systems were built to serve people, not to answer software. Stock, customer-specific pricing, promotions, availability, and order state live in systems that nobody ever asked to answer an outside question. When those systems stay silent, the agent cannot quote, and a buyer without a price moves on.
How to Test Whether Your Platform Can Answer an Agent
The pillar comes down to one idea: whether your business can expose its operational data, such as stock, pricing, and order status, to software, rather than just its content to crawlers. Three practical tests sit under that idea:
- Access: Can an agent retrieve products, stock, pricing, and promotions through a documented interface?
- Answers: Does a question asked in a customer’s own words get an answer rather than a filter tree?
- Speed of change: How long does a fix take to ship? A platform where every fix becomes next quarter’s backlog item cannot keep pace with demand that reroutes in weeks.
Pass all three, and the agent reaches the final step. That is where the quietest failure waits.
Transactable: Reaching Checkout Is Easy, Selling Through AI Is Not
The last failure is the quietest, because everything works until the final step. The agent has found the product, understood it, priced it, and placed it in the basket. Then checkout begins: bot protection challenges the automation, payment controls assume a cardholder at a keyboard, and a commercial policy written years ago effectively requires that the buyer be a person holding a browser. The agent cannot fall back on the human behaviors your current journey assumes. It may retry, ask the buyer for help, or move on to a supplier that works, and nobody phones you to say so.
Removing Checkout Barriers Answers Only the Easy Question
Removing those barriers is the visible half of Transactable. Whether software can click buy is the easy question. The harder one is whether your systems can trust the software that clicks.
An agent buying for someone spends that person’s money within limits they set. The industry calls this delegated authority, and the payment protocols are being designed around it: mandates, scoped credentials, and accountability for every step. In practice, your systems need to distinguish an authorized agent from an unknown bot, understand whose authority it acts under, apply the right commercial rules, and leave an auditable record of what happened. They also need to know what can be reversed, by whom, and how to handle partial failure when one step succeeds, and the next does not. A checkout that merely tolerates bots answers the easy question and leaves the hard one open.
Are You Commercially Ready to Sell Through an AI Intermediary?
Transactable also covers everything after the purchase: order status, amendments, returns, and re-ordering, the things that today end in a phone call. And it includes the one question in the framework that isn’t technical at all. Before agents buy from you at scale, your leadership team needs answers to these:
- Customer ownership: Who owns the customer relationship when an assistant sits between you and the buyer?
- Margin: What margin or fee will you accept to sell through an AI intermediary?
- Authorization: Who can authorize an agent to buy on a trade account, and up to what limit?
- Commercial rules: Which pricing tiers, contract terms, and credit limits apply to an agent’s order?
- Returns: What happens when an agent-placed order is wrong, partly fulfilled, or sent back?
- Decision owner: Who in your business owns the decision to sell through AI assistants at all?
In most companies, nobody has been asked to form a position. Trace three of these journeys side by side, and the problem’s shape appears.
broken at U
Assistant reaches the product page → cannot establish fitment for the buyer's vehicle → journey ends. Discoverable held. Understandable broke.
broken at C
Assistant understands the product → cannot retrieve live stock or the buyer's trade price → journey ends. Two pillars held. Capable broke.
broken at T
Assistant selects the right product → checkout blocks automated interaction → journey ends. Three pillars held. Transactable broke.
Your AI Readiness Is Your Weakest Pillar, Not Your Average Score
Score the four pillars out of five, and the arithmetic will try to comfort you. Picture a scorecard: Discoverable 4.5, Understandable 4.2, Capable 3.8, Transactable 1.5. The average says a respectable 3.5. Operationally, the business behaves much closer to 1.5, because the journey can progress only as far as its weakest link. So we give our own team one instruction for reading any scorecard, and it is worth stealing.
Read the lowest pillar, not the average. The average describes overall maturity. The weakest pillar determines where the journey can fail.
Discoverable
4.5
Understandable
4.2
Capable
3.8
Transactable
1.5
arithmetic reading
Average 3.5. Comfortable, and irrelevant to a journey that has to pass through every pillar.
operational reading
1.5. An AI customer can find, understand and interrogate this business, and cannot reliably buy from it.
The most expensive mistake is spending at the wrong end of the chain. Buying visibility while your product data cannot support a recommendation sends better-qualified machine traffic into a failure that now happens faster. That raises a natural question: should you then fix the pillars in order?
DUCT Is a Diagnostic, Not a Maturity Ladder
Not necessarily, and this misreading is worth closing off. A company does not graduate from Discoverable to Understandable to Capable to Transactable. These are four simultaneous properties of the same commerce system, and the order describes the buyer’s journey rather than an implementation roadmap.
Your implementation priority comes from where the chain actually breaks and what the break costs. Often that is the earliest broken link, because everything downstream inherits whatever the upstream links pass along. Sometimes it is not. If your position on agent selling is strategic and entirely unformed while your discoverability has a minor weakness, the fourth pillar deserves attention first. Fix the constraint that prevents the next meaningful customer outcome. DUCT diagnoses; it does not prescribe a sequence. Finding that constraint starts with four questions.
Where Does Your AI Buying Journey Break? Run the Four-Question Test
You can take a first reading in one meeting. Put these four questions to your leadership team, one per pillar:
| Pillar | Ask your team | What a “no” costs you |
|---|---|---|
| Discoverable | Can AI find us, including the information a buyer actually decides on? | An assistant shortlists someone else before a person ever sees you. |
| Understandable | Can AI understand what we sell, what fits, and what goes with what? | AI recommends the supplier it can reason about, not the better one. |
| Capable | Can AI get stock, price, and answers out of our platform without a human session? | The agent stops at the edge of your website and cannot quote. |
| Transactable | Could an authorized agent complete a purchase with us, end to end, tonight? | Orders fail at checkout, and nobody reports the loss. |
Treat “probably” as a no. If you haven’t tested the journey, you don’t know whether it works. When we run this pillar with clients, “it would probably work” is the most common starting answer on agent checkout. If any of your four answers is “not reliably,” you have found your weakest link. Your AI commerce roadmap starts there, whether or not it is the fashionable pillar.
FAQs
DUCT is Net Solutions’ framework for assessing whether a digital business can be discovered, understood, used, and transacted with by AI systems. It has four pillars: Discoverable, Understandable, Capable, and Transactable, with 12 capability areas underneath, three per pillar.
No. Agentic commerce is the hardest test DUCT applies, not its definition. The framework covers ownership of AI visibility, product information, content, search, integration, architecture, post-purchase operations, and commercial policy, and most of that work also pays off with human buyers who arrive through AI.
The order describes the buyer’s journey, not an implementation sequence. Start from the weakest pillar, as the evidence shows it, weighed against what the break is costing. Often that is the earliest broken link. Not always.
Score the 12 capability areas from 1 to 5, for today and for where your growth plans need you to be, then read the lowest pillar rather than the average. The free AI Readiness Score gives you a structured version in about 15 minutes, and the AI Readiness Diagnostic adds external evidence and our assessment alongside your own.
Find Your Weakest Link Before a Buyer’s Agent Does
The free AI Readiness Score is a 15-minute self-assessment that shows which of the four DUCT pillars breaks your AI buying journey first. Where it flags a weak pillar, the AI Readiness Diagnostic adds the evidence: what your platform exposes, what the assistants say about you, and where your view and the facts disagree.
1. Take the Score
2. Identify your weakest pillar
3. Add the Diagnostic’s evidence
4. Prioritize the fix
