Selling to Machines
E-commerce, retail and growth in the agentic era β visibility, protocols, operations and economics. A self-paced course on selling when the shopper may be a model or an agent: how generative engines choose what to cite, how to make a catalogue machine-legible, how to win the off-site citations you do not own, how to measure demand you cannot see, and what to do about agentic checkout. Thirteen modules, each producing an artefact.
Most organisations don't have an AI problem. They have a placement problem.
Roughly 95% of enterprise generative-AI pilots produce no measurable return, and Gartner expects more than 40% of agentic projects to be cancelled by the end of 2027. Read the post-mortems and the pattern is rarely that the model was incapable. It is that a model was given work a query, a rule or a scheduled script would have done more cheaply and more reliably β while the two steps that genuinely needed judgement were never identified, governed or measured.
This course teaches the discipline that prevents that: where AI belongs, how to connect it to systems you already own, how to govern it under GDPR, HIPAA and the EU AI Act, and how to redesign the organisation around what actually changed. It is written for people who will never write the code and are accountable for the outcome.
Frameworks, not features
Vendor landscapes date in months. A test for whether a step needs a model, and a control set that satisfies three regulatory regimes at once, do not.
Sourced, dated, falsifiable
Claims are cited and time-stamped β from Akerlof and Coase to the 2026 AI Act omnibus that moved the high-risk deadlines most course material still gets wrong.
Every module ships an artefact
A labelled portfolio, a no-list, an integration map, a control map, a scorecard, a layer audit β assembling into a board-ready twelve-month roadmap.
What you will be able to do
By the end of the course, participants will be able to:
- Explain how generative engines and shopping agents select, cite and recommend products
- Distinguish the four AI surfaces in commerce and locate where your demand actually sits
- Audit a storefront for machine-legibility and produce a prioritised remediation specification
- Specify a product feed and structured-data layer an engine and an agent can both consume
- Set and defend an agent-access policy, balancing visibility against crawl cost and control
- Win and correct the third-party citations that ground answers about you, and stay the right side of the law that governs them
- Measure AI-influenced demand when the referrer is missing, and report it credibly to a CFO
- Evaluate the agentic checkout protocols and decide what to implement, when, and what to defer
- Design trust, fraud, dispute and returns handling for agent-initiated orders
- Adapt merchandising, pricing and assortment for a buyer that compares exhaustively
- Plan paid media across AI surfaces, separating live placements from announced ones
- Assess where durable advantage sits in agentic commerce, and what commoditises
- Reshape the team and the agency relationship around the work that actually changed
- Differentiate evidenced findings from vendor claims in a market generating both at speed
- Produce a twelve-month plan with sequencing, owners, measurement and a rollback position
13 modules, about 22 hours
Each module states its objectives, works through the material, and ends in an exercise that produces something you keep. All 13 are published in full, with a single running case, cited sources and a capstone assessed against a public rubric.
- 01What Actually ChangedThree shifts in how demand reaches a retailer β and the much longer list of things that did not move at all.You produce: A one-page brief separating what has changed from what has not1.5h
- 02The Four AI SurfacesAI search, on-platform assistants, third-party agents and your own β four different games, four different owners, one budget.You produce: A surface map showing where your demand actually sits and who owns each surface1.5h
- 03LLM Visibility I β How Engines Choose What to CiteThe retrieval pipeline behind a generative answer, what the evidence says actually moves citation, and how to measure your position before you spend anything.You produce: A prompt set and a dated baseline of your citation position2h
- 04LLM Visibility II β Making the Catalogue LegibleFeeds, structured data, rendering and access policy β the four layers that decide whether a machine can read what you sell.You produce: A machine-legibility audit and a prioritised remediation specification2.5h
- 05LLM Visibility III β Winning the Citations You Do Not OwnMost of the sources grounding answers about you sit on somebody else's domain. This is the module about them.You produce: A citation source register and a 90-day off-site plan with named owners2h
- 06Measurement When the Referrer DisappearsCounting AI-influenced demand you cannot see in analytics, and reporting it in a form a finance director will accept.You produce: An AI-influenced demand model, with its stated confidence, that your finance team accepts2h
- 07Agentic Checkout and the Protocol StackACP, UCP and the payment rails beneath them β what exists, what it asks of you, and how to decide what to implement.You produce: A protocol decision with a dated implementation position and a review trigger2h
- 08Trust, Fraud, Disputes and ReturnsWhat breaks operationally when the buyer is software, and the policy that has to exist before the first agent order arrives.You produce: An agent-order policy covering authorisation, evidence, disputes and returns1.5h
- 09Merchandising for a Machine BuyerWhat changes when the shopper compares exhaustively, never sees a banner, and reads your specification instead of your copy.You produce: A revised attribute, pricing and assortment specification1.5h
- 10Paid Media on AI SurfacesWhich placements exist, which are being sold as though they do, and what happens to performance marketing when the click is not the point.You produce: A paid plan that separates live placements from roadmap items1.5h
- 11Advantage, Moats and the Operating ModelWhat commoditises, what compounds, and how the team and the agency relationship have to change.You produce: An advantage thesis and a team and agency design1.5h
- 12Appendix: The VocabularyEvery term this course uses, defined once β because in a market this noisy, a shared definition is a negotiating position.You produce: A vocabulary you can hold a vendor and an agency to0.5h
- 13Capstone: The Twelve-Month PlanOne assessed deliverable, assembled from the artefacts you built along the way.You produce: The completed plan, assessed against a published rubric2h
Vantis Diagnostics
A single running case, Marlowe & Finch, threads through every module: a fictional European home and lifestyle retailer of β¬182m revenue, with direct e-commerce, marketplaces, wholesale, eleven stores and an agency on retainer β chosen because it exercises every profile in the audience at once.
One company, several systems, and deliberately different answers: an internal tool that is minimal-risk, and a patient-facing one that engages special-category data, automated-decision rules and possibly medical-device regulation. Seeing why the same organisation lands in different tiers is the point.
The question a client's chief executive asked mid-walkthrough. The honest answer β most of it, yes β is where this course starts, and why its first test is a test for saying no.
Built for the people who fund, govern and deploy β not build
None beyond running or advising an e-commerce or retail business. No engineering background is required; the technical layers are covered at the level a leader needs to specify and challenge work.
E-commerce growth managers
Leave with a visibility audit, a feed and schema specification, and an attribution model that survives the loss of the referrer.
E-commerce and retail leaders
Leave with a protocol decision, an agent-order policy, and a board-ready position on where advantage now sits.
Omnichannel retail leaders
Leave with the assortment, pricing and inventory consequences of machine buyers, and how marketplace mix changes.
Marketing agencies
Leave with a repeatable client engagement: the audit, the reporting model, the paid-surface classification and the retainer they justify with it.
A timed exam, and a certificate worth having
25 questions in 40 minutes, drawn at random from a larger bank so every module is represented and no two papers are the same. 20 correct to pass. Pass and a signed certificate arrives by email.