The policy shelf

A customer asks her agent to book a Tuesday haircut for her dog. The agent is told to ignore any groomer that uses cage drying, only book if there is parking within five minutes, and prefer places that refund if the groomer runs more than 20 minutes late.

The agent does not care that one salon has a charming photo of a spaniel in a bow tie. It does not read the founder story. It does not notice the pastel homepage. It goes straight for the cancellation terms, drying policy, lateness clause, parking note and refund conditions.

One business looks better to a human. The other is easier for a machine to buy from.

That is the shift most commerce teams are missing. AI agents will not shop like distracted humans with a bit more memory. They will shop by constraint. They will ask whether a delivery can avoid Friday, whether an invoice can carry a purchase order, whether a child under eight can attend, whether oat milk costs extra, whether a technician can enter without the owner present, whether a gift card stacks with a member discount.

The old internet made products searchable. Agent commerce makes conditions searchable.

Small print is moving from the basement to the shelf.

The boring part becomes the product

A Shopify product has a SKU, price, image, description, variant, stock count and fulfilment setting. That is enough when the buyer is a person who can tolerate ambiguity, ring the shop, read between the lines, or give up.

An AI buyer needs a different object. Beside the SKU sits a ruleset:

• eligibility
• cancellation windows
• refund conditions
• substitution rules
• guarantees
• proof requirements
• approval paths
• expiry dates
• human handoff triggers

This is the policy shelf: the structured, maintained layer of promises that agents can inspect before they buy, book, return, cancel or substitute.

It is not a prettier FAQ. It is not legal copy with headings. It is not a chatbot trained on the help centre. Those are human surfaces. The policy shelf is operational inventory.

A return window becomes a field. A rain reschedule becomes a condition. “No cage drying” becomes an exclusion. “Net 30 available for verified business domains” becomes an eligibility rule. “Refunds over £100 require approval” becomes a handoff path.

Some of this exists in fragments already. Schema.org has MerchantReturnPolicy. Google’s booking systems have structured rules for reservations. Enterprise procurement portals ask suppliers for insurance documents, data retention policies, accessibility answers and invoice terms before anyone discusses the actual service.

The difference is distribution. These fields used to be compliance plumbing. Agents turn them into sales surfaces.

Every business becomes a little airline

Airlines learned this disease early.

A travel agent never really sees “Paris for £300”. Behind the fare sits a thicket of rules: advance purchase, minimum stay, change fee, refundability, stopovers, baggage, seat selection, same-day changes, upgrade eligibility and the delightful nonsense of married segments. A cheap fare is not the same product as a flexible fare, even when the seat is physically identical.

Airline distribution systems made boring rules sellable. IATA’s New Distribution Capability is partly about packaging and distributing those richer offers. ATPCO’s Routehappy exists because passengers and sellers need to know what is actually included.

Ordinary businesses are about to inherit the same logic, minus the glamour, plus an angry dog groomer at 8am.

The haircut is the same haircut. The promise envelope changes:

• Basic booking: cheapest slot, strict cancellation, no lateness refund.
• Semi-flex: one reschedule, partial refund, named groomer if available.
• Fully-flex: same-day cancellation, no cage drying guarantee, parking instructions, refund if late.

A café can sell a breakfast meeting with three different policy envelopes: non-refundable table, weather-safe terrace booking, corporate invoice bundle. A tutor can sell the same lesson with different reschedule rights. A plumber can charge more for “arrival before school run” and “tenant can grant access”. A childcare provider can make eligibility machine-readable without asking a receptionist to answer the same question 40 times a week.

Refundability is not customer service. It is inventory.

Agents shop by constraint

Human ecommerce has been optimised around desire. Search query, product image, review count, basket, checkout. The policies usually appear after the customer wants the thing.

Agents invert the order. They often need the rules before the desire.

“Find me the cheapest hotel near the station” becomes:

• allows 11pm arrival if my train is delayed
• free cancellation until the train departs
• can split the invoice between personal card and company card
• quiet room not facing the road
• no resort fee surprises
• accepts late checkout request before booking
• stores accessibility confirmation

The prettier hotel loses to the hotel with structured late-arrival rules. The independent clinic loses to the boring chain because the chain exposes insurance acceptance, cancellation terms and chaperone policy in a way the agent can verify. The caterer with better food loses the corporate lunch because it cannot express allergen substitutions and invoice terms without a phone call.

That will feel unfair to owners who think their service quality should carry them. It often will. Distribution has never been fair. It rewards what the channel can understand.

A policy is only cheap while nobody can compare it.

The shelf needs an owner

Most small businesses already have these rules. They live in heads.

The hotel manager knows which regulars get late checkout. The owner knows when to waive a no-show fee. The senior dispatcher knows which estates need two-person delivery. The receptionist knows the consultant will accept one extra patient if the family has travelled far. The café owner knows which regular can run a tab.

Agents force private judgement into a public interface. That is useful, ugly and politically awkward.

A real policy shelf product would look less like a website builder and more like a policy CMS. The merchant writes policy atoms:

• return_window: 14 days, unopened goods only, exceptions for faulty items
• rain_reschedule: once, forecast above 50%, requested before 6pm previous day
• dog_cage_drying: never, unless owner approves in writing
• invoice_terms: net 30 for verified business domains only
• handoff_required: refund over £100, allergy conflict, safeguarding concern, medical claim

The endpoint exposes a subset to buyer agents. The dashboard shows attempted purchases blocked by missing conditions. A weekly drift report says: agents asked 83 times about pram access; no structured answer exists. The owner can turn it into a field, keep it vague, or require human approval.

The interface details matter:

• “Agent visible” toggle beside each rule.
• Draft, publish to humans, publish to agents, require approval.
• Confidence score beside each policy.
• “Last confirmed by Emma, 12 days ago.”
• Conflict warning: “Late arrival allowed” clashes with “doors locked after 8pm.”
• Test console: run a pretend buyer agent with constraints and see whether the booking passes.

This is not glamorous work. Someone will own policy inventory. In many firms it will sit with operations, not marketing. Monday morning becomes: review agent rejections, approve new rule atoms, retire seasonal exceptions, simulate top buyer agents.

“Policy merchandiser” is a horrible job title. It is also probably real.

Flexibility becomes priced

The pricing angle is stranger than the automation angle.

When agents can compare policies, flexible terms stop being a mushy service promise and become paid components. Not “premium support” as a vague bundle. Granular rights:

• £8 extra for rain-safe reschedule.
• £15 for no-questions cancellation until the morning of service.
• £4 for guaranteed oat milk substitution without surcharge.
• Business tier with purchase order invoicing, SLA, audit log and named escalation contact.
• Family tier with two illness reschedules per term.

Cheap tiers may become deliberately less agent-friendly. “Phone us” is friction, but friction can protect margin. If every waiver is structured, agents will find the cheapest path through the rule system. They will book four refundable slots and cancel three. They will choose the weather clause most likely to trigger. They will split orders to stay below approval thresholds.

Airlines know this well. Rules do not merely describe the product. They defend revenue.

Ordinary businesses will learn fare-family thinking. Basic, semi-flex, fully-flex. Same service, different promise envelope. Customers will hate some of it. Businesses will pretend it is about choice. Both sides will be partly right.

The receipt becomes the trust object

A customer currently trusts a brand page, a confirmation email, a screenshot, a staff member’s promise, or nothing at all.

An agent-mediated purchase creates a different trust moment:

“Booked because cancellation rule was verified at 10:42. Merchant endpoint signed. Screenshot stored. Refund condition: groomer delay over 20 minutes.”

That tiny receipt changes the support conversation. The customer is no longer saying, “I thought your website said...” They arrive with a machine log.

“My agent selected this option because your rule said stroller access was true.”

“My agent booked the repair because your policy allowed tenant access.”

“My agent accepted the deposit because your cancellation field said one free reschedule.”

Support loses the comfortable escape route of “that is not what we meant”. The blame shifts to the rule publisher.

This will cut some low-value support. Fewer people will ask whether they can cancel, substitute, reschedule or bring a child. It will also create more expensive disputes. Frontline staff become policy interpreters. Their discretion shrinks in the common case and becomes more valuable in the exception case.

Status changes too. The old power of a good receptionist, dispatcher or store manager was knowing the unofficial rulebook. In agent commerce, that rulebook gets partly compiled. Humans move from discretion kings to exception approvers.

Some will hate that. They will be right to.

The dark shelf is necessary

The bad version of this future is transparency maximalism: every rule exposed, every exception encoded, every human judgement flattened into fields.

That would be commercially naive.

Some policies should be agent-actionable. Stable promises belong on the shelf: return windows, access requirements, invoice eligibility, published guarantees, safety exclusions.

Some should be askable but not automatic. These are soft promises: late checkout, goodwill refunds, unusual substitutions, special access, exceptions for regular customers. The agent can request them. A human approves.

Some should remain unavailable. Call it the dark shelf if you like, though it sounds more sinister than it is. Businesses need private discretion. They need fraud controls. They need room for generosity that does not become entitlement. They need to avoid turning “we’ll try our best” into a contract surface.

Ambiguity has economic value. It protects margin, relationships and common sense.

The product decision is not “make every policy readable”. The product decision is which promises are stable enough to publish to machines, which require human approval, and which should never become an API field.

Compliance becomes distribution

Corporate buyers will move first because agents already need rules to buy safely.

Travel, catering, training, clinics, maintenance, childcare, healthcare-adjacent services, rentals and events all have high-friction purchasing. The buyer cares about cancellation, insurance, safeguarding, accessibility, invoicing, data retention, substitutions and audit trails. A supplier with an empty policy shelf becomes harder to buy from, even if it is cheaper.

That is how compliance becomes a distribution tax.

Marketplaces will use this as leverage. Amazon, Booking.com, Deliveroo, DoorDash, Shopify, Google and vertical booking platforms can force merchants into their schema. Independent businesses will need lightweight policy shelves or they will vanish from agent shopping. SEO will gain an uglier cousin: policy search optimisation. Someone will sell “refund clause optimisation for buyer agents” by lunchtime.

Most consumers will still click the cheapest thing. Agents will misunderstand missing rules. Merchants will refuse endpoints. Too much structure will kill some charm, local discretion and goodwill. The first wave will be messy.

But the direction is hard to dodge. Commerce used to hide policies until after desire. AI buyers read policies before desire.

The next competitive advantage for many ordinary businesses will not be a better chatbot or a flashier booking page. It will be maintained small print, priced flexibility and a clean answer when a machine asks, “Can my human safely buy this?”

---