Loyalty When an Agent Shops: Keeping the Customer Relationship and Member Pricing

What does loyalty mean in agent shopping?

Loyalty in agent shopping means a customer keeps their points, member pricing and free-shipping entitlement even when the purchase starts on an AI surface. The classic loyalty programme rested on one assumption: the customer comes to your site, signs in, sees their points and uses them. In the agent era that chain can break. The user finds the product through ChatGPT or Google AI Mode, the agent prepares the basket and moves to payment; if the customer never signed in on your site, neither member pricing nor accumulated points come into play. The UCP update Google published on 19 March 2026 targets exactly this: the Identity Linking capability lets an agent act on the user's behalf in the brand's systems through an OAuth 2.0 authorisation flow, and once linked the agent can show earned rewards, apply them to the cart and explain eligibility.

This article is the May 2020 loyalty programme guide rewritten for the agent era. Segmentation and the frequency model still work; what changed is where the customer relationship is formed. Our thesis at Webtures: in the agent era loyalty is not a campaign mechanic but a question of identity and data; a brand that loses the customer relationship becomes one line in the agent's catalogue.

What is the disintermediation risk?

The biggest structural risk for a brand in agentic commerce is disintermediation: the agent layer moves between brand and customer. The user finds the product on the agent's recommendation, the agent runs the comparison, and first contact is often with the assistant rather than the brand. The consequences show up in three places:

• Loss of data. Which user looked at which product, what they compared it with, why they walked away; most of that stays with the agent. Site analytics often shows only a "Direct" visit.
• Loss of relationship. When the order is tied to the assistant's account (as in Amazon's Buy for Me flow), marketing permission, email address and preference data do not accumulate with the brand.
• Margin pressure. The agent selects on price, delivery and ease of returns. Without a rules engine that turns into trying to be cheapest on every query and eroding margin.

The antidote is making loyalty data readable by the agent. If the agent cannot see member pricing it evaluates you at list price; if it can, it may put you ahead for the same product. Loyalty is no longer only a retention tool; it becomes a ranking signal.

How does loyalty data work on the agent layer?

The table also shows why being the merchant of record matters. After the first wave of in-chat checkout was pulled back, the dominant model became "discover in AI, buy on site", and that model lets the brand keep the customer relationship. If the transaction completes on your site, sign-in, membership and permission flows stay with you. An agent that prepares the basket and hands it to your site is a better scenario for loyalty than in-chat completion.

How do you build a loyalty programme for the agent era?

1. Keep segmentation, widen the signals. The classic model built on frequency, basket value and campaign behaviour still works; add an agent-origin tag to the order. Which customer arrived via an agent, and did they move to the direct channel after the first order?
2. Turn member pricing into data. A membership discount should be a value an API can return, not only a rule computed in the basket. An agent cannot use a discount it cannot see.
3. Prepare for Identity Linking. An endpoint supporting the OAuth 2.0 authorisation flow, scope definitions (read balance, apply reward) and a revocation path. On platforms such as Shopify this support comes from the platform; on custom stacks it needs development.
4. Write reward rules explicitly. The agent has to explain eligibility: which products it applies to, until when, which combinations it excludes. An ambiguous rule is not applied by the agent.
5. Tie margin protection to the rules engine. Track shipping margin erosion and return rates on agent orders; below defined thresholds, no discount applies.
6. Move permission collection after the order. Email permission is not requested on the agent surface; after delivery, ask for permission and an account through your own channel (an in-parcel code, the order tracking page, the returns flow).
7. Measure. Repeat purchase rate of agent-sourced orders, the rate of movement to the direct channel, and the margin difference between members and non-members. Those three numbers say whether loyalty works in the agent era.

What are the most common mistakes?

• Keeping member pricing behind sign-in only. The agent sees the list price and you fall behind in the comparison.
• Showing points only on the account page. If the agent cannot query the balance the customer cannot use their points and the experience breaks.
• Not tagging agent orders. A brand that does not know which customer arrived via an agent runs retention blind.
• Rule tangles. Overlapping campaign rules cannot be explained by the agent, and a discount that cannot be explained is not applied.
• Thinking loyalty means discounts. In the agent era priority delivery, free returns and stock priority are stronger signals than a discount, and both cost less margin.

What changes for cross-border brands?

In multi-market selling loyalty data is defined per market: in which country the membership discount applies, in which currency the free-shipping threshold sits, within which window returns are accepted. If these three are not given per country the agent treats them all as unknown. In our own export matrix, loyalty is formed directly with the consumer in B2C markets such as the United States and the United Kingdom; in B2B-heavy markets such as Germany its equivalent is a contracted price list and an API opened to the supplier's agent. In the Gulf competition is low and membership data is almost entirely unstructured; a brand that moves early can be the first name an agent evaluates at member pricing.

Read the agent-side counterpart of loyalty together with the agent makes the recommendation now and return and trust signals; we covered UCP's cart, catalog and identity capabilities in the protocol map. You can see whether your loyalty data is readable by agents with the Agentic Commerce Readiness scan and plan the identity and authorisation flow with our Agent Experience team.