When an AI assistant shops for a customer, a loyalty programme works only if the merchant can recognise the member, calculate the eligible benefit, show the final price before approval and reconcile the reward after purchase. In 2027, retailers can test that with a single order from search to refund, comparing what the customer gets on the brand’s own checkout with what happens through the assistant.
- Run the same basket through direct and AI-assisted checkout.
- Test member pricing and points redemption separately.
- Reconcile the order and the points balance after a return.
One basket and two prices
“Order my usual skincare bundle. Use my member discount and apply points if that brings the total under ₹2,700.”
On the brand’s website, that instruction is simple. The customer logs in. The ₹3,000 basket drops to ₹2,700 with the member discount. The customer applies ₹100 worth of points and pays ₹2,600. If the programme awards one point per ₹100 paid, they earn 26 points. These are illustrative rules for a fictional Indian retailer.
Now give the basket to an AI shopping assistant. If it sees only the public price, it might move on. If it assumes discounts and points can be combined, it might quote a price the merchant cannot honour. If the order misses the loyalty ledger, the customer still has to chase the 26 points.
That last failure is easy to miss. The commerce team sees a successful payment. The customer sees a promise the brand failed to keep.
What changed heading into 2027
Agentic commerce is shopping in which an AI system can take steps on a customer’s behalf, from comparing products to preparing or placing an order, depending on the integration and the customer’s permission.
The Universal Commerce Protocol (UCP) now documents a loyalty extension for showing membership, tiers, eligible benefits and rewards that can be earned. Its specification treats redeemable balances separately. In practice, recognising a member and spending their points are two capabilities to test independently. Universal Commerce Protocol loyalty specification
Mastercard expanded Agent Pay in September 2026 with services for AI-initiated transactions. Visa’s November shopping benchmark reported a 42.8% failure rate across its test runs. Retailers can run a separate test for their own checkout rules. Mastercard’s September announcement and Visa’s November benchmark.
A five-step test for your 2027 pilot
Use one customer account, one basket and one benefit. Run the journey first on your site, then through the assistant or platform you plan to support. Keep the applicable offer rules and stock conditions the same. Compare the outcome at each step.
| Step | What to test | A pass looks like |
|---|---|---|
| Recognise | Try a signed-in member, a guest and a customer who declines account linking. | The merchant verifies membership with the customer’s permission. The guest sees the public terms. |
| Quote | Try an eligible basket and one with an excluded item or expired offer. | The merchant applies current rules and returns an itemised total the customer can understand. |
| Approve | Try combining a member discount and points. Change the basket before approving it. | The merchant checks whether both benefits can apply. The customer sees and approves any changed total. |
| Post | Repeat the order confirmation or send it twice. | Points are earned or debited once, with an order reference that support can look up. |
| Reverse | Cancel the order, then test a separate partial return. | The refund and points adjustments follow the programme rules for the items returned. |
Try a third run in which the assistant can recognise the member price but cannot redeem points. The customer should get a clear choice to continue at the confirmed price or move to the brand’s checkout with the basket intact where supported. This is especially useful for benefits that depend on a participating store, a specific payment method or proof of purchase after checkout.
What should teams measure?
Benefit accuracy is the percentage of checked baskets where the assistant’s final quote agrees with the merchant’s eligible total, including exclusions. Break it down by offer type. An average across all offers can hide a problem that affects an entire member tier.
Reward reconciliation is the percentage of eligible orders and returns where the loyalty ledger agrees with the order record. Check full and partial returns separately. An order can pass on price and fail on what the customer gets back.
Customer effort shows up in member sign-ins that need a retry, failed handoffs and contacts about missing points. Review the cause of each failure in the pilot. A clear explanation that an offer expired is a different outcome from a support ticket asking why it disappeared.
Set pass criteria before the pilot. An unauthorised points debit or repeated award should block rollout for that scenario. Once the checkout is reliable, use a comparison group to learn whether making the benefit easier to use improves conversion or repeat purchase after reward and support costs.
Three decisions for 2027
Which benefits should travel first? Start with the benefit customers already look for at checkout and that the merchant can calculate consistently. Member pricing or a simple earn rule may be easier to test than a promotion that varies by store, SKU and payment method. Test redemption as its own step.
Where does the merchant take over? An AI assistant may help a shopper compare offers. Eligibility, the final price and the reward ledger should come from the merchant or its authorised programme partner. Agree what can be shown, what needs approval and where the customer goes if the platform cannot complete a step.
What happens after checkout? In India, a purchase-linked campaign may verify eligibility through an order record, an invoice or a QR claim. The confirmation should tell the customer whether the benefit has been applied, is pending or requires a claim. Give support a way to see the same status. This is where programme design, purchase verification and fulfilment meet.
The quickest way to find the weak spot is to ask commerce, loyalty and customer service to look at the same order. If they each tell a different story about the final price or the points balance, fix that journey before giving another channel access to it.
If you are planning a 2027 rollout, talk to RewardPort about mapping one live benefit from eligibility through purchase verification, fulfilment and returns.
Frequently Asked Questions
Can AI shopping assistants apply loyalty points?
They can when the merchant and the shopping platform support points redemption, verify the member and collect the necessary approval. An assistant may be able to show member status or an eligible discount even when it cannot spend points.
Do retailers need to rebuild their loyalty programme?
A pilot can start with an existing programme. Give the supported channel access to the merchant’s current eligibility rules and test whether checkout, rewards and returns agree. The results will show where to invest first.
What is the difference between a member price and points redemption?
A member price changes what an eligible customer is charged. Points redemption uses value that the customer has already earned. The merchant needs to show whether both can apply to the same order and what happens to each if an item is returned.
Who should settle a loyalty dispute on an AI-assisted order?
Assign an owner before launch. The team handling the dispute needs the order, the benefit shown to the customer and the relevant ledger entries so it can resolve the issue without sending the customer between services.
Can this test work before an AI integration is live?
Yes. Compare your direct website with another supported checkout, such as an app or a partner channel. A simulated assistant can also reveal which rules are ambiguous. A live pilot is needed to verify the actual integration.
