Every January, a wave of holiday returns lands on the customer service team. Retailers surveyed for NRF and Happy Returns' 2025 returns report expected 17% of holiday sales to be returned. Most of those are ordinary requests, each needing someone to read the order against the policy.
Retailers are bracing in the usual ways
Asked how they'd handle holiday returns, retailers in the same survey said they'd lean more on third-party logistics partners (49%), hire seasonal staff (43%) and extend return windows (37%), according to the American Journal of Transportation's write-up. Each of those adds people or time to the same manual decisions.
Most return decisions don't need judgement. The policy says what's allowed: the window, the condition, the categories that can't come back, the proof needed. The work is finding the order, checking it against those rules and writing the answer.
Write the policy down first
An agent can only apply a policy that's written down in full. Most return policies are a public page plus a set of exceptions that live in the team's heads: final-sale items bought in a promotion, gift receipts, orders split across two shipments, the extended window for items bought after a certain date.
Before October ends, get those exceptions into one document. Decide the edge cases now, while there's time to argue about them.
What the agent checks
Connect the agent, read-only to begin with, to order history and the returns queue. For each request it checks:
- whether the order exists and the item is on it
- whether the purchase date falls inside the window for that category
- whether the item is in a category that can't be returned
- whether the stated reason matches what the policy accepts
- whether the account shows a pattern worth a closer look
It then recommends approve, refuse or send to a person, and attaches the policy clause that applies. A request outside the written rules always goes to a person.
Let the agent approve first
Refusals reach customers and cost goodwill when they're wrong, so a person confirms each one at the start. Approvals inside the policy are lower risk. Once a few hundred recommendations have been checked and the agent applies the policy the way your team would, let it approve the clear cases on its own and keep refusals and exceptions with people.
Watch the overrule rate by category. If people keep reversing the agent on gift returns, the gift rules in the policy document are probably unclear.
Run it before the rush
The time to test this is November, while volumes are still normal. Run the agent alongside the team on real requests, compare its recommendations with what people decided, and fix the policy document where they differ.
Frequently asked questions
How many holiday purchases get returned?
Retailers surveyed by NRF and Happy Returns for their 2025 returns report expected 17% of holiday sales to be returned. The same report found 9% of all returns were fraudulent.
What can an agent do with return requests?
Check each request against the written policy, using the order record, the purchase date, the item category and the reason given, and recommend approve, refuse or send to a person. Clear cases can be approved automatically once the logs show the agent applies the policy correctly.
Should an agent refuse returns on its own?
Not at first. Refusals affect customers directly, and a wrong one costs the most goodwill. Have a person confirm every refusal, and let the agent approve only the cases the policy clearly allows.
Can AI help with return fraud?
It can flag patterns for review, such as repeated returns from one account or a mismatch between the item ordered and the item described. NRF's 2025 report found 85% of the large retailers surveyed were using AI to detect or prevent return fraud.