Research report Rank 03
Agent Purchase Returns Desk
A merchant tool verifies delegated purchase records and routes authorized returns when an AI agent placed the order.
- Funding announcement
- Agentcard · $8M Seed
- Sector
- Commerce
- Business model
- B2B SaaS
- Gap type
- layer
13of 15
- Funding activity
- 4 out of 5
- Market gap
- 4 out of 5
- Ease of testing
- 5 out of 5
Customer hypothesis
Hypothesis: merchants may need a simple way to distinguish a buyer’s approval from an agent’s later return request. Start with a permission receipt and a human-reviewed return handoff, without issuing cards, moving money or replacing a returns platform.
Inside the buying decision
Who to recruit
Start with a merchant’s head of returns or customer experience. Screen for a store that can identify orders placed through a shopping assistant and show what happened when a shopper requested a change. An agent-platform developer can explain the authorization record, but only the merchant can tell you whether the exception consumes meaningful support time.
Follow one case
Trace one order from purchase confirmation to a request for a return. Who supplied the order reference, who asked for the return, which items were covered, and what did the merchant need to verify before releasing a label? A shopper approving the original purchase and an agent asking to return one item are two different events. Have an operator mark the point where the available evidence stops.
The alternative to beat
Set up two competing runs: the merchant’s existing returns portal plus an ordinary support ticket, and the same flow with a small authorization record attached. Include both an ordinary self-service request and a disputed delegated request. Only the second case might justify buying another service. Keep the current portal as the routing destination.
First customer test
Ask three merchants and two agent developers to walk through a synthetic purchase and return, then identify the exact approval fields both sides would accept.
Run the pilot
With merchant and developer permission, use test orders to cover an exact match, a partial return and an expired delegation. Give a support operator an order reference, allowed items and actions, expiration, and a link to the approval evidence. Have a human choose approve, reject or ask the original shopper. Count unresolved decisions and minutes per case. Keep the trial in review mode; it need not issue refunds or collect payment credentials.
- Advance if
- Run a paid pilot only if an operator can reproduce the permission decision in each test, the merchant can identify enough real cases to cover the price, and an authorized employee will own the exception queue after handover.
- Stop if
- Stop if stores cannot distinguish delegated transactions, if approval evidence cannot be associated with a specific item, or if their existing return form resolves the same cases in comparable time.
Pilot economics
40 exceptions × 6 minutes saved × $30/hour = $120 per month
The example values support time only. At a hypothetical $99 monthly price, the margin of benefit to this buyer is just $21 before integration and training. If the merchant sees ten exceptions per month, the time value is $30. Ask for the previous month’s actual exception count before doing an integration. Record verified avoided losses separately if a merchant has concrete evidence for them.
Test these numbers in the calculatorRisks to investigate
- Merchant platforms or payment providers may standardize the evidence before a new vendor gains adoption.
- A permission receipt may be insufficient for disputed orders, making human handling costly.
The comparable round
Agentcard
Seed ·
Named investors 3
Peak XV, Fin Capital, Y Combinator
Market evidence
Coverage proxy: Agentcard says its commerce tools can shop across more than 5,000 online stores. That is claimed store coverage, not active merchants, paying customers or transaction volume.
Why now
Agentcard now supports user approvals through passkeys, Face ID or voice. Capturing how authorization should survive a return or cancellation is a timely merchant-side question.
Where the merchant already makes a decision
Shopify documents that creating an open return through its returnCreate operation assumes the merchant has already approved the request. That makes the evidence available before approval a concrete place to test the idea. The trial can hand the decision back to the store’s current workflow.
shopify.devExisting players
The funding case
The $8m seed round shows early financing for agent payment infrastructure.
Agentcard’s distinction between saved-card authorization and newly issued cards suggests different evidence paths to test.
Loop’s existing returns workflows offer a possible integration boundary; the proposal is a delegation record rather than another full returns suite.