Forty return claims, and one of them is lying.
Sorting honest returns from the odd ones without treating every customer like a suspect.
Read this claim, the photo and the order. Does it fit our return policy? Tell me what you checked and anything that looks off.
Claudeputs the claim, photo and order side by side
Geminichecks it against your policy and its exceptions
Codexsorts all claims into a review list
Returns, September
- 11 of 38 claims were the same lamp, cracked base
- Listing photo makes the shade look larger than it is
- Asked supplier about packaging on the lamp
Every claim looks the same at first. A photo, a sentence or two, an order number. To decide, someone on your team has to open the order, check the product listing, look at the photo properly, and see whether this customer has done this before.
- Good customers wait days for a simple refund
- Decisions depend on who is tired that afternoon
- Nobody sees that one product keeps coming back
Try it yourself in Luminair
Six steps. No setup beyond the AI plan you already have.
- 1
Install Luminair and sign in
Download Luminair and open it. Go to Settings → Models → Add account and sign in with an AI plan you already pay for, like Claude, ChatGPT or Gemini.
- 2
Make a folder for this job
In the sidebar, choose New folder… and name it “Returns”. Then choose New session in this folder. Everything for this job stays together.
- 3
Bring the information in
You need: customer message, photo, order details and return policy. Connect WooCommerce and Intercom in Settings → Connectors, or export the files and drag them into the chat.
- 4
Ask for the result
Paste this into the session, or say it in your own words:
Read this claim, the photo and the order. Does it fit our return policy? Tell me what you checked and anything that looks off.
- 5
Get a second opinion
Pick a different model in the model picker and ask it to check the first answer:
Does the photo actually match what the customer describes?
- 6
Make it repeat
Open Flows → New and add the steps: Read claim, photo and order history → Sort: clear, needs a look, unusual → Draft replies for clear ones, hold the rest. Choose when it runs, like every Monday at 09:00. Nothing is sent or changed until you approve it.
How it plays out
One folder for claims
Create a Returns folder. Write your return policy into Journal once, in your own words, so every answer follows the same rules.
Hand over the evidence
Paste in the customer's message, the photo and the order details. Ask for a first read: does the photo match the description, does it fit the policy, is anything odd. It tells you which detail made it hesitate.
Sort, then decide
A Flow can sort each claim into clear, needs a look, or unusual, and draft the reply for the clear ones. The unusual ones land in Actions with the reason attached, waiting for a person.
You stay in charge
A refusal is always a human decision. Luminair prepares the file and says what it noticed, your team makes the call.
Once a month, ask Journal what the claims have in common. A bad batch or a misleading photo shows up as a pattern long before it shows up in your margins.
Sign in with the AI plan you already have and start with this job.