Most makers decide what to promote by vibes. You feature the product you’re proudest of, or the one a customer complimented last week, or whatever photo you happen to like. Meanwhile the actual best-seller, the quiet workhorse that pays the rent, sits un-promoted because it’s boring to you. Your shop already knows the answer. It’s sitting in your order history, and until now reading it meant a pivot table you were never going to build. Here’s how to analyze your Etsy sales data with AI so a plain order export tells you which products to promote, review, and pause this week. No spreadsheet skills required.
The unlock is that ChatGPT, Claude and similar tools can now read a file and do the math for you. File uploads and limits vary by plan, so check what yours includes. You upload a CSV, ask questions in plain English, and it comes back with revenue, units, and trends by product. That used to be a spreadsheet job. Now it’s a conversation, which means the excuse “I don’t do numbers” finally stops costing you sales.
Step 1: export the last 90 days and clean it first
In Etsy, pull your orders CSV (Shop Manager → Settings → Options → Download Data → Orders). Shopify has the same under Orders → Export. Ninety days is enough to see what’s selling now without dragging in ancient seasonal noise.
Before that file goes anywhere near AI, strip the personal data: customer names, addresses, emails, and order notes. You don’t need any of it to answer “what should I promote,” and you shouldn’t be uploading buyers’ details to a chatbot. Delete those columns, keep the product, quantity, price, date, and status columns. This one habit is the difference between a useful workflow and a privacy problem, and it’s the kind of back-office discipline that separates a real business from a hobby. Same spirit as the rest of the maker back-office basics.
Step 2: make AI check the file before it trusts it
Don’t ask for insights yet. Exports are messy: refunds counted as sales, duplicate rows, the same product listed under three slightly different names. If you skip straight to “what’s my best-seller,” you’ll get a confident answer built on dirty data. Open with this instead:
Check this file for missing values, refunds, duplicate orders, inconsistent product names, and any columns that shouldn’t be used for revenue analysis. Explain what you changed and why before you calculate anything.
Making it show its work first is the whole game. This is the same maker-brain-first way to prompt: you’re not outsourcing the judgment, you’re making the tool expose its assumptions so you can catch the one that’s wrong. If it says it’s merging “Oak Sign” and “Oak sign - personalized” into one product, you get to say yes or no.
Step 3: analyze Etsy sales data by asking real questions
Now the fun part. In plain English:
For the cleaned data, give me revenue, units sold, average order value, and refund rate by product or product family. Then show me week-over-week trend by product: what’s climbing, what’s fading.
You’ll see things vibes never told you: the “boring” product doing quiet volume, the darling you love that barely moves, the item whose refund rate is quietly eating its profit. Ask follow-ups the way you’d ask a smart assistant: “which products grew fastest in the last month?”, “what’s my AOV if I exclude the one big custom order?” No formulas, just questions. This is the everyday version of what I mean when I say one grounded AI workflow beats a pile of tools: one file, real questions, decisions you can act on today.
Here’s the kind of thing that turns up (illustrative numbers, so check your own). A shop’s favorite product, a big layered shadow box, brought in $1,400 over 90 days, but 3 of its 20 orders were refunded. Meanwhile a plain engraved keychain nobody thinks about sold 210 units for $2,300, with almost no refunds, and climbed every week. The shop had been promoting the shadow box.
Step 4: turn it into a promote / review / pause table
End every session with a decision, not a dashboard:
Give me a three-column action table (Promote, Review, Pause) and put the numbers behind every recommendation so I can sanity-check each one.
- Promote: strong and climbing. This is your feature this week, your next photo, your email.
- Review: decent but flat, high refund rate, or a name a buyer would find confusing. Worth a listing fix before you spend on it.
- Pause: dead weight. Stop giving it prime shelf space and attention.
Now “what should I post about this week” has an answer with a number attached. And if the table says a product’s worth promoting, that’s the one to point paid traffic at too. Before you pay for Etsy ads, make sure the money’s going behind something the data already likes.
The honest part
A sales export shows what sold, not what was profitable. Revenue is not margin. Before you make a real decision, like killing one product or doubling down on another, you have to add the costs the CSV doesn’t know: materials, your labor, marketplace and payment fees, packaging, and ad spend. A product can top your revenue chart and still lose money once you load in the blanks and the hours. So use AI to find the pattern fast, then run your best candidates through your actual cost math before you bet on them. For laser and UV work, the free Laser Pricing Calculator does that from real cut time, material cost and true margin, and Your Product Catalog keeps one record per product, so the size, material and price you check against don’t drift. The spreadsheet points; you still decide.
Use the speed. Keep the soul.
Keep reading
- A custom order DM just landed. Here’s how to turn the mess into a quote.
- A corporate gift order just landed in your inbox. Here’s the 12-minute reply.
- Etsy Banned the Slot Machine. Good.
Ready to put this to work? Join the Maker Growth Hub.