Before You Start: What Data to Export

These prompts only work if you feed ChatGPT the right raw data. Here's exactly what to pull:

‍

1. Shopify Orders Export Go to Admin → Orders → Export. Include: Order ID, created date, product title, SKU, variant, quantity, total price, tags, fulfilment channel, discount codes used.

‍

2. Shopify/ Returns Apps Data Export Go to Admin → Analytics → Reports → Returns. Include: return reason, return date, refund amount, resolution type (refund/exchange/store credit), items returned.

‍

3. If you use Return Prime Export directly from the Return Prime dashboard. You'll get richer fields: standardized return reasons and sub reasons, returns converted to exchanges, refund methods, amount captured for higher exchanges/upsell, refund TAT per return, approval TAT per return, product details etc. You also get customer NPS scores and feedback if you have opted for it.

‍

4. Combine into one CSV Join on Order ID. This is the master file you'll paste into ChatGPT. For large datasets (10K+ rows), use ChatGPT's Advanced Data Analysis mode - it can process the full file without you needing to chunk it.

‍

On data privacy: Before pasting any CSV into ChatGPT, strip PII - customer names, emails, and phone numbers. You don't need them for any of these. Keep Order IDs as your join key.

‍

10 ChatGPT Prompts to Analyze Your Shopify Store's Return Data

Each prompt is copy-paste ready. Text in [SQUARE BRACKETS] is the only thing you change.

Prompt 1: The Bleed Rate by SKU

Category: Revenue Leak

Your return rate at the store level means almost nothing. You need it at SKU level. One problem product returning at 45% can inflate your whole store average and you'd never know until you break it down.

‍

Prompt:

I'm sharing a CSV of my Shopify store's orders and returns data. Each row has: Order ID, product title, SKU, variant, order date, return date, return reason, refund amount, and resolution type.

Analyze the data and give me:

1. Return rate by SKU (returns ÷ units sold), ranked highest to lowest. Show SKU, product name, units sold, units returned, return rate %.
2. The top 10 highest-returning SKUs.
3. For each of those top 10, show the breakdown of return reasons (e.g. "wrong size" 40%, "defective" 30%, "not as described" 30%).
4. Total refund value lost per SKU across the period.

Format output as a table. Flag any SKU with return rate above [YOUR THRESHOLD - e.g. 20%] in the output.

‍

What to act on:

• SKUs where "not as described" or "wrong colour" dominates → product page problem, not a product problem. Fix the images and copy before you discontinue the SKU.
• SKUs where "sizing" is the top reason → consider adding a size guide, fit notes, or a size recommender.
• SKUs where "defective/damaged" is high → supplier or packaging issue. Pull the batch info.

Return Prime note: If you export from Return Prime, return reasons are already standardized into categories. This makes the ChatGPT analysis significantly cleaner and more reliable.

‍

Prompt 2: Return Rate by Acquisition Channel

Category: Revenue Leak

Customers from different channels return at radically different rates. Meta ads and Google Shopping customers often have inflated return rates because they bought impulsively. Knowing this changes how you calculate true channel ROAS.

‍

Prompt:

Using the same orders + returns CSV:

My Shopify orders include a UTM source or discount code column that identifies acquisition channels. The channels in my data are: [LIST YOUR CHANNELS - e.g. meta_ads, google_shopping, email, organic, influencer_collab].

1. Calculate return rate by channel (returns ÷ orders from that channel).
2. Calculate average order value by channel.
3. Calculate net revenue per order by channel (AOV minus average refund value for that channel).
4. Rank channels by net revenue per order, not just by AOV.

Add a column: "Adjusted ROAS impact" - describe qualitatively whether each channel's return rate makes it more or less valuable than the raw ROAS suggests.

‍

*Note: Raw RoAS data would make this analysis richer. This will not be available in Shopify or return app. 

‍

What to act on:

• If Meta ads show 30%+ return rate vs 12% for email → your Meta creative may be overpromising. Audit your top-spend ad creatives against what you're actually shipping.
• If influencer collab codes have high return rates → that influencer's audience doesn't match your product. Factor this into repeat partnership decisions.
• Low return rate from email → these are your most loyal, accurate buyers. Double down on retention.

‍

Prompt 3: The Sizing Problem Detector

Category: Product Intelligence

For apparel brands, sizing returns are frequently misread as a product quality problem when they're actually a product page problem. This prompt separates the two and tells you exactly which size runs are causing the most friction.

‍

Prompt:

Using the returns CSV, filter for all returns where the reason includes "size", "fit", "too big", "too small", or "sizing".

For this subset:

1. Which product categories have the highest sizing-related return rate?
2. Which specific size variants (e.g. XS, S, M, L, XL) are returned most - as a % of units sold in that size, not just raw volume.
3. Is there a consistent pattern - e.g. does XL have higher return rate than XS across multiple products? 
4. Are sizing returns concentrated in specific product lines (e.g. trousers vs t-shirts)?

Output a table of: Product | Size Variant | Units Sold | Units Returned for Sizing | Sizing Return Rate %

Then write a 3-sentence hypothesis about what's driving the sizing returns based on the data.

‍

What to act on:

• If one size (e.g. L) has 3x the return rate of adjacent sizes → that size likely runs small. Add a fit note on the product page: "Our L runs small - size up if you're between sizes."
• If sizing returns are concentrated in one category → that category needs a dedicated fit guide.
• If sizing returns are evenly spread → the issue is likely the overall size chart accuracy, not a variant-specific problem.

‍

Prompt 4: The Serial Returner Segment

Category: Customer Behaviour

Across eCommerce, a small percentage of customers account for a disproportionate share of returns. The Wall Street Journal reported retailers flagging "serial returners" whose return rates exceed 50% of purchases. This prompt finds yours, without profiling by identity.

‍

Prompt:

Using the orders + returns CSV (with customer IDs or hashed email, NOT raw emails):

1. Calculate each customer's personal return rate: total items returned ÷ total items ordered.
2. Segment customers into four buckets:
   
   • Never returned (0%)
   • Occasional returner (1-20%)
   • Frequent returner (21-50%)
   • Serial returner (50%+)
3. Show: how many customers are in each bucket, what % of total customers they represent, and what % of total refund value they account for.
4. For the serial returner segment: what is their average order value vs store average? What are their most common return reasons?
5. What % of serial returners made a repeat purchase after returning? (This indicates whether they're loyal but hard to fit, or genuinely exploiting the policy.)

‍

What to act on:

• If serial returners still buy repeatedly → they're not bad customers. They may need better fit guidance or a different product recommendation at checkout.
• If serial returners have zero repeat purchases → consider tightening your return policy for this segment (e.g. requiring photos, limiting free returns after N returns).
• High AOV serial returners → worth a 1:1 outreach or personal shopper program rather than a blanket policy change.

Return Prime note: Want automated flags return frequency per customer? Reach out to us to configure this for your store.

‍

Prompt 5: The Discount Code Return Correlation

Category: Revenue Leak

Discounted orders frequently have higher return rates. But not all discount types behave the same. A 10% loyalty discount for instance behaves very differently from a flash 40%-off sale code. This prompt maps the relationship precisely.

‍

Prompt:

My orders data includes a discount code column. Using the full orders + returns CSV:

1. Group orders by discount type. Classify them as:
   
   • No discount
   • Loyalty / repeat customer code
   • Seasonal sale code (e.g. BFCM, SUMMER40)
   • Influencer / affiliate code
   • First-purchase code
   • Other / unknown (Use the discount code names I'll describe: [LIST YOUR DISCOUNT CODES AND THEIR TYPES])
2. For each discount type, show:
   
   • Number of orders
   • Average order value
   • Return rate %
   • Average refund amount
   • Net revenue per order (AOV minus average refund)
3. Which discount type generates the highest net revenue per order?
4. Which discount type has the worst return-to-revenue ratio?

Output as a ranked table, worst ratio to best.

‍

What to act on:

• If first-purchase codes have high returns → your acquisition offer may be drawing in the wrong buyer. Test smaller discounts with better product education instead.
• Low return rate on loyalty codes → rewards programs reduce returns. Consider shifting the budget from acquisition discounts to retention incentives.

Prompt 6: Return Rate vs. Review Score Correlation

Category: Product Intelligence

If you have product review data, combining it with return data reveals whether your reviews are actually predicting returns or missing complaints entirely. Sometimes a 4.7-star product has a 30% return rate because negative buyers don't leave reviews.

‍

Prompt:

I'm giving you two datasets:

1. Returns data by product (SKU, product name, return rate %, top return reason).
2. Product review data (SKU, product name, average star rating, number of reviews, most common review tags if available).

Please:

1. Merge the two datasets on SKU / product name.
2. Is there a statistically meaningful correlation between review score and return rate?
3. Flag any products that have: high review score (4.5+) BUT high return rate (20%+). These are "silent failure" products.
4. Flag any products that have: low review score (below 4.0) AND low return rate. These are products customers keep despite disliking - a potential loyalty risk.
5. For "silent failure" products, what are the most common return reasons? This tells you what reviewers aren't saying.

‍

What to act on:

• "Silent failure" products with high returns despite good reviews → the return reasons expose a product description gap that reviewers have accepted but new buyers haven't expected.
• Products with low reviews AND low returns → customers are keeping something they're unhappy with. Follow up with a post-purchase email survey. This is churn risk hiding in plain sight.

‍

Prompt 7: Time-to-Return Analysis

Category: Customer Behaviour

How quickly a customer returns tells you a lot about what went wrong. Returns within 48 hours of delivery usually signal a description mismatch. Returns at Day 25-28 in a 30-day window usually signal policy gaming or buyer's remorse after use.

‍

Prompt:

Using the returns data (with order date, delivery date if available, and return initiation date):

1. Calculate "days to return" = return initiation date minus delivery date (or order date if delivery date is unavailable).
2. Segment returns into time buckets:
   
   • 0-3 days: Immediate return
   • 4-10 days: Early return
   • 11-20 days: Mid-window return
   • 21-30 days: Late return / policy-limit return
   • 30+ days: Out-of-window (if you accept these)
3. For each time bucket: number of returns, % of total returns, most common return reason, average refund value.
4. Is there a spike in Day 25-30 returns? If so, what products dominate that bucket?
5. Do immediate returns (0-3 days) have a different reason profile vs late returns?

‍

What to act on:

• Spike in 0-3 day returns with "not as described" → your product images or descriptions are misleading. Immediate fix.
• Spike in Day 25-30 returns → customers may be using the product and returning at the last minute. Consider shortening your return window or adding a "used item" exception.
• Consistent mid-window returns → usually genuine dissatisfaction. Dig into the reason codes for that window specifically.

‍

Prompt 8: Exchange Rate and Revenue Recovery Audit

Category: Operations

A return processed as an exchange retains the revenue. A return processed as a refund loses it. Most brands have no idea what % of their returns could have been exchanges - or what that represents in actual money.

‍

Prompt:

Using the returns data with resolution type column (refund / exchange / store credit):

1. What % of all returns were resolved as: refund vs exchange vs store credit?
2. What is the total refund value that went out as cash refunds?
3. What is the total value retained through exchanges and store credit?
4. For returns resolved as refunds, what were the most common return reasons? Are any of those reasons - e.g. "wrong size", "wrong colour" - ones that could have been resolved as an exchange instead?
5. Calculate: if [X%] of "wrong size" and "wrong colour" refunds had been converted to exchanges instead, how much revenue would have been retained? Use the average order value from the dataset.
6. What is the current exchange conversion rate, and how does it compare to the apparel benchmark of 30-40%?

‍

What to act on:

• If the exchange rate is below 20%, you are leaving significant revenue on the table. A self-serve exchange option in your return portal immediately changes this.
• If "wrong size" is a top refund reason, this is the highest-value exchange opportunity. Offer an automated size swap with zero friction in the return flow.
• If store credit uptake is low, your store credit incentive may not be compelling enough. Test adding a 10-15% bonus credit (e.g. return £50 → get £57.50 in credit).

‍

Return Prime note: Wonder Smart Exchange automatically surfaces the correct size/colour variant during the return flow - before the customer selects a refund. This single feature typically moves exchange rate from sub-20% to 30-45% for apparel brands within 60 days of enabling it.

‍

Prompt 9: Return Rate by Geographic Region

Category: Revenue Leak

If you ship to multiple regions or countries, return rates vary significantly by location - driven by delivery time, carrier damage rates, and regional sizing norms. Brands that don't track this attribute all regional issues to product quality when the real cause is logistics.

‍

Prompt:

Using the orders + returns CSV with shipping city/state/country data:

1. Group orders and returns by region. Use: [CHOOSE: country / state / city / postcode prefix - depending on your data granularity].
2. Calculate return rate by region.
3. For the top 5 highest-returning regions:
   
   • What are the most common return reasons?
   • What is the average delivery time for orders from that region (if the data has fulfilled date and delivery estimate)?
   • Are returns from that region concentrated in specific product categories or SKUs?
4. Are there any regions where "damaged / arrived broken" is a disproportionately high return reason? This signals a carrier or packaging problem, not a product problem.

Output: Region | Orders | Returns | Return Rate % | Top Return Reason

‍

What to act on:

• High "damaged" returns from a specific region → switch carrier or improve packaging for that route.
• High "sizing" returns from a specific country → sizing standards differ. Add region-specific size guides (UK vs US vs EU) on the product page.
• High return rate in regions with long delivery times → set better expectations at checkout, or explore a local fulfilment option.

‍

Prompt 10: The Full Revenue Leakage Summary - Your Action Prioritization Matrix

Category: Operations

After running the individual prompts, this is the synthesis step. It produces a ranked list of the highest-impact interventions sorted by estimated revenue recovery potential - so you know where to start first.

‍

Prompt:

I've run several analyses on my Shopify return data. Here is a summary of the key findings:

[PASTE YOUR FINDINGS FROM PROMPTS 1-9 HERE AS BULLET POINTS - e.g.:

• Top returning SKU: Blue Linen Shirt L, 42% return rate, primary reason: sizing
• Serial returners: 8% of customers, 34% of refund value
• Exchange rate: 14% (vs 30-40% benchmark)
• Highest return channel: Meta ads, 28% return rate
• Region spike: Germany, 35% return rate, top reason: damaged in transit]

Based on this data:

1. Identify the top 5 highest-impact fixes ranked by estimated revenue recovery potential.
2. For each fix, specify: the problem, the exact action to take, who owns it (marketing / ops / product / tech), estimated effort (high/medium/low), and estimated revenue impact.
3. What are the quick wins I should tackle in the next 2 weeks (low effort, high impact)?
4. What are the structural fixes to plan for next quarter (high effort, high impact)?
5. Are there any anomalies or patterns in this data that I haven't mentioned but that you'd flag as worth investigating?

Format as a prioritized action matrix.

‍

What to act on:

• This prompt turns your analysis into a project plan. Don't skip it - the individual prompts give you data, this one gives you decisions.
• Share the output directly with your ops, marketing, and product teams. It's already structured for a stakeholder meeting.
• Run this every quarter to track whether your interventions are working and what new leaks have appeared.

‍

‍FAQs

What data do I need from Shopify to use these prompts?

Export your Orders report and Returns report from Shopify Admin → Analytics → Reports. You need order ID, product, SKU, return reason, refund amount, and order date at minimum. If you use Return Prime, you can export a richer dataset including exchange rate, return resolution type, and customer segment.

Can I use these prompts with Claude or Gemini instead of ChatGPT?

Yes. These prompts are model-agnostic. For large CSV files, Claude's 200K context window handles bigger datasets without needing to chunk the file.

Do I need to know how to code?

No. Every prompt is written for plain English output - tables, ranked lists, and bullet-point summaries.

How often should I run this analysis?

At minimum, monthly. For stores doing 1,000+ orders/month, run the SKU and exchange rate prompts weekly. During peak seasons, run the time-to-return and discount correlation prompts during the 14-30 day return spike window post-sale.

My Shopify return reasons are customer free text, not structured categories. What do I do?

Add a pre-step: paste a sample of 200-300 return reason entries and ask ChatGPT to classify them into 6-8 categories (sizing, defective, not as described, wrong item, changed mind, delivery issue, other). Use those classified categories in the main prompts. Going forward, Return Prime standardizes reason codes at the point of customer submission - so you won't have this problem on future exports.

‍