---
name: d2c-catalog-regrouping
description: Regroup a D2C product catalog by what items actually are before analysing sales, stock, or performance. Use whenever the user hands over a Shopify, WooCommerce, or similar catalog and asks a question about a product group ("how are our T-shirts doing?", "is upperwear selling?", "should we cut winterwear?"). The category tags in these catalogs are almost always wrong or inconsistent — Kurtis get filed under "Ethnic", Henleys under "Winter Essentials", hoodies under "Lounge" — so trusting the labels will skew the answer. Trigger this skill even when the user does not ask for regrouping; if the task involves judging how a product type is performing and the input is a raw catalog, regroup first.
---
# D2C catalog regrouping
## The problem this skill solves
A D2C founder asks a simple question: "how is our upperwear performing?" You feed the Shopify catalog to an LLM and get a clean-looking report. The report is wrong.
It is wrong because the catalog's own category tags are a mess. One merchandiser tagged Kurtis as "Ethnic". Another tagged Henleys as "Winter Essentials". A third dropped hoodies into "Lounge". The word "upperwear" does not appear anywhere in the data, so the LLM reports on whatever it finds labelled "T-shirts" and "Polos" and calls it a day. Growth looks flat. The founder nearly cancels an inventory order for a line that is in fact selling out.
Labels lie. Treat them as a hint, not the truth.
## When to use this skill
Use it whenever all three of these are true:
- The user has given you a product catalog (Shopify export, WooCommerce CSV, Magento dump, spreadsheet of SKUs, or similar).
- The user is asking a question about a product group, not a single SKU. Examples: "how are our T-shirts doing", "should we reorder winterwear", "which category is underperforming", "is upperwear selling".
- The answer depends on grouping SKUs correctly.
If any one of those is true, regroup before you answer. Do not skip this step because the catalog looks tidy. Tidy catalogs lie too.
Do not use this skill for single-SKU questions ("how many units of SKU 1234 did we sell last week"), for questions that are purely about tags themselves ("list every tag in the catalog"), or for non-product data.
## The core method
Think of the catalog as a spice box where every tin is mislabelled. Do not read the label. Open the tin.
Work through these steps in order.
### Step 1. Ask what the user means by their group
The user said "upperwear". What counts? T-shirts, polos, shirts, kurtis, henleys, hoodies, sweatshirts, jackets? All of these cover the upper body, so all of them count. Write the list down before you touch the data. If the list is not obvious, ask the user one short question. Do not guess silently.
Do the same for any other fuzzy group the user names: "winterwear", "loungewear", "festive", "basics". Pin down the items first.
### Step 2. Read the SKUs, not the tags
For each product in the catalog, look at the product title and description, not the category or tag fields. A product titled "Men's Cotton Henley Full Sleeve" is a henley, no matter what the tag says. A product titled "Floral Kurti with Palazzo" is a kurti.
The title is almost always honest. The tag is often not.
### Step 3. Build your own group
Put every SKU whose title or description matches the user's group into one bucket. Ignore the original category. Keep a second bucket of SKUs you are unsure about and flag them for the user.
### Step 4. Answer from the new bucket
Now run the user's question against your regrouped bucket, not the catalog's. If the question is about sales, pull sales for every SKU in your bucket and sum. If the question is about stock, do the same.
### Step 5. Show your working
Tell the user what you put in the bucket and what you left out. Something like: "I counted 142 SKUs as upperwear, including 31 kurtis that were tagged 'Ethnic' and 18 henleys that were tagged 'Winter Essentials'. I left out bralettes and camisoles — tell me if you want those in." The user needs to be able to correct you.
## What a good answer looks like
**User prompt:** "Here is our Shopify export. How is upperwear performing this quarter?"
**Bad answer:** Filter by category = "Upperwear", report the numbers, done. (The category field is either empty or wrong.)
**Good answer:**
