Set up weekly voice-of-customer alerts from support tickets with AI

Support hears every complaint first, but those insights reach product, ops and marketing late or not at all. Monthly tag reports lag and only count categories, CX leads don't have time to read hundreds of transcripts, and new issues get noticed only once the volume hurts. Product and merchandising decisions get made without the customer's own words in the room.

1. Connect Claude to Yuma through the Yuma MCP, and connect Slack so Claude can post the report.
2. Load the weekly voice-of-customer prompts (https://docs.google.com/document/d/1QtFIs_iuWhsXgDKGUfCImKZXzkDzHyWkKkGoJnFSRLc/edit?usp=sharing) into a Claude project so the same instructions run every week.
3. Run the weekly scan: the last 7 days of tickets against a 4-week baseline. Ask for the top themes, the fastest-rising themes and any brand-new ones, each with a count, ticket IDs and a customer quote.
4. Map every theme to an owner: product for defects and sizing, ops for shipping and missing items, marketing for promo confusion, CX for policy gaps.
5. Audit before you send. Open two or three ticket IDs per flagged theme and drop anything that doesn't hold up.
6. Post the digest to a shared Slack channel: three things to act on with owners, three to watch, and what customers praised.
7. When a theme looks like a real problem, switch to a root-cause investigation for that one issue and pull the affected orders from Shopify.
8. Schedule the scan to run every week. During peak season, run it daily over the last 24 hours.
9. Close the loop. When a theme turns out to be a policy gap, update the Handbook so the agent answers it correctly next time.

Support becomes the company's fastest customer research channel. Product, ops and marketing get a short weekly read of what customers are saying, in their own words and with evidence attached, instead of a lagging tag report. Emerging issues get flagged in the week they start rather than the month after, and praise reaches the teams who earned it.

Compare every week against a rolling 4-week baseline so seasonal volume doesn't trigger false alarms. Give Claude last week's digest as context and keep theme names stable, so trends stay comparable week to week. Ask for ticket IDs on every theme and spot-check them before posting. Keep the Slack post short and put the full theme list in a thread. Include what customers praised, since that keeps teams outside support reading the channel. Try it on your own tickets with a 30-day free Yuma trial (https://yuma.ai/support-ai).

Coverage is limited to conversations that flow through Yuma. Themes are AI-generated groupings that can merge or split differently between runs, so a human review stays in the loop. Stores with low ticket volume may not have enough data for weekly trends and should run it every two weeks or monthly. Posting to Slack requires Slack connected to Claude. The digest surfaces signals, not root causes; confirming a cause still needs a dedicated investigation.