A whole month of tickets, in seven minutes.
A monthly summary of 9,001 support tickets against our help center: 16 draft FAQs and two real production bugs.
I asked for a monthly summary of the help center, not just yesterday’s daily package.
I wanted to know what customers ask that help.e-vrit.co.il doesn’t answer. Which answers are weak, and what’s worth highlighting.
Background:
Since October 4, a daily routine has run at 10:27.
It reads yesterday’s Glassix (our ticketing system) tickets, and cross-checks them against the roughly 239 articles in the help center.
Then it writes draft Q&As in Hebrew. Each draft is marked as new, overlapping, or strengthening. Nothing gets published without me.
What happened:
On October 8, I asked for the full month.
Within about 7 minutes, 30 days of Glassix exports were gathered, from September 8 to October 7. 9,001 unique tickets, without a single missing day.
Every recurring topic was ranked by volume and trend, and marked as missing, weak or covered.
The big ones:
- Shipping: about 1,350.
- A paid-for book won’t open: about 850.
- Coupons: about 480.
- Stuck payment: about 430.
- Login: about 410.
- Gifts: about 300.
Some topics are rising sharply. Part of an order out of stock is up 3x. Also rising: a damaged or wrong printed book, “delivered but never arrived”, and cancellations.
It was all merged into one file of 16 drafts. Six of them cover about 2,700 tickets a month, and can be published with almost no policy decisions.
Bugs that an FAQ (frequently asked questions) can’t solve were split out.
The shaliach also corrected its own wrong guess about one-click purchase in the app. It did that after checking a live article. (Shluchim, singular shaliach, is Hebrew for “emissaries”. It’s what I call my AI agents.)
The follow-up that same night:
I asked which email exactly had the broken “contact us” link.
The shaliach read full ticket threads, and found it. It’s the delay email, “The books you ordered are running a little late”, sent at around 18:00.
It’s quoted in 146 tickets, not 37. There are 16 explicit complaints, and it’s still live.
A second bug wasn’t in an email at all. The “order received” SMS links to the QA server (the test server).
The shaliach opened two bugs in Linear (task tracker), E-369 and E-370, using the BUG template. They were assigned to a developer, with evidence and a request to reproduce.
A checkout failure with a kibbutz address went to Nitza. She reproduced it against QA, and suggested a fix to the site’s copy.
Original
Pinch, double-tap or scroll over the image to zoom. Drag to move.
Why this is good:
One picture of the month: what’s missing, what’s weak and what’s rising.
Ready drafts, ranked by impact. And two real bugs on the live site, with proof.
All from the tickets, without me digging by hand even once.
The shluchim involved: the customer service shaliach for the analysis, the drafts and Linear, the Glassix puller for the daily pulls, and Nitza for the UX (user experience) reproduction and the recommendation.
What to take from this:
A daily routine builds an archive. Because there were 30 daily exports, the monthly summary took minutes.
Ask “where exactly?”. One follow-up question turned a vague problem into a specific email, with the real number, 146 and not 37.
Separate FAQ from bug. Some of the pain is content, and some is code. Each goes to the right place.
The shaliach corrects itself against a live source. Ask it to check a real article before it draws a conclusion.