---
name: voice-of-customer-digest
description: Synthesize the week's support tickets into themes and route the top signal to product. Use this skill when: Every Monday, 8:00 AM — Weekly digest of last week's tickets.
---

# Weekly support-ticket synthesis for product

Goal: synthesize the week's support tickets into themes and route the top signal to product.

Connected tools this skill uses: zendesk, intercom, slack, linear.

## When to use

- Every Monday, 8:00 AM — Weekly digest of last week's tickets.
- Or when the user asks directly to synthesize the week's support tickets into themes and route the top signal to product.

## Method

Run the weekly voice-of-customer digest.

1. Pull every ticket opened or closed in {{support_tools}} since the last
   digest ({{last_run_date}}).
2. Cluster the tickets into root-cause themes, not raw tags. For each
   theme, count tickets and list 2-3 representative ticket links.
3. Diff this week's theme volumes against last week's saved digest: what
   grew, what shrank, what's new.
4. Validation: for every theme, confirm each linked ticket actually
   supports the theme's description. Drop any ticket that's a loose match
   instead of force-fitting it.
5. Save the digest to `reports/voc-digest/{{week}}.md` and post the top 3
   themes, with trend and ticket links, to {{delivery_channel}}.
6. For any theme appearing 3+ weeks running with no product movement, file
   a Linear issue in {{product_team}} summarizing the pattern and linking
   the source tickets.

Do not respond to or close a customer ticket directly, the digest is for
the CS lead and product to act on.

## Theme clustering

Group raw tickets into root-cause themes instead of relying on surface-level tags.

## Method
1. Pull every ticket opened or closed in the support tools since the last digest.
2. Read each ticket's actual content, not just its assigned tag.
3. Group tickets by the underlying root cause they describe.
4. Merge tickets that use different tags but describe the same root issue.
5. Split a tag that actually covers two distinct root causes into separate themes.
6. Count tickets per theme and select 2-3 representative examples.

## Rules
- Cluster by root cause, never by the ticket's existing tag alone.
- A theme needs a minimum ticket count of (set on first run, then stored in memory) to be reported.
- Keep the theme taxonomy in memory and refine it as new patterns emerge, don't restart it weekly.
- Prefer a specific root-cause label over a vague catch-all theme name.

## Output
A list of themes, each with a ticket count and 2-3 representative ticket links.

## Trend diffing

Compare this week's theme volumes against the last digest to surface real movement.

## Method
1. Load last week's saved digest from reports/voc-digest/.
2. Match this week's themes to last week's by root cause, not by label text.
3. Compute the change in ticket count per theme: grew, shrank, unchanged, new.
4. Track how many consecutive weeks a theme has appeared, even as its ticket count fluctuates.
5. Note whether a shrinking theme is resolved (fixed) or just quiet this week.
6. Save this week's digest as the baseline for next week's diff.

## Rules
- Always diff against the immediately prior digest, never an older one.
- A theme counts as "new" only if it did not appear in the prior digest under any label.
- Track consecutive-week streaks precisely — the escalation threshold depends on this count.
- Keep every weekly digest on file for the running trend history.

## Output
A trend line per theme: grew, shrank, unchanged, or new, plus its consecutive-week streak.

## Verbatim sourcing

Keep every theme backed by real, verifiable ticket links, never paraphrase alone.

## Method
1. For each theme, attach 2-3 actual ticket links, not summarized descriptions.
2. Re-read each linked ticket and confirm it genuinely supports the theme's stated description.
3. Drop any ticket that is only a loose or partial match — do not force-fit it to hit a quota.
4. Prefer tickets with a direct customer quote over ones requiring inference.
5. Carry the same verified links forward if a theme repeats across weeks.

## Rules
- Every theme ships with real ticket links attached, never description-only.
- Drop a loosely matching ticket instead of forcing it to fit a theme.
- A theme with zero genuinely supporting tickets does not get reported.
- Preserve verbatim customer language in representative excerpts where used.

## Output
Each reported theme paired with 2-3 verified ticket links and, where available, a verbatim excerpt.

## Recurring jobs

This skill pairs with scheduled jobs. Offer to set each one up; only create the ones the user approves:

- Weekly voice-of-customer digest run (Every Monday, 8:00 AM) — https://eluu.ai/cookbook/voice-of-customer-digest/jobs-setup/weekly-voice-of-customer-digest-run.md