ع
Start Topics Teams Reference What's new Saved
playbook

Run a weekly support operating system

Triage the queue, draft the high-volume replies, escalate the real bugs, refresh the macros, and roll the patterns into a voice-of-customer report — one repeatable rhythm that shrinks the queue at its source.

advanced ~half a day
when to reach for this

Run support reactively and you spend every week emptying a bucket that refills overnight — answering the same questions, missing the bug behind fifty of them, and never getting upstream of the cause. The fix is a weekly operating rhythm where each piece feeds the next: triage finds the patterns, replies clear the volume, bugs get escalated, the macro library and help center absorb the repeats, and a report rolls it all up so product fixes the source. This is the capstone that ties the other Customer Support playbooks into one loop — so the queue gets smaller every week instead of just emptier each night.

gather this first
  • The week's ticket export as tickets-week.csv — drop it into the chat in Claude Desktop (or open the folder it lives in), scrubbed of names, emails, and account numbers to [customer] first.
  • Your reusable assets: macros.md from the canned-response library and your help-center article list, so this week's patterns update them instead of starting cold.
  • A fixed slot on the calendar — same half-day each week — because the power of an operating system is that it runs on a rhythm, not when you remember to.
the workflow

  1. Triage the week to find the patterns

    Start by running the Find what the queue is really about playbook on the week's export — cluster by root issue, rank by volume, separate the bugs from the noise. Everything downstream keys off this map.

    you ask
    Read tickets-week.csv. Confirm the row count and date range, then group by underlying issue (not wording), give me the top 6 clusters with counts and one example each, and flag which clusters look like a recurring bug versus normal support. This is the input for the rest of the week's work.

    what you get back A ranked, count-backed map of the week — "login loop (52), double-charge (38, looks like a bug), shipping delay (29)..." — with the bug candidates flagged. The same output the triage playbook produces, now feeding the loop.

    Verify the headline counts against the raw export here — every later step trusts this map, so a wrong number propagates all the way to the report.

  2. Clear the volume with drafted replies

    Run the Draft a reply that sounds like a person playbook against the biggest clusters, pulling from macros.md where a template already fits — so you clear the most tickets per minute without going robotic.

    you ask
    For the top 3 clusters, draft a warm reply in my voice for each — own the issue, one clear next step, [name] and [order number] placeholders, and pull from macros.md if a template fits. Do not invent any order number, refund, ship date, or policy. Flag any cluster where a single reply can't fairly cover the variation.

    what you get back Three on-voice draft replies keyed to the highest-volume clusters, placeholders for every private detail, and a flag where a cluster is too varied for one template — clearing the bulk of the queue fast.

  3. Escalate the real bug, not the symptoms

    Take the flagged bug from triage and hand engineering a tight, evidence-backed report instead of forwarding forty tickets — the count and shared trigger are what get it prioritized over answering it fifty more times.

    you ask
    The double-charge cluster is flagged as a bug (38 tickets). Write a 4-sentence escalation for our tracker: what's happening, who's affected and how many, the shared trigger, and 2 example ticket IDs as evidence. Neutral, specific, no guesses about the code.

    what you get back A paste-ready bug report with the pattern, the count, the trigger, and example IDs — the one escalation that stops the cluster at its source instead of generating thirty-eight more replies next week.

  4. Refresh the macros and help center from the repeats

    Feed this week's recurring clusters into the canned-response library and help center playbooks — promote the questions that keep coming back into a template or an article, so next week's volume is lower by design.

    you ask
    Two clusters repeat from last week: 'how do I change my plan' and 'where's my invoice.' For each, draft a macro for macros.md AND a short help-center article (numbered steps, plain language, 'still stuck?' line, [confirm the steps] where you don't know the real flow). These should absorb the question so it stops hitting the queue.

    what you get back A new macro and a draft help article per repeating cluster — the mechanism that converts a recurring ticket into a self-serve answer, shrinking the queue rather than re-emptying it.

    This is the step that makes the system compound — every week a few repeats become permanent self-serve answers, so the baseline volume trends down.

  5. Roll the week up into the report

    Close the loop by running the voice-of-customer playbook on the week's clusters — themes, friction, requests, bugs, with counts and impact — so leadership and product fix the causes you can't fix from the queue.

    you ask
    Roll this week's clusters into a short voice-of-customer report: a 4-bullet summary, the themes split into friction / requests / bugs / praise with counts and anonymized quotes, and the top 3 things product should look at with owners. Frame the rankings as recommendations to verify. Paste-ready for our team channel.

    what you get back A factual, count-backed weekly report with a prioritized handoff to product — the upstream half of the system, sending causes to the people who can remove them so next week's queue starts shorter.

make it your own
  • Run the loop component by component: each step is a full playbook on its own — Find what the queue is really about, Draft a reply that sounds like a person, Build a canned-response library you'll actually reuse, Build a help center from the questions you actually get, and Turn a month of tickets into a voice-of-customer report — so you can dive into any one when it needs more depth than the weekly pass gives it.
  • Monthly zoom-out: keep the weekly rhythm tactical, then run the voice-of-customer playbook on a full month every fourth week for the trend lines a single week can't show.
  • Automate the rhythm (Power Track): wire the whole loop into a /support-week custom command or a scheduled agent (see the Playbook's Capabilities tab) that drafts the triage map and the report cut from Monday's export, so you start each week editing instead of from a blank queue. Custom commands and scheduled agents are the opt-in Power Track — the whole loop runs perfectly well by hand in Claude Desktop until you're ready to wire it together.
watch out for
  • The loop only compounds if you verify at each handoff — confirm counts against the raw export, check that drafted replies promise nothing unauthorized, and validate help-center steps in the real product before any of it propagates downstream.
  • You'll touch a full week of tickets in one sitting — scrub names, emails, and account numbers to [customer] before upload, keep [placeholders] in every reply, and anonymize quotes in the report.
  • Claude runs the rhythm; a human owns every decision in it — which bug to escalate, what to promise a customer, whether a recommendation in the report is right. The system makes you faster at the work, not absent from it.

you'll end up with A repeatable weekly rhythm that empties the queue and shrinks it — patterns found, volume cleared, the real bug escalated, repeats converted into self-serve answers, and the week rolled into a report — so support gets upstream of its own work instead of refilling the bucket every night.

Questions people ask

How is this different from running the individual support playbooks?
It chains them into one weekly loop where each piece feeds the next: triage finds the patterns, replies clear the volume, bugs get escalated, the macro library and help center absorb the repeats, and a voice-of-customer report rolls it all up — so the queue shrinks at its source instead of just emptying each night.
How long does the weekly loop take, and how often do I run it?
Plan about a half-day, in a fixed slot on the same day each week. The power of an operating system is that it runs on a rhythm rather than when you remember to — and the repeats it converts into self-serve answers make each week's baseline volume trend down.
What makes the queue actually get smaller over time?
The step that feeds recurring clusters into your macro library and help center: every week a few repeats become permanent self-serve answers or escalated bug fixes, so the cause is removed instead of re-answered. That compounding is what shrinks the baseline.
Does Claude make the decisions in the loop?
No. Claude runs the rhythm, but a human owns every decision — which bug to escalate, what to promise a customer, whether a report recommendation is right — and verifies counts and help-center steps at each handoff. It makes you faster at the work, not absent from it.