VoC digest
Synthesizes a weekly voice-of-customer digest from support tickets, sales call notes, reviews, and community threads in Computer — themed, counted, quoted, and routed to product. Use when the user mentions voice of customer, VoC, customer feedback summary, "what are customers saying," or product asks for feedback themes.
When to run
- Recurring: weekly digest to the product channel (recommended)
- On demand: "what's the feedback on the new billing flow"
Computer scopes
- Support: tickets and CSAT verbatims
- Sales + CS: call notes and meeting summaries
- Synced external sources: review sites, community forum threads
- Product: existing feedback/backlog items for linkage
Workflow
- Collect the week's feedback-bearing records across all sources; dedupe cross-channel echoes of the same customer voice.
- Cluster into themes; for each theme count distinct accounts (not mentions — one loud account is a signal, ten quiet ones are a different signal; report both).
- Attach 2–3 representative verbatims per theme, cited, chosen for specificity over spice.
- Link each theme to existing backlog items where they exist; mark genuinely new themes.
- Compute deltas vs. prior weeks: rising, falling, new. The delta is the headline.
- Draft the digest for the product channel; post only via the configured channel with L1 approval the first four runs, then L2 with audit if the team promotes it.
Guardrails
- Starts L1 (drafted post, human approves), promotable to L2 for the posting step only.
- Verbatims are quoted exactly and attributed to account tier, never to named individuals in the shared channel.
- Counts are account-distinct and auditable; no theme without at least two independent sources unless flagged single-source.
Output
Weekly digest: delta headline → theme table (theme, accounts, trend, linked backlog item) → verbatim blocks → new-theme flags. Archive stored in Computer for trend history.Eval fixtures
- Set: 4 weeks of sandbox feedback (600 records) with gold theme assignments and known deltas.
- Rubric: theme clustering F1; account-count accuracy; verbatim fidelity (exact quotes); delta correctness vs. seeded history.
- Ship gate: clustering F1 ≥ 0.85; count and quote errors 0.
MORE SKILLS
Customer Support & CX · L0
Escalation brief
Generates the complete engineering escalation pack for a support ticket — reproduction steps, environment details, timeline, prior contacts, account impact, and related incidents — from Computer history. Use when a ticket needs engineering escalation, when the user mentions escalating, filing a bug from a ticket, or an engineer asks for context on an escalated issue.
Customer Support & CX · L1
KB gap miner
Mines resolved support tickets for recurring questions that have no knowledge-base coverage, ranks the gaps by deflection value, and drafts the missing articles from actual resolutions. Use when the user mentions knowledge base gaps, deflection, self-service improvement, "we keep answering the same question," or content planning for the help center.
Customer Support & CX · L3-gated
Refund approval pack
Assembles a policy-checked refund or credit recommendation — eligibility against written policy, account context, precedent cases, and financial impact — then halts for human approval before anything moves. Use when the user mentions a refund request, credit request, billing dispute, or make-good, or a support ticket contains a refund ask.
Customer Support & CX · L0
Root-cause clusterer
Collapses a spike or backlog of support tickets into probable root-cause clusters, linking related incidents, known bugs, and recent releases from Computer. Use when ticket volume spikes, when the user asks "what's driving all these tickets," mentions a backlog surge, or during incident review.
