Customer Operations Templates
Two templates covering the post-sale relationship: proactive retention and multi-channel support.
create-leafmesh my-project --template customer-success # or:
create-leafmesh my-project --template customer-supportcustomer-success — Proactive Retention
16 agents · 3 crons — protecting the revenue you already have.
| Agent | Type | Role |
|---|---|---|
cs_operations_agent | Programmatic | CS operations — sweeps the whole customer base every 6 hours, plus on-demand DAU trends, feature breadth, quiet streaks for one account |
health_score_agent | LLM | Usage + support + billing + engagement → 0–100 score, *and* the "why" in plain English — quotable by CSMs |
risk_detector_agent | LLM | Usage decay, exec turnover, ticket spikes; high-ARR accounts always get human review |
playbook_trigger_agent | Programmatic | Pattern × ARR × segment → save / QBR / renewal / expansion play |
save_play_agent | LLM | Drafts the save play; top-tier urgent accounts force executive signature |
qbr_drafter_agent | LLM | QBR deck: usage data, ROI achieved, roadmap, asks |
renewals_agent | LLM | Renewals analyst — forecasts renewal probability + expansion potential; daily cron catches accounts entering the 30/60/90-day renewal window |
account_intel_agent | LLM | Account intelligence analyst — expansion/upsell signal scans plus champion-change detection and stakeholder-map rebuilds |
customer_onboarding_agent | Programmatic | Daily cron — post-sale milestones, time-to-first-value tracking, overdue → rescue play |
competitor_intel_agent | Programmatic | Displacement signals (reviews, tickets, CSM notes), channel-weighted severity |
csm / senior_csm / renewals_manager / ae / vp_cs | Human | Five-role roster matched to account stakes |
cs_audit_logger_agent | External | Immutable log of every decision and customer touch |
How it works. The 6-hourly sweep is the heartbeat — no waiting for customers to complain. Health scores feed risk detection feed the playbook matrix, which dispatches the right intervention to the right CSM tier. Three parallel sentinels run alongside: onboarding (the first 90 days), the renewals analyst's daily window check, and competitor displacement. Nothing customer-facing auto-sends — every play passes a human.
Value. Churn announces itself months early — in usage decay, champion departures, competitor mentions — but CSMs managing 40 accounts can't watch all of it. This mesh watches all of it, every 6 hours, and converts signals into ready-to-send plays. One saved enterprise account typically pays for the system for years.
customer-support — Multi-Channel Triage
14 agents · 1 cron
| Agent | Type | Role |
|---|---|---|
ticket_intake_agent | LLM | Intake owner for every channel (Slack, email, chat, phone, web, Telegram; Discord and WhatsApp coming) — canonical ticket, intent + priority classification, and voice-call handling (human / callback / self-serve) in one role |
sentiment_agent | LLM | Urgency / frustration scoring; churn-risk flags |
cross_channel_agent | Programmatic | Unifies one customer's threads across all channels |
response_agent | LLM | Retrieves the knowledge-base-grounded answer *and* drafts the reply — grounded, tone-aware |
qa_agent | LLM | Tone / accuracy / policy / PII gate before any customer sees anything |
crm_case_generator_agent | LLM | At resolution: categorised CRM record → Salesforce / HubSpot / Zendesk |
support_insights_agent | Programmatic | Support insights analyst — CSAT/NPS ledger + rollups plus trending/spiking ticket clusters → feeds the KB gap filler |
kb_gap_filler_agent | LLM | No KB hit? Drafts the missing article from the resolution (human-reviewed) |
sla_watchdog_agent | Programmatic | Every minute — past-SLA tickets escalate |
tier_1 / tier_2 / sme / queue_manager | Human | Three support tiers + queue owner |
support_audit_logger_agent | External | Immutable log of every reply and escalation |
How it works. Whatever channel a ticket arrives on — voice included — intake classifies it and sentiment reads it; cross-channel lookup gives the responder the customer's full history — a repeat customer is never a stranger. The responder pulls the grounded KB answer and drafts the reply, which passes the QA gate: wrong tone, policy breach, or PII never reaches a customer. Resolution feeds three loops: the CRM record, the insights ledger, and — if the knowledge base had no answer — a drafted article so the next identical ticket self-serves.
Value. Roughly 70% of support volume is routine, and routine tickets are exactly where humans burn out and SLAs slip. The mesh resolves the routine share with grounded, QA-gated replies and gives humans the 30% that needs them — with full context attached. The KB-gap loop means the self-serve share grows over time.