The brain that watches our ops — and flags what we'd otherwise miss.
~20+ anomaly rules in production. Billing detection cut from 7 days to ~24 hours. Knowledge writes ~23× faster after the Vertex AI move.
Small service teams hold the whole customer relationship in WhatsApp and a CRM — but the context is scattered across thousands of chats, leads, invoices and call logs that no one can hold in their head. Cold leads slip, renewals come due unnoticed, vendor threads get lost. Most “AI” answers hand the conversation to a bot, which owner-operated businesses aren't ready to do.
From setup to second nature.
Ingest your history
The brain reads your WhatsApp chats, CRM leads, customers, invoices and call records, then distils them into structured facts, lifecycle events and relationship links — a knowledge graph of how your business actually runs.
Emit suggestion cards
A background service watches for the moments that matter (“this lead has gone cold for 4 days,” “this renewal is due”) and writes a card to a staff desk inside the CRM — with a drafted message ready to go.
Staff review with full context
Each card carries its reasoning plus a Customer Intelligence Capsule — a cross-source mini-brief so the staffer sees the whole customer without tab-hopping.
A human sends
The staffer sends with one click via a wa.me link from their own WhatsApp, or places a click-to-call. The system never sends on its own.
Inside Operations Brain.
Knowledge graph over your own data
Years of WhatsApp + CRM history become a structured, queryable store of facts, lifecycle events, pricing signals and relationship links — built from your operations, not a generic template.
Suggestion desks that coach, not replace
Staff-facing desks surface the next best action as a reviewable card with a drafted message. The Lead Desk is in production; Customer Care and Channel Partners desks are in development.
Per-card reasoning
Every card shows why it was suggested, so staff trust it and learn from it instead of blindly clicking. No black-box prompts to send.
Customer Intelligence Capsule
A cross-source brief on the customer — BLUF summary, status pills, recent activity, open suggestions and provenance — rendered inline, with graceful degradation when an ID can't yet resolve.
Human-in-the-loop by design
The brain suggests and drafts; your staff member always acts. There is no autonomous send.
Self-health monitoring
~20+ automated health rules run on a live tile, a daily anomaly check and a daily narrative pulse — catching token, delivery, stuck-card and cost regressions before they bite.
Numbers from running it ourselves.
Every figure here is measured on our own business — customer zero. No external clients, no projected numbers.
- Vertex AI fallback: the fact-extraction backend was failing ~half its calls (HTTP 503) on AI Studio; moving the same model to Vertex AI dropped that to ~0 — an availability win, not an accuracy claim.
- ~23× faster knowledge-write after refactoring to batched upserts (~4× on the full pipeline).
- ~20+ automated health rules run live across a dashboard tile, a daily anomaly cron and a daily narrative pulse. Billing detection cut from 7 days to ~24 hours.
- Three staff desks, soft-claim ownership, per-card reasoning and a wired Customer Intelligence Capsule ship in the live CRM module — with their own smoke tests.
- All of it runs on our own business first. Customer zero — we'd lose money if it broke.
Who actually uses this.
The owner-operator drowning in WhatsApp
You run the relationship in chat and a CRM, you have years of history, and nothing surfaces the lead that went quiet or the renewal due next week.
The small team not ready for a bot
You won't hand customer conversations to autonomous AI, but you'd take an assistant that drafts the message and hands it to a human. Human-in-the-loop is the point.
The relationship-heavy service business
Virtual offices, agencies, professional services — where cold leads, due renewals and vendor relationships all live in messages, and context is the whole job.
The stack
- WhatsApp. Read-only ingestion into the knowledge graph; outbound via wa.me click-to-send from staff devices (no WABA required).
- Perfex CRM. Desks, cards and the Capsule render inside the CRM sidebar; reads leads, customers, invoices and call records.
- Cloud telephony. Click-to-call from a card; call records feed back into the brain.
Straight answers.
Does it message my customers automatically?+
No. The brain drafts the message and surfaces the context; a human reviews and sends every message. Autonomous AI auto-reply is on the roadmap and intentionally paused — it is not shipped.
Where does the knowledge come from?+
Your own operational history — WhatsApp chats, CRM leads, customers, invoices and call records — distilled into a structured knowledge graph. It learns your business, not a generic playbook.
Which desks are actually live?+
The Lead Desk is in production. The Customer Care and Channel Partners desks are in development. The Customer Intelligence Capsule is v1, surfacing across views.
Will it work if my data is messy / mixed-language?+
The pipeline was built on a real, code-mixed (Manglish/English/Hindi) corpus. Messy, multi-year, multi-language history is the design target, not an edge case.
Where this honestly stands
Customer-zero work, running on our own business first. Lead Desk is in production; Customer Care and Channel Partners desks are in development; the Customer Intelligence Capsule is v1. Autonomous AI auto-reply (the AI Worker) is roadmap and paused — not shipped.
Works with the rest of the platform.
Scope a pilot of Operations Brain.
Tell us your workflow on a 30-minute call. We'll show you how this runs for us and how it would run for you.
30 minutes · no obligation · we reply within 1 business day