WhatsApp and email, one customer graph

From first target
to closed deal.

An AI-native revenue platform for businesses that sell through conversations. Build audiences, run outbound, answer every WhatsApp and email in one inbox, qualify with agents you can actually govern, and connect it all back to revenue.

Replacesa shared WhatsApp number

No sign-up. Real product, seeded with demo data.

Arjun DeshpandeSundaram TextilesWA
Hi, what are your rates Mumbai to Jebel Ali?
Happy to help — we run that lane weekly. What are you moving, and roughly how many containers a month?

Sales Agent

Cotton yarn, around 24 TEU a month
That's a good running volume. Are you buying FOB or EXW?

Sales Agent

FOB. First shipment in about 3 weeks
Rohan here from pricing — USD 780 all-in per 40'HC, weekly direct. Can hold it 14 days.

Rohan · took over

Qualify freight enquiry6/6 captured
OriginNhava ShevaDestinationJebel AliCommodityCotton yarnVolume24 TEU/moIncotermFOBTimeline3 weeks

Built on

WhatsApp Cloud APIAmazon SESMeta AdsGoogle AdsPostgres + pgvectorAWS ap-south-1Gemini

One inbox, one pipeline, one answer to “what worked?”

Every conversation lands in one governed inbox

WhatsApp and email in a single queue with ownership, SLA timers and an agent that qualifies — until a human takes over, explicitly.

  • The agent asks, captures and knows when to stop
  • Collision detection — two people never answer the same thread
  • Hindi and English are both first-class

Arjun Deshpande

FOB. First shipment in about 3 weeks

AI12m

Meera Iyer

Can you send the revised quote today?

Human4m

Faisal Khan

What's the transit time to Rotterdam?

AI38m

Deals move only when the fields say they can

Qualification against fields you define, a lifecycle that refuses illegal moves, and a board that shows who owns what.

  • Stages and fields are configuration, not code
  • Scoring you can read rule by rule
  • Approvals for anything that spends money

Qualified2

Sundaram Textiles

₹4.2L

Apex Polymers

₹1.8L

Proposal1

Veda Interiors

₹2.4L

Kohinoor Freight

₹3.1L

Revenue walks back to the touch that caused it

Attribution is computed over the timeline itself — ad click to cold email to WhatsApp reply — and frozen the moment a deal closes.

  • Cost per qualified lead, not cost per form fill
  • One person, many addresses, one record
  • Closed history never moves under a model change

Revenue by first touch

₹18.6L

Ad clickCold emailWhatsAppWon

Every stage walks back to the touch that caused it

Most stacks lose the thread somewhere between the ad click and the closed deal. Here the whole chain is one query over one graph, so “which campaign produced this revenue” has an answer rather than an estimate.

  1. Audience

    18,400
  2. Touched

    6,820
  3. Conversation

    612
  4. Qualified

    209
  5. Opportunity

    63
  6. Won

    17

Five surfaces, one system

Not five products stitched together — they share a contact, a consent record and a timeline, which is the only reason the attribution above is possible.

Rules are enforced in code, not requested in a prompt

An instruction in a system prompt is a suggestion the model may or may not follow. Here the rules live outside the model, at three enforcement points: a gate before any tool runs, a gate before anything is sent, and a verifier on every reply.

  • The agent never denies being an AI — refused by the verifier, not asked for in the prompt
  • Every outbound message passes a consent, suppression and sending-tier check
  • Anything that spends money is surfaced for approval, never applied silently

Never quote a rate before the lane and volume are captured

Tool refused

The pre-tool gate blocks it. Not a prompt instruction — a refusal.

Hazardous cargo always goes to a human

Escalated

Forced handover on topic, before the agent can answer.

Discounts above 10% need manager approval

Approval required

Lands in the approvals queue with the number attached.

The agent asks, captures, and knows when to stop. The handover to a human is explicit — and that is the whole trick.
The operating principle behind every agent here

One person, many addresses, one timeline

Somebody clicks an ad, replies to a cold email, then messages your WhatsApp number from a phone you have never seen. Most systems make that three records. Here it is one.

  1. 1

    Ad click

    Meta lead form, 3 months ago

  2. 2

    Cold email

    Sequence step two, opened twice

  3. 3

    WhatsApp reply

    New number, matched by signed token

  4. 4

    Closed won

    ₹2.4L, attribution frozen at close

Built for businesses that qualify before they quote

Nothing here is hardcoded to a vertical. What you capture, the stages you move through, the rules the agent obeys — all configuration, which is why two businesses that sell completely differently run on the same system with no bespoke code.

See it running, with data in it

The demo is the actual product. Two seeded businesses, real conversations in English and Hindi, and every governance rule enforced exactly as it would be in production.