An AI lead qualification agent that runs BANT in 60 seconds on WhatsApp

How the AI qualification agent runs end to end across an Indian inbound funnel

1. 01

   Capture and language detection in under five seconds

   Inbound from WhatsApp, your website form, click-to-WhatsApp Meta Ads, JustDial, IndiaMart or your CRM landing form lands in one unified inbox. The agent detects Hindi, English, Hinglish, Marathi, Tamil, Telugu, Kannada or Bengali in the first message and replies in the user's preferred language immediately. No awkward English-only default that signals robot.

2. 02

   Five to seven BANT questions in 60 to 90 seconds

   The Gemini-powered agent asks budget, authority, need and timeline questions one at a time in conversational tone, adapts the next question based on previous answers, skips questions where the lead has already volunteered the data, and never feels like a chatbot interrogation. The whole exchange completes in 60-90 seconds end to end with structured BANT data captured.

3. 03

   Score, route and seamless human handoff

   Each lead receives a 0-100 score against your explainable rubric with the contributing signal weights visible to the human reviewer. Hot leads above your configurable threshold ping the on-call human inside HubSpot, Zoho CRM, Salesforce, Sell.do, LeadSquared or your CRM of choice. Cold leads enter a 21-day nurture sequence and are continuously re-scored on each reply.

• B2B SaaS demo-request inbound triage

  Demo requests get qualified for company size, use-case, current tooling and decision-maker fit before a human SDR ever opens the thread or wastes a Calendly slot on a lookalike.

  See AI workflow automation for service businesses
• Real-estate site-visit booking and routing

  Property enquiries qualified on budget, configuration preference, end-use vs investor intent and possession timeline; warm leads routed to the right RM with the full BANT transcript already attached.

  See WhatsApp lead automation for real estate India
• Clinic appointment triage and routing

  Patient enquiries categorised by symptom urgency, insurance partner, preferred specialist and slot availability before the receptionist sees them. DCI-compliant on tone and never making outcome promises.

  See doctors and clinics vertical playbook
• Professional services qualification on WhatsApp

  CA, advocate, architect and consultancy enquiries scored by engagement size, urgency and partner-fit before a human partner replies. ICAI and BCI compliant on solicitation language.

  See B2B services vertical playbook for India
• Missed-call to WhatsApp conversion loop

  Every missed call rings back on WhatsApp inside 60 seconds via Exotel or Knowlarity handoff. The AI agent picks up the qualification thread the moment the caller replies in chat.

  Use the missed lead cost calculator for India
• Multi-channel followup orchestration

  Once qualified, the follow-up cadence routes across WhatsApp, email and CRM tasks based on the lead's preferred channel, declared explicitly during qualification or inferred from reply patterns.

  Compare CRM vs WhatsApp vs email for followup

Frequently asked questions

• How is the AI trained? Is this just a generic ChatGPT wrapper with a system prompt?

  No. The agent is built on Gemini via Vertex AI (we have zero OpenAI dependency and no plans to add one) and configured per deployment with your ideal customer profile, your qualification rubric, your tone-of-voice guide and your edge-case library extracted from past sales transcripts. The prompt and the rubric are version-controlled artefacts in a Git repository you own from day one. Generic ChatGPT wrappers fall over on Hinglish code-switching, Tier-2 and Tier-3 city language patterns and India-specific objections. Ours is tuned on exactly those signals.
• What is the false-negative rate, and what happens to leads the AI misclassifies?

  Our production baseline is roughly 4-7% false-negative rate (warm leads scored cold) once the agent has 200+ training conversations and the scoring rubric has stabilised. Every cold-scored lead still enters the 21-day nurture sequence and is automatically re-evaluated on each new reply, so a misclassified false-negative surfaces within days, not weeks. Hot leads where the agent's confidence score is low are routed to a human review queue with a flag, never silently parked or lost. The full audit trail is visible in the dashboard.
• What is the escalation logic? When does a human take over the conversation?

  Three escalation paths in the default configuration: (1) lead score above your team's configurable threshold triggers an immediate human ping on Slack, Telegram or your CRM activity feed, (2) the lead explicitly asks for a human and the agent recognises the intent in eight Indian languages including Hindi, English, Hinglish, Marathi, Tamil, Telugu, Kannada and Bengali, (3) the lead asks anything outside the qualification rubric scope. The handoff always includes the full WhatsApp transcript and the structured BANT answers, so the human picks up mid-conversation, not at zero.
• Does this integrate with HubSpot, Zoho, Salesforce, Sell.do and LeadSquared?

  Yes, these are the five CRMs we have shipped against most often across our Indian deployments and the integrations are battle-tested. Integration approach is bidirectional: lead capture and structured BANT answers push into your CRM as a new lead record with custom fields, and status changes in your CRM trigger nurture sequence changes in the agent. We have also shipped against Freshsales, Pipedrive, Microsoft Dynamics and bespoke MySQL or Postgres CRMs via the standard webhook API. The integration is part of the 14-day build, never an add-on charge.
• How is data privacy and DPDP Act 2023 compliance handled?

  All conversation transcripts are stored encrypted at rest in Indian-region Supabase (Mumbai data centre) with row-level security per tenant. Personally identifiable data is access-logged and exportable on demand for DPDP Act subject-access requests, with a documented seven-day response SLA. We do not send any PII to OpenAI, Anthropic or any non-Vertex AI provider. Google's India-region inference is the only model endpoint touched. Per-deployment you can configure a 90, 180 or 365 day retention window after which conversation transcripts auto-delete while preserving aggregated scoring telemetry.
• What does this cost? Is it a SaaS subscription or a build-and-own engagement?

  Build-and-own, never SaaS. You pay a one-time implementation fee for the 14-day build, then a predictable monthly run cost that covers Gemini token usage, BSP messaging fees, monitoring, and the supervised tuning loop in the first 90 days. Most Indian sales teams fit into the Growth Engine tier; high-volume teams with 5,000+ inbound enquiries a month graduate to the AI Transformation tier. We never quote prices on landing pages because the right tier depends on volume. The 20-minute audit produces a fixed quote within 48 hours that holds for 30 days.