Inbound interest goes cold fast. A qualification agent works the moment someone shows up — on the website, WhatsApp, or email — asks the right questions, and hands a warm, qualified lead to a rep with context. Here's how to build one.
The use case
A visitor asks "do you support SSO?" or "what's pricing for 50 seats?" Instead of a form and a 24-hour wait, an agent answers from your real docs, learns who they are, and books the demo while intent is high.
The agent
One agent, scoped to sales. Attach knowledge: pricing, security overview, and product docs so answers are grounded and cited — not improvised. Choose a fast model for snappy back-and-forth (switch per agent any time).
The qualifying prompt
You qualify inbound leads for Acme. In a warm, brief tone, one question at a time,
learn: team size, primary use case, and timeline. Answer product/pricing questions
only from the attached knowledge and cite the source.
Never invent discounts, SLAs, or commitments.
When the lead is a fit and ready, offer the next step:
- self-serve trial → share the start link
- wants a walkthrough → book a demo and capture name + work email
If they ask for custom pricing, security review, or procurement, hand off to sales.The tools
- Knowledge search — grounded, cited answers.
- Book appointment (built-in) — put time on a rep's calendar.
- CRM connector (e.g. HubSpot/Salesforce via MCP) — create or update the lead.
- Slack connector — ping the owning rep with the transcript.
The workflow
- Trigger: conversation tagged "sales intent."
- LLM + KB: qualify and answer questions.
- Branch: qualified & ready → booking path; needs human → hand off.
- Booking path:
book_appointment→ CRM create/update lead → Slack notify the rep with summary + citations. - Run record: every booked demo leaves a reviewable trail for RevOps.
Keep it honest
The fastest way to lose a technical buyer is an agent that overpromises. Ground every product claim in the knowledge base, and route security/procurement questions to a human. A qualification agent's job is to advance the deal, not to close things it can't back up.
Measure
Track chats → qualified → demos booked → showed. Tune the qualifying questions and the "ready" threshold from real transcripts.
Next: writing agent system prompts and tool calling with MCP.