Intercom + OpenAI Lead Qualification Playbook (That Sales Teams Actually Use)

I Was Losing Good Leads in the Inbox Pile

A few weeks ago, I checked our inbound conversations and found a pattern I did not like.

High intent leads were landing in the same queue as support questions, partnership pitches, and random "can you build my app for $300" messages. Everything looked urgent, so nothing got handled fast enough.

The painful part was response time. A real buyer would ask a pricing or integration question, then wait hours for a useful follow-up. By then, they were already talking to competitors.

So I rebuilt the intake flow with one goal: qualify serious leads fast, route them to sales immediately, and keep noise away from the pipeline.

This is the exact setup I am using now with Intercom and OpenAI.

What This Workflow Actually Does

When someone starts a chat, the assistant asks a short set of qualification questions, scores the lead, and sends the conversation down one of three paths:

1. Sales now for high intent prospects
2. Nurture for leads that are real but early
3. Support or dead-end for non-buying conversations

The key is that the AI does not replace your sales team. It handles early triage so your team spends time where revenue is most likely.

The 5 Qualification Signals That Matter Most

I tested longer qualification forms first. They hurt conversion.

Five signals gave me enough accuracy without killing chat completion rate:

• Use case clarity: Can they explain the problem they need solved?
• Company fit: Team size, industry, or business model match
• Urgency: Are they buying now or "just exploring"?
• Authority: Are they a decision maker or just researching?
• Technical fit: Do they need integrations you can support?

This is better than asking for ten fields and getting half-empty answers.

Intercom Setup: Keep the First Message Short

In Intercom, I use a custom inbound bot flow with a short opener:

"Happy to help. Are you looking for support with an existing account, or evaluating us for your team?"

That single split removes a lot of junk from sales routes.

Then I collect four data points:

• Team size range
• Primary goal
• Timeline
• Work email

If a visitor refuses all context and asks for a demo right away, I still pass them to a human. But if they engage, the AI can qualify with much better precision.

OpenAI Prompt Design: Ask, Score, Route

The model prompt is simple and strict. It has three jobs:

1. Extract lead signals from chat replies
2. Return a lead score from 0 to 100
3. Return one route: sales_now, nurture, or support

I keep output in JSON so it plugs cleanly into automation.

Example scoring logic:

• +25 if they have a clear use case
• +20 if timeline is under 60 days
• +20 if they are likely decision maker
• +15 if company size fits ICP
• +20 if requested integrations match your stack

Then thresholds:

• 75-100: sales_now
• 45-74: nurture
• 0-44: support

I learned this the hard way: if your prompt is vague, routing gets weird fast. Be explicit about what counts as buying intent.

Routing Rules That Improved Close Rate

Here is the routing logic that worked best for me.

Route 1: sales_now

When score is 75+, the flow:

• Creates a high-priority lead in CRM
• Posts a Slack alert to sales with summary and score
• Offers instant calendar booking inside chat

The Slack alert includes:

• Company and role
• Use case summary in one sentence
• Urgency level
• Any integration blockers

This cut first-response delay for hot leads from hours to minutes.

Route 2: nurture

For mid-score leads, I do not force a demo.

Instead, the bot offers:

• A relevant case study
• A short implementation guide
• Optional follow-up in 2 weeks

These leads are added to a nurture list with the captured context. Sales can still review them, but they are not clogging the urgent queue.

Route 3: support or non-fit

If someone is clearly a support request, they go to support.

If they are not a fit, the bot still responds politely with the best next step. That protects brand experience and keeps your team focused.

The Human Handoff Script Matters More Than You Think

My first version had robotic handoff lines like "You have been qualified for a sales conversation."

Terrible.

Now I use this style:

"Thanks, this is helpful. Based on what you shared, you are likely a strong fit. I can connect you with our sales team now, or you can grab a time directly on the calendar."

Small change, big impact. More people complete booking when the handoff sounds human.

Metrics to Watch in Week 1

Do not just track booked demos. Watch the full funnel:

• Chat-to-qualified rate
• Qualified-to-meeting rate
• Meeting show rate
• Lead response time
• False positive rate (bad leads sent to sales)

After tuning for one week, my false positives dropped sharply and sales stopped complaining about "junk demos."

If false positives are high, tighten scoring on urgency and authority. If you are filtering too hard, relax the threshold from 75 to 70 and retest.

Common Mistakes I Made

1) Asking budget too early

Early budget questions scared people off. I now ask budget only after basic fit is clear.

2) Over-automating edge cases

I tried to automate every branch. Waste of time. Handle common paths first, then patch edge cases after real traffic.

3) Sending giant AI summaries to Slack

Nobody reads walls of text in a sales channel. Keep alert summaries under 5 lines.

4) No weekly prompt review

Language drift happens. New offer, new ICP, new objections. Review and tune prompts every week.

Practical Stack (Minimal Version)

If you want this live quickly, use this minimal setup:

• Intercom: capture responses and trigger webhook
• OpenAI API: score and classify lead
• Zapier or n8n: route by score and write to CRM
• CRM (HubSpot/Pipedrive): owner assignment and lifecycle stage
• Slack: real-time hot lead alert

You can build this in one afternoon if you keep scope tight.

Is It Worth It?

If inbound volume is low, manual triage is fine.

If your team handles enough chats that hot leads wait in queue, this workflow pays for itself fast. The value is not only automation. It is speed and consistency at the exact moment someone is ready to buy.

My advice: launch a simple version this week, monitor real conversations, then tune scoring based on what your sales team calls "actually qualified."

That feedback loop is where the real gains happen.

And once this is stable, you can layer in account enrichment, intent data, and personalized follow-ups without rebuilding the foundation.