This article isn't about responding quickly to leads — we cover that in automating lead follow-up. This is about the step before that, the one almost nobody systematizes: deciding who is actually worth calling.
A sales team that works leads in order of arrival is spending its most expensive hour on its cheapest contact.
In This Article
- The two axes of every qualification model
- Which signals to score, and how much weight to give them
- What AI adds that classic scoring doesn't
- How to set the threshold
- False positives: the mistake that kills the model
The Two Axes of Every Qualification Model
Every model that actually works keeps two things separate that people tend to lump together:
- Fit. Is this the type of customer you want to sell to? Industry, company size, job title, geography. It doesn't change over time.
- Intent. Are they in a buying moment right now? Behavior, stated urgency, budget. It changes week to week.
Blending both into a single number is the most common mistake. A lead with a perfect fit but low intent needs nurturing; one with high intent but a poor fit needs to be told no. If both score 45, your sales rep has no idea what to do with either one.
How we set it up: two separate scores plotted on a four-quadrant matrix — call, nurture, disqualify, and watch. The rep sees the quadrant, not the raw number.
Which Signals to Score, and How Much Weight to Give Them
| Signal | Axis | Weight | Why it matters |
|---|---|---|---|
| Company size within your target range | Fit | High | Determines deal size and sales cycle |
| Job title with decision-making authority | Fit | High | No authority, no close |
| Industry within your specialty | Fit | Medium | Affects win rate |
| Corporate email address | Fit | Low | Cheap filter for noise |
| Visits the pricing page | Intent | High | The single most predictive signal |
| Mentions a specific timeline | Intent | High | Real, stated urgency |
| Asks for a comparison with a competitor | Intent | High | Actively evaluating options |
| Opens emails without clicking | Intent | Low | Passive interest |
| No activity in 30 days | Intent | Negative | Going cold |
Start with six to eight signals. Nobody keeps a thirty-variable model maintained, and within three months it's out of sync with reality.
What AI Adds That Classic Scoring Doesn't
Traditional lead scoring rates what a lead does: clicks, visits, opens. AI also rates what a lead says.
- It reads free-text form fields. "We need to solve this before September" isn't a click, but it's the strongest signal you'll ever get.
- It qualifies through conversation. An agent asks the two or three follow-up questions you're missing in the moment, instead of waiting for the call.
- It flags early disqualifiers. Students requesting info for an assignment, competitors snooping around, vendors fishing for a sale. Filtering them out saves hours.
- It summarizes context so the rep walks into the call already knowing what's bothering the lead.
This is what an AI agent does in practice — it's the difference between a form and a conversation.
How to Set the Threshold
Don't pick it by gut feel. Pull it from your own closed deals:
- Take your last 50 leads that converted into customers.
- Score them retroactively with your model, as if they had just come in.
- Find where the bottom 20th percentile falls. That's your starting threshold.
- Revisit it after a month with fresh data.
If you don't have 50 closed deals yet, start with a deliberately low threshold and raise it over time. It's better for a rep to disqualify a few extra leads than for the model to bury real opportunities for three months.
False Positives: The Mistake That Kills the Model
The risk isn't that the model lets a good lead slip through. It's that it sends junk to sales for two weeks straight — because after that, the team stops checking the alerts, and the system dies.
Two safeguards that work:
- Explicit negative scoring. Competitor domains, edu email addresses, intern job titles. Subtracting points matters as much as adding them.
- Weekly review for the first six weeks. The rep marks each alert "good" or "bad," and weights get adjusted. Half an hour a week for about six weeks.
Conclusion
An automatic lead qualification system that actually works isn't a complex algorithm. It's two separate axes, six to eight signals, a threshold pulled from your real closed deals, and weekly reviews at the start.
What determines success isn't how sophisticated the model is — it's whether the sales team trusts the alerts. And that trust gets destroyed by two weeks of false positives.
Want to build the model with your own data? Request a free assessment.
