Lead Scoring Is Broken. Here's How to Fix It.

The Point System Problem

Downloaded whitepaper: +10 points. Visited pricing page: +15 points. Job title contains "Director": +20 points. Score hits 100: "Marketing Qualified Lead."

Sales calls. Nobody answers. Or worse - they answer and have no idea why they're being called.

This is lead scoring at most companies. Arbitrary points, arbitrary thresholds, arbitrary handoffs. The model was set up years ago. Nobody remembers why "webinar attendance" is worth 25 points.

Why Points Don't Predict

Activity isn't intent. Someone binge-reading your blog might be a student writing a paper. Someone who visited once and went straight to pricing might be ready to buy tomorrow. Points can't tell the difference.

Firmographic fit isn't timing. Perfect ICP company, perfect title, zero budget until Q3. Your score says "hot lead." Reality says "not now."

Decay is ignored. Activity from six months ago still inflating scores. That engaged prospect moved on. Your model doesn't know.

Gaming is easy. Marketing wants to hit MQL targets. They adjust point values until numbers look good. Sales gets the same garbage leads with higher scores.

What Actually Predicts Buying

Forget points. Look at patterns.

Velocity matters more than volume. Three pricing page visits in one day beats thirty blog visits over six months. Compressed activity signals active evaluation.

Specific pages tell you everything. Integration docs, security/compliance pages, pricing - these indicate serious evaluation. Generic content consumption doesn't.

Return visits beat first visits. Someone who comes back after going dark is re-entering the market. That's a trigger, not just another data point.

Multiple stakeholders appearing. Second person from same company hits your site? Deal is real. Someone's building internal consensus.

Building Scoring That Works

Start with closed-won analysis. What did actual buyers do before they bought? What pages did they visit? What was the timeline? Build your model backward from success.

Separate fit from engagement. Two scores, not one. Fit score: are they the right company? Engagement score: are they actively evaluating? Both need to be high.

Add decay aggressively. Points should expire. Activity from 90 days ago shouldn't count the same as activity from yesterday.

Validate with sales feedback. Are high-scoring leads actually converting? If sales says the leads are garbage, the model is wrong. Full stop.

Once you've built a better scoring model, the next challenge is actually building nurture sequences that convert those scored leads into opportunities.

The Machine Learning Question

Yes, ML-based scoring outperforms manual point systems. No, you probably don't have enough data to build one.

You need thousands of closed-won and closed-lost examples to train a decent model. Most companies don't have that. And if your CRM data is messy (it is), garbage in, garbage out.

Start with behavioral signals and pattern matching. Graduate to ML when you have the data foundation.

The Real Fix

Lead scoring isn't a set-it-and-forget-it system. It's a hypothesis that needs constant testing.

Review monthly: Are high scores converting? Are low scores getting ignored that shouldn't be? What's changed in buyer behavior?

The model is never done. The market keeps moving. Your scoring has to move with it.

And if you're still trying to figure out which channels are actually driving those leads, you might find that attribution is more broken than you think.