There’s a specific kind of meeting every marketing ops person dreads. Sales comes in with a spreadsheet, points at a list of MQLs, and says some version of: these aren’t real leads.
I’ve been in that meeting. And the uncomfortable truth I’ve had to sit with more than once is that they had a point.
Not always. But often enough that it made me look hard at the scoring models I’d built and ask whether they were actually measuring what I thought they were.
What lead scoring is supposed to do
The premise is simple. Not every contact in your database is equally ready to talk to Sales. Some are actively evaluating. Some are casually curious. Some downloaded a whitepaper two years ago and haven’t been back since. A lead scoring model sorts them out. It assigns a number that reflects how ready someone actually is, so Sales knows who to call first.
When it works, it works well. Reps spend time on contacts who are actually in market. Marketing gets a signal on which programs are producing real pipeline. The handoff between both teams stops being a point of friction.
When it doesn’t work, you get that meeting.
Where it usually goes wrong
The most common problem I’ve seen isn’t technical. It’s that the model gets built once and nobody touches it again.
Lead scoring models are built on assumptions. This behavior signals intent. This job title means it’s the right buyer. This number of page views means someone is evaluating. Those assumptions might be right when the model is first built. But products change. The buyer profile shifts. The model keeps running on the original logic and quietly diverges from reality.
I’ve audited scoring models that were still giving points for visiting a pricing page that no longer existed. Models that weighted a job title heavily that the company had stopped selling to a year earlier. Nobody meant for that to happen. Recalibration just never made it onto anyone’s list.
The second problem is simpler. The model measures activity, not intent. Downloading three ebooks isn’t the same as being ready to buy. Attending a webinar doesn’t mean someone is evaluating your product. But scoring systems often treat engagement as a proxy for intent, and those two things are not the same. A contact can be very engaged and completely unqualified. A contact can visit your pricing page once, from a competitor benchmarking you, and light up your entire model.
Sales figures this out fast. They call a few high-scoring leads, get nowhere, and stop trusting the number.
What actually made a difference
The models that held up over time had a few things in common.
Sales was involved in building them. Not just consulted at the end. Actually involved. The people who talk to buyers every day have pattern recognition that doesn’t exist in any database. They know which questions signal real evaluation versus early curiosity. They know which company types close versus which ones go dark after two calls. That knowledge belongs in the scoring logic. The only way to get it there is to have the conversation before you build the model, not after.
The criteria were honest about what they could and couldn’t measure. Behavioral signals are useful but noisy. Fit signals like company size, industry, and job title are more reliable indicators that someone is the right type of buyer, even if they haven’t done much yet. The models I’ve trusted most treated fit and behavior as separate dimensions. They surfaced both rather than collapsing everything into a single number that hid what was actually driving it.
There was a feedback loop. Someone was looking at what happened to high-scoring leads after they reached Sales. Were they converting? Where were they dropping off? That data existed in Salesforce. It just needed someone to pull it and use it to update the model. In most teams, that wasn’t anyone’s explicit job. So it didn’t happen.
The harder conversation
Lead scoring only works if both sides agree to it. Marketing has to believe the model is worth maintaining. Sales has to work the leads it surfaces and say something when it’s wrong.
That second part is where it usually falls apart. Sales gets burned a few times and stops giving feedback. They route around the system quietly. Marketing keeps optimizing for MQL volume without knowing the quality signal has degraded. Both teams are busy. Nobody has the conversation.
The most useful thing I’ve found isn’t a better model. It’s getting both teams in a room regularly enough that the model stays honest. A monthly check. Looking at what converted and what didn’t. Adjusting together.
It sounds obvious. It’s surprisingly rare.
Lead scoring isn’t a set-it-and-forget-it system. It’s a living record of what Sales and Marketing currently agree a good lead looks like. When that agreement is current, it works. When it drifts, you end up back in that meeting. Looking at a spreadsheet. Trying to explain why a 90-point lead never picked up the phone.