Ask a sales team about lead scoring and you will often get a polite laugh. The model marketing built is somewhere in the CRM, generating numbers nobody looks at.
It does not have to be like this. Lead scoring fails for predictable reasons, and the fixes are more about design and trust than technology.
Why lead scoring models get ignored
- Too many signals. Points for every page view and email open produce high scores for curious students and competitors.
- Fit and interest mixed together. A perfect-fit company that has done nothing looks the same as a poor-fit company that visited ten pages.
- No sales input. Marketing designs the model alone; sales never agrees with the output.
- No decay. Someone who was active a year ago still looks hot today.
- Unreliable signals. Email opens, inflated by Apple Mail Privacy Protection (Litmus), still carry points.
Score fit and engagement separately
The most useful change is to score two dimensions.
Fit: “Should we want them?”
- Industry and sub-industry
- Company size and region
- Job role and seniority
- Technology or other qualifying attributes relevant to your offer
Engagement: “Are they interested now?”
- Demo or contact requests (the strongest signal)
- Pricing and product page visits
- Repeat visits within a short period
- High-intent content, such as comparison pages or case studies
- Email clicks, webinar attendance, replies
Plot the two together and the priorities become obvious: high fit and high engagement go to sales now; high fit with low engagement goes to nurture or ABM; low fit gets little attention regardless of activity.
Keep it simple
Start with a handful of signals for each dimension. You can always add more; it is much harder to remove signals once people rely on them. Every point value should be something you can explain to a sales rep in one sentence.
Build it with sales
- Review recent wins and losses together. What did good leads have in common before they converted?
- Agree definitions. What exactly is a marketing-qualified lead? What is sales accepted?
- Agree the handoff. Who follows up, how quickly and what happens if they do not?
- Agree a feedback loop. Sales flags leads that were not ready, so the model improves.
Salesforce found sales reps spend around 70% of their time on non-selling tasks (Salesforce). A scoring model that sends them fewer, better leads is one of the most practical ways marketing can help.
Add decay
Interest fades. Reduce engagement scores over time so that recent behaviour counts more than old behaviour. A pricing page visit this week should outweigh five blog visits last year.
Watch for buying groups
B2B purchases involve several people. 6sense’s research shows buying decisions are largely shaped before first contact with sellers (6sense). Scoring accounts, not just individuals, helps you spot when several people from the same company are researching at once, which is often the clearest sign of real intent.
Where AI helps
- Finding patterns in historical data that predict conversion.
- Enriching records so fit data is complete.
- Explaining scores in plain language: “high fit, recent pricing visits, three contacts active this week”.
- Monitoring accuracy, comparing predicted scores against actual outcomes.
Predictive scoring needs enough historical deals to learn from, and it must stay explainable. If sales cannot see why a lead scored highly, they will not trust it.
The takeaway
Good lead scoring is simple, transparent and co-owned with sales. Separate fit from engagement, use a few meaningful signals, decay old activity and review it regularly. AI can sharpen the model, but trust is what makes it useful.
We design and implement scoring and lifecycle automation as part of our email marketing work. For putting it into practice, see HubSpot workflows worth setting up first.
