Key takeaways
- Most models over-weight engagement (opens, clicks, page views) because it is the easiest data to collect, but engagement measures interest, not the ability to buy.
- Fit (company size, industry, role, ICP match) predicts revenue better than activity. The fix is to gate on fit first and treat engagement as a tiebreaker among fitting leads, not as additive points that let clicking compensate for poor fit.
- AI helps most on the fit and signal side (enrichment, inferring fit, real buying signals) and least when pointed at the same engagement metrics that were already misleading.
Ask a revenue team how their lead scoring works and you will usually hear a list of actions: opened three emails, clicked a link, visited the pricing page twice, downloaded a whitepaper. Each action adds points; cross a threshold and the lead is “sales-ready.” It feels rigorous, and it is easy to build, because every one of those actions is already sitting in the marketing automation platform waiting to be counted. It is also, in a lot of cases, measuring the wrong thing.
Interest is not the same as fit
The core problem is a confusion between two different questions. Engagement scoring answers “how interested is this person?” The question that actually predicts a sale is “is this person able and likely to buy?” Those are not the same, and conflating them is where most scoring models go wrong.
A lead can be extremely interested and completely unable to buy. A student writing a paper downloads your whitepaper and reads three blog posts. A competitor studies your pricing page every week. An individual with no budget and no authority signs up for everything because the content is genuinely useful to them. All three generate strong engagement scores. None of them will ever close. Meanwhile, the person who can actually sign a six-figure contract might read one page, forward it to their team, and go quiet, scoring almost nothing in a model that rewards clicks.
When a sales team complains that “the high-scoring leads don’t convert,” this is almost always why. The score was optimized for the signal that was easy to collect, not the signal that predicts revenue.
Fit, then engagement, in that order
The fix is not to throw out engagement. It is to put the two signals in the right relationship. Fit should come first, as a gate, not a bonus.
Fit is the set of attributes that describe who the lead is: company size, industry, revenue, the role and seniority of the person, whether the account matches the profile of customers who actually buy and stay. This is what the major scoring platforms call a fit score, kept distinct from an engagement score. Engagement is what they do. The failure mode in most scoring models is treating these as additive, a lead accumulates fit points and engagement points into one blended total, which means enough clicking can compensate for being a terrible fit. That is exactly backwards. A perfect-fit account that hasn’t engaged yet is a prospecting opportunity. A perfect-engagement lead with no fit is a time sink dressed up as a hot lead.
The two-by-two makes the point. A pure engagement score only reads the horizontal axis, so it cannot tell the priority account (top-right) apart from the trap (bottom-right): both look “hot.” Gating on fit first collapses the trap quadrant out of your “hot leads” pile, which is exactly the pile reps are spending their limited hours on.
Where AI actually helps, and where it doesn’t
Lead scoring is now a heavily marketed AI use case, and it is worth being precise about where the technology genuinely helps, because it is not evenly useful across the problem.
AI helps most on the fit and signal side. It is good at enriching thin lead records into a fuller firmographic picture, the job of a dedicated data-enrichment layer, inferring fit when you only have an email and a company name, and detecting real buying signals that a static points model never sees, an account hiring for a role that implies they need your product, a funding round, a leadership change, a shift in their technology stack. These are predictive of buying intent in a way that “opened an email” is not, and they are hard to capture without something doing the enrichment and pattern-matching for you. It is the same lesson that separates the AI SDR tools worth using from the ones selling hype: the constraint that matters is intelligence quality, not activity volume.
AI helps least when it is simply pointed at the same engagement metrics that were already misleading. A machine-learning model trained to predict “lead score” from click behavior will faithfully reproduce the original error with more decimal places. A more sophisticated model of the wrong signal is still the wrong signal. If the training target is “engagement,” the output is a better engagement predictor, not a better revenue predictor, and the gap between those two is the whole problem.
The takeaway
The reason so many scoring models disappoint is not that the math is wrong. It is that they answer “who is paying attention?” when the revenue team needed “who can actually buy?” Attention is easy to measure, which is exactly why it dominates models that were built from whatever data was closest to hand.
A scoring system worth trusting gates on fit first, treats engagement as a timing signal among already-qualified accounts rather than a way to earn your way in, and points its AI at the parts of the problem where better data genuinely changes the answer, enrichment and real buying signals, not a higher-resolution picture of who clicked. Get that ordering right and the score starts predicting revenue instead of enthusiasm. Get it wrong and you have built a very efficient machine for sending your best reps after people who were never going to buy.
The Counter Brief — one email, every Monday.
The week's AI-for-revenue moves in a 5-minute read: which tools are worth the budget and which to skip, plus what to do this week. Source-checked, no vendor decks.
Edited by Aditya Marin Gasga
Free. One click to unsubscribe.
Frequently asked questions
What is the difference between fit and engagement in lead scoring?
Fit is who the lead is: company size, industry, role, budget, whether they match your ideal customer profile. Engagement is what the lead does: opens emails, clicks links, visits pricing pages, downloads content. Fit predicts whether they can buy; engagement predicts whether they are paying attention. Both matter, but fit is the stronger predictor of a closed deal, and most models over-weight engagement because it is easier to measure.
Why do high-scoring leads often fail to close?
Usually because the score rewarded activity that does not indicate buying ability. A student researching for a paper, a competitor studying your product, or a curious individual with no budget can all rack up high engagement scores. If the model does not gate on fit first, it surfaces enthusiastic non-buyers and sends reps chasing them.
Can AI improve lead scoring?
Yes, but mostly on the fit and signal side rather than the activity side. AI is useful for enriching firmographic data, inferring fit from limited information, and detecting genuine buying signals (a company hiring for a relevant role, funding events, technology changes) that a simple points model misses. It is less useful when pointed at the same engagement metrics that were already misleading, because a better model of the wrong signal is still the wrong signal.
What should a lead score actually weight?
Fit first, as a gate: does this lead match who actually buys? Only among fitting leads does engagement become a useful tiebreaker for timing and priority. A common failure is treating fit and engagement as additive points in one total, which lets high engagement compensate for poor fit. Gating on fit first prevents that.