Lead Scoring Sales Actually Trusts

Marketing Automation

Almost every lead scoring model I have ever inherited from a client was built with good intentions and quietly ignored within a quarter. Marketing sets it up, feels clever about it, and moves on. Sales glances at the numbers, notices that a “hot” lead is a student doing research and a “cold” lead is the head of procurement at a company that just signed, and stops trusting the whole thing. From that point the model is dead. It still runs, it still assigns points, but nobody makes a decision because of it. At Identiti Design we have spent enough time cleaning up these graveyards to have a firm opinion: a modest scoring model that sales actually believes in will beat a sophisticated one they roll their eyes at, every single time. This piece is about how to build the first kind and avoid the second.

Why Most Lead Scoring Quietly Dies

The failure is rarely technical. The CRM did its job. The points added up correctly. The issue is that the model was built as a marketing artefact rather than a shared operating agreement between two teams who have to live with the results. Marketing wants to prove that its campaigns produce qualified pipeline, so it rewards the behaviour it can easily measure: email opens, page views, form fills, webinar sign-ups. Sales wants to know one thing, which is whether this person can and will buy. Those two goals overlap far less than people assume, and when the score reflects marketing’s goal instead of the buying reality, sales learns to distrust it fast.

The second reason models die is that the points are arbitrary. Someone decided an ebook download is worth ten points and a pricing page visit is worth fifteen, and nobody can explain why fifteen and not five, or why an ebook is worth anything at all. When you ask where those numbers came from, the honest answer is usually “they felt about right.” A model built on gut feelings dressed up as arithmetic has no defence the first time it produces an obviously wrong result, and it always produces an obviously wrong result eventually. Sales does not need the model to be perfect. They need it to be explainable, and arbitrary points cannot be explained.

The third reason is the absence of a feedback loop. Most models are set up once and never revisited. Deals close, deals fall through, and none of that information ever flows back to adjust the scoring. So the model keeps rewarding the same signals even after the sales team has learnt, from painful experience, that those signals mean nothing. A scoring model that never learns from outcomes is a guess frozen in time, and the market moves faster than the guess.

Underneath all three of these is a single cultural failure: the model was owned by one team instead of shared by two. Lead scoring is not a marketing deliverable, it is a handover protocol. It exists precisely at the seam where marketing passes a person to sales, and a protocol that only one side helped write is not a protocol at all, it is an instruction the other side is free to ignore. Every time I have seen a model survive past its first quarter, it was because both teams felt ownership of it. Every time I have seen one die, it was because marketing built it alone and expected sales to simply comply. The technology is almost never the problem. The relationship is.

The Vanity Signal Trap

The single most common mistake I see is a model stuffed with vanity signals. These are the actions that are easy to track and feel like engagement but say almost nothing about intent to buy. Opening an email is the classic one. In an era of automatic image loading and preview panes, an open barely proves a human saw the subject line. Rewarding it with points inflates scores for people who are politely ignoring you while doing nothing for the person who read one carefully chosen page and picked up the phone.

The same trap applies to raw form fills. A form submission feels like a commitment, but half the time it is someone grabbing a gated PDF with a throwaway email address, and treating that identically to a demo request tells sales nothing useful. What matters is not that a form was filled but which form, by whom, and what they did next.

Vanity signals are seductive because they produce big, satisfying numbers. A campaign that generates five thousand opens looks like it is working, and it is tempting to build a model that celebrates that volume. But the score is supposed to help a salesperson decide who to call next, not to make the marketing dashboard look busy. When you reward shallow engagement, you flood the top of the leaderboard with people who will never buy, and you bury the quiet, serious buyer who does not need to click through six nurture emails before making a decision. Sales notices this within days, and it is often the first crack that breaks their trust.

The fix is to be ruthless about which behaviours earn points. I ask one question of every signal: does a genuine buyer do more of this than a tyre-kicker? If the answer is no, or if I cannot tell, the signal earns nothing. Repeated visits to a pricing page, viewing a case study relevant to the buyer’s own industry, requesting a demo, returning to the site after weeks of silence, these are the actions that separate intent from idle curiosity. I have written before about how the small, early actions that reliably precede a purchase are worth far more than the loud, hollow ones, and that thinking sits at the centre of good scoring. If you want to go deeper on identifying those early tells, our piece on micro conversions as leading indicators covers the ground properly.

Fit And Behaviour Belong On Separate Axes

Here is the structural mistake that quietly poisons more models than any other: collapsing fit and behaviour into a single number. Fit is who the person is. Are they the right role, at the right kind of company, in the right region, with the right budget reality? Behaviour is what they are doing. Are they showing signs of active interest right now? These two things are completely different, and when you add them into one score you get nonsense.

Picture two leads who both score eighty. The first is a perfect-fit head of operations at your ideal customer profile who has visited twice. The second is a freelancer who has consumed every piece of content you have ever published but will never have budget or authority to buy. Same score, wildly different value. A salesperson who chases the second lead because the number told them to will lose faith the moment they get on the call, and they will be right to. One number cannot carry both meanings without lying about at least one of them.

The answer is to score fit and behaviour on separate axes and keep them separate all the way through. I usually express fit as a simple grade, A through D, and behaviour as a temperature, hot through cold. Now a lead is not “eighty,” it is “A and warming” or “C but very active.” That two-part label tells a salesperson exactly what they are looking at. The A leads who are heating up are the ones you drop everything for. The A leads who are cold are worth a patient, human nurture. The C and D leads who are on fire are a signal to marketing that its content is attracting the wrong audience, which is useful information rather than a false alarm. Separating the axes turns the score from a single misleading verdict into an actual diagnosis.

Build A Model Sales Can See And Argue With

Transparency is the whole game. If a salesperson cannot look at a score and understand, in under a minute, why this lead is rated the way it is, they will not trust it, and they should not. The instinct among marketers is to make the model clever, with dozens of weighted signals and decay curves and conditional logic. I push hard in the opposite direction. The model should be simple enough to explain on a whiteboard, and every point should trace back to a rule sales agreed to.

The way to earn that trust is to build the model with sales in the room, not to present a finished model to them for approval. Those are completely different exercises and they produce completely different levels of buy-in. Sit down with your two or three best closers and ask them what a good lead looks like. They already have a scoring model in their heads, refined over hundreds of calls. Your job is to write it down and formalise it, not to overrule it with marketing’s theory. When a rep can point at a rule and say “yes, I told you that job title matters,” the model stops being marketing’s toy and becomes a tool they helped make.

Then make the scores visible and challengeable inside the CRM itself. Any modern CRM can show the score breakdown on the lead record, so a rep sees not just “A, hot” but the specific reasons: right role, viewed pricing twice, requested a demo, based in a target market. And give them a channel to push back. When a rep flags a lead that scored high but was obviously junk, that is not a complaint to manage away, it is the most valuable calibration data you will ever get. A model that sales is allowed to argue with is a model sales will use. A black box they are told to trust is a model they will quietly route around.

Negative Scoring And Decay Keep The Model Honest

Most models only ever add points, and that is why they slowly fill up with stale, misleading scores. A serious model has to subtract too. Negative scoring is how you stop the wrong people from floating to the top. A personal email domain instead of a company one, a job title that clearly is not a buyer, a company far too small or in a country you do not sell to, a visit to your careers page rather than your product pages: these should pull the score down, because they are genuine evidence that this is not your buyer. Subtracting points feels uncomfortable to marketers who are measured on lead volume, but it is exactly what makes the surviving high scores meaningful.

Decay is the other half of keeping the model honest. Intent is perishable. Someone who visited your pricing page three times last week is hot. Someone who did the same thing four months ago and has gone silent since is not hot, they are history. If behaviour points never decay, your hot list gradually fills with the ghosts of interest long past, and a salesperson who works that list wastes their day on people who have moved on or already bought elsewhere. Behaviour scores should fade over time so that the temperature reflects current reality, not an accumulated record of everything the person has ever done.

The important nuance is that fit and behaviour decay differently, which is another reason to keep them apart. Fit barely changes. A head of procurement at your ideal customer is still a great fit in six months even if they have gone quiet. Behaviour, by contrast, should cool steadily the moment activity stops. When you apply decay only to the behaviour axis and leave fit largely stable, the model naturally surfaces the right follow-up: strong-fit leads who have cooled are perfect for a patient re-engagement, and getting that re-engagement right without becoming a nuisance is a craft of its own, one I have written about in our guide to lead nurturing without spam.

Close The Loop With Won And Lost Outcomes

This is the step that separates a model that improves from a model that slowly rots, and it is the step almost everyone skips. The outcomes of your deals, closed-won and closed-lost, are the only real answer key you will ever get. Every closed deal is the market telling you whether your score was right. If you never feed that answer back into the model, you are studying for an exam and never checking your answers against the marks.

The mechanics are not complicated. When a deal closes, look at what the lead scored before sales engaged, and look at which signals it had accumulated. Do this across enough closed-won deals and clear patterns emerge. You may find that the buyers who actually convert almost always viewed a specific case study, or that a particular job title closes at twice the rate of the one you were weighting more heavily. You will also find signals you were rewarding that show up just as often in closed-lost as in closed-won, which means they have no predictive value and should be stripped out. The closed-lost data is as valuable as the wins, because it tells you which of your “hot” signals were noise all along.

There is a subtler payoff to closing the loop, and it is about trust rather than accuracy. When sales can see that the model changed because of the deals they closed, the whole thing stops feeling like something imposed on them and starts feeling like something that listens. A rep who flags a signal as worthless in July and sees it removed in the October calibration will believe in the model in a way no launch presentation could ever achieve. The feedback loop is not only how the model gets smarter, it is how the model earns and keeps its credibility with the people who have to act on it. That is why I treat the quarterly review as non-negotiable even when everyone is busy. Skip it twice and the model is back to being a static guess, and sales can smell a static guess.

I recommend a standing calibration rhythm, a short session every quarter where marketing and sales look at the closed deals together and adjust the model. Not a rebuild, just a tune: raise the weight on signals that predicted wins, cut the ones that did not, adjust the fit grades if the ideal customer has shifted. This is also where trustworthy data earns its keep, because a feedback loop is only as good as the tracking underneath it. If your conversion data is unreliable, your recalibration will be too, which is why we treat clean measurement as a prerequisite and have written a full piece on building trustworthy conversion tracking before you lean on it for decisions. A model with a working feedback loop gets sharper every quarter. A model without one is at its best on the day you launch it and gets worse from there.

A Worked Example You Can Copy

Let me make this concrete with a model close to ones we have actually shipped for B2B service clients. Start with the fit axis, graded A to D. A company in a target industry and size band earns strong marks; a decision-making or influencing job title adds more; a target region adds a little; a personal email domain or an obviously junk title subtracts. Add it up and bucket it: A is a near-perfect match, D is someone you should probably not spend a call on regardless of how active they are. This axis moves slowly and rarely needs decay.

Now the behaviour axis, expressed as a temperature. A demo request is the strongest single signal and might be worth thirty points on its own. Two or more pricing page visits in a fortnight, twenty. Viewing an industry-specific case study, fifteen. Returning to the site after a month of silence, ten, because a comeback often signals a project restarting. Meanwhile the vanity signals earn little or nothing: an email open is worth zero, a single blog visit maybe two. Every behaviour point decays, halving over roughly thirty days, so the temperature always reflects recent intent. Map the total to hot, warm, or cold.

Put the two together and the label does the work. An A that is hot goes to a salesperson today with the full breakdown attached. An A that has gone cold goes into a considered nurture track segmented to their profile, which is exactly the kind of situation where getting your email segmentation right pays off, because the message to a cooled A-grade buyer is nothing like the message to a fresh one. A C that is hot triggers a quiet conversation with marketing about why the content is pulling the wrong crowd. And crucially, when any of these leads closes, its pre-engagement label and signals get logged so the next quarter’s calibration has real data to work with. Notice how little cleverness is in this model. That is deliberate. It is simple enough that a rep understands it instantly and simple enough that you can actually maintain the feedback loop, which matters far more than sophistication.

Start Simple, Earn The Right To Add Complexity

The strongest temptation, especially for anyone who enjoys building systems, is to over-engineer the first version. Resist it completely. A model with two clear axes and a dozen honest signals that sales trusts and uses is worth more than a forty-signal masterpiece that nobody looks at. Complexity is not a sign of maturity in lead scoring, it is usually a sign that nobody had the discipline to decide what actually matters. You earn the right to add a signal by proving, through the feedback loop, that it predicts outcomes. Until then it is just noise wearing a lab coat.

There is a sequencing point here too. Scoring sits on top of your funnel, and if the funnel underneath is leaking, a beautiful score just helps you rank leads who were going to bounce off a broken experience anyway. Before I invest heavily in a scoring model I want the basics working: a site that does not fight the visitor, forms that do not repel them, a path to conversion that is not full of friction. If you have not looked hard at that layer, a conversion friction audit is usually a better use of the first week than tuning point values, because it fixes the ground the scoring stands on.

To close, here is the short checklist I run through whenever we build one of these. Score fit and behaviour on separate axes and never collapse them into a single number. Build the model with your best salespeople in the room, not for them in a slide deck. Reward only signals that a genuine buyer does more than a tyre-kicker, and pay nothing for vanity metrics. Use negative scoring to push the wrong people down, and decay behaviour so the temperature stays current. Make every score visible and challengeable inside the CRM, and treat a rep’s pushback as calibration data rather than complaint. Feed closed-won and closed-lost back into the model on a quarterly rhythm so it learns. And start simple, adding a signal only once outcomes prove it earns its place. Do those things and you will have the only kind of lead scoring worth having: the kind sales actually trusts.