Score with points, decide with thresholds
A lead-scoring model turns "they feel serious" into a number you can route on, and the threshold, not the score, is the decision.
Once you know what fit and intent look like, you need a way to apply that judgement at scale without thinking about it each time. That is what scoring does. You assign points to the signals that predict a good customer, add them up automatically, and let a threshold decide what happens next. The point of scoring is not precision to the decimal; it is consistency. The same lead gets the same verdict whether it arrives on a Monday or while you are on holiday.
What a scoring model is
A lead-scoring model is a set of point values attached to signals, summing to a number that ranks each lead by how worth-your-time it is.
You score two planes separately. Fit points come from firmographics: company size in range, right industry, decision-maker job title, a use case you serve. Intent points come from behaviour: visited pricing, requested a demo, opened the last three emails, attended a webinar. Keeping the two planes separate matters, because a lead can be high-fit and low-intent (nurture) or high-intent and low-fit (disqualify), and a single blended number hides that difference.
Why thresholds do the real work
The score is just a position on a scale; the threshold is where you draw the line that triggers an action. Without thresholds, a score is a vanity number nobody acts on. With them, scoring becomes routing.
Set at least two lines. The MQL line is where marketing hands a lead to sales: enough fit and enough intent to be worth a human's attention. The SQL line is where that lead is qualified for a real sales conversation, usually fit-confirmed plus a buying trigger. Below the MQL line, the lead stays in automated nurture and never touches your calendar.
How to build the model
- List every signal you can capture, split into fit and intent columns.
- Assign points by predictive weight, not by feel. A signal that strongly predicts a closed deal (decision-maker job title, demo requested) gets more points than a weak one (opened an email). Negative points are valid: a free-email-domain or a student job title can subtract.
- Set the MQL threshold from your best customers. Score your last ten closed-won deals retrospectively. Where did they sit? Put the MQL line just below that band so real buyers clear it.
- Set the SQL threshold as fit-confirmed plus a trigger. Reaching MQL gets attention; reaching SQL gets a calendar invite.
- Automate the sort in your CRM. A scoring model that lives in a spreadsheet you update by hand will not survive a busy week. It has to run on every lead, every time, untouched.
- Review the thresholds monthly against what actually closed, and move the lines.
A worked example
A 22-person B2B services firm was scoring leads in a shared spreadsheet, which meant scoring happened only when the founder remembered. They rebuilt the model inside their CRM with HubSpot, splitting fit points (company-size band, industry, job title) from intent points (pricing visits, demo requests, email engagement), then set an MQL line calibrated to where their last twelve closed deals had sat and an SQL line that required a confirmed budget conversation. Sales stopped chasing every form fill; the founder's call list became the leads above the SQL line only. Time spent on unqualified demos dropped by about a third, and the close rate on the leads that did reach a call climbed because the floor was now real. Used here, the CRM is the scoring engine inside a qualification system, not the topic itself.
Pitfalls
- Scoring vanity signals. A blog-post view or a careers-page visit is noise. Only score signals that correlate with deals you have actually won.
- Blending fit and intent into one number. It hides the high-intent poor-fit trap. Keep two scores or at least two dimensions.
- A threshold you never move. Markets and your ICP shift. A static line slowly drifts out of truth; review it monthly.
- Manual scoring. If a human has to run it, it will not run. Automate or do not bother.
With a model that scores and thresholds that route, you have a working engine. But scoring tells you who and when, not how to run the conversation. Next we choose the qualification framework that shapes the call itself.