Section 1
What a score is actually for
A lead score exists to answer one operational question: when you have ninety minutes for pipeline today, who gets them? Without a score, that decision defaults to recency or loudness, whoever emailed last or pushed hardest, which systematically favors demanding poor fits over patient good ones. Goldratt's line is the warning label here: tell me how you measure me, and I will tell you how I will behave. If your de facto measure is 'most recent inquiry wins,' your pipeline behaves accordingly. A score replaces that accident with intention. It does not predict the future or replace judgment on calls; it sequences attention, ensuring the limited selling time Salesforce finds reps actually have, a minority of their week by its State of Sales research, lands on the leads most likely to become profitable clients. A useful companion to this piece is [Lead Qualification and Nurture: The System Between Attention and Revenue](/blog/lead-qualification-and-nurture-the-system-between-attention-and-revenue).
Section 2
The five signals worth scoring
Resist the urge to score everything trackable. Five signals, four from your intake form and one from behavior, carry nearly all the predictive weight for a service business. Fit signals come straight from your ICP: budget range, problem type, and decision role. Intent signals show motion: stated timing, meaningful page visits such as pricing or case studies, and engagement with your follow-up. The table below shows a complete starter model with points. Calibrate against history: score your last ten closed clients and your last ten dead deals retroactively, and check that the model separates them. If it does not, adjust weights, not by adding signals but by re-weighting the five. A model you can explain in one breath is a model your team will actually use.
Section 3
Thresholds: where the score becomes a decision
Points without thresholds are trivia. Define two lines and the actions they trigger, and the score becomes an operating rule. In the starter model above, a lead at seventy points or higher gets human outreach the same business day, ideally within the hour, since HBR's lead-response research showed contact attempts within an hour were roughly seven times more likely to qualify the lead than attempts made even sixty minutes later. Between forty and seventy, the lead enters your nurture sequence with a score-raising path: each meaningful action, a reply, a pricing visit, a webinar attendance, adds points until the threshold trips. Below forty, send the respectful decline or redirect. Write the thresholds where the whole team can see them, because a score that one person interprets privately is just intuition with decimals. The thinking here builds on [From Lead to Booked Call: Designing the Follow-Up System That Closes](/blog/from-lead-to-booked-call-follow-up-system).
Section 4
Keeping the model honest without a data team
Scoring models rot quietly. The fix is a thirty-minute monthly review, not a data scientist. Pull the month's closed-won and closed-lost deals and ask three questions. Did any low scorers close? Find the signal you missed and consider adding its points. Did any high scorers waste calls? Find the signal that inflated them and trim it. Are mid-band leads escaping nurture by raising their score, or is the middle a graveyard? If nothing has moved in two cycles, your nurture content is not generating scoreable behavior, which is a content problem, not a scoring problem. Keep version notes so you know which model produced which quarter's numbers. When clients install LeadOS, this review is a standing ritual on the calendar, and it is the difference between a model and a relic; a strategy call can show you the template. To see how this connects to the wider system, read [How Small Teams Achieve Big Results with AI](/blog/how-small-teams-achieve-big-results-with-ai).