What is lead scoring?

At its core, lead scoring translates two questions — "Is this the right company and person?" and "Are they showing buying interest?" — into a single number. Marketing and sales agree on which attributes and actions matter, and how many points each is worth. Those values are loaded into a CRM or marketing automation platform (MAP), which updates each lead's score automatically as new data arrives.

When a lead's cumulative score crosses the agreed MQL threshold, the system routes them to sales. Below the threshold, leads stay in a nurture track. The threshold itself should come from your own closed-won data — the score at which your sales team's close rate meaningfully exceeds your baseline — not from a vendor default.

In practice, high-performing teams layer at least two models: a fit model that scores firmographic and demographic attributes (title, industry, company size, geography), and an engagement model that scores behaviors (pricing page visits, demo requests, email clicks). Presenting both as a two-axis view — rather than collapsing them into one composite number — gives reps far more context than either dimension alone.

Rule-based scoring is the most common starting point. Marketers manually define criteria and point values, making the model transparent and easy to audit. The drawback is ongoing maintenance: rules drift out of sync with what sales actually closes, and the model can only capture signals the team thought to include upfront.

Predictive lead scoring uses machine learning — typically trained on historical won and lost opportunities in your CRM — to weigh features automatically and assign a conversion-probability score. A 2025 peer-reviewed study published in Frontiers in Artificial Intelligence tested 15 classification algorithms on 16,600 B2B CRM records; the Gradient Boosting Classifier achieved 98.39% accuracy (AUC 0.9891) and substantially outperformed the case company's existing manual model. Predictive scoring needs a meaningful volume of clean, closed-deal records to be reliable — most practitioners treat several hundred closed contacts as the working minimum before training a model.

A hybrid approach applies rules as hard gates (wrong industry or company size = automatic disqualify) and ML for behavioral intent signals, then combines both into a composite score. This is the practitioner standard for mid-market and enterprise B2B teams: rules handle the disqualifiers that do not require probabilistic reasoning, while the model handles the nuanced interplay of signals that humans cannot efficiently weigh.

The evidence is strong. Organizations with lead scoring frameworks report lead-to-opportunity conversion rates 77% higher and marketing-driven revenue 79% higher than peers without scoring, per data aggregated by Outfunnel from HubSpot research. Separately, companies implementing lead scoring achieve 138% ROI on lead generation versus 78% for those without it.

Sales-marketing alignment explains much of the lift. When both teams agree on what score constitutes an MQL, hand-off friction drops and reps stop treating the queue as unreliable. Industry benchmarks show MQL-to-SQL conversion rates ranging from 12–21% across B2B sectors; teams with structured scoring and fast follow-up consistently land at the top of that range or above it.

The key caveat: 61% of B2B marketers still send all leads to sales without scoring or qualification. Rule-based models can also produce false positives when fit alone is used without intent or engagement data — an account can look great on paper and be nowhere near a buying cycle. Intent data overlays and behavioral signals are what close that timing gap.

Rule-based scoring is explicit and human-authored: a marketer decides that a VP title is worth 15 points and a pricing-page visit is worth 20. It is easy to explain to sales, straightforward to audit, and can be running the same day. The cost is ongoing maintenance — as your ICP evolves and new channels emerge, the rules need manual updates, and they can only reflect patterns the team consciously anticipated.

Predictive scoring inverts this: a machine learning model ingests your historical CRM data — won, lost, and churned deals — and discovers which combinations of attributes and behaviors most strongly correlate with conversion. It updates as you close more deals and can surface non-obvious predictors — for instance, that leads from a specific referral source convert at 3x the baseline even at lower engagement scores, something a rule-based model would never capture unless someone noticed it manually.

The practical decision for most teams: start with rule-based scoring to create alignment and a working MQL threshold. Layer in predictive scoring once you have enough clean closed-deal history to train a reliable model. The two reinforce each other — rules handle hard disqualifiers that should never reach sales regardless of ML probability, while the model handles nuanced intent signals that rules cannot efficiently represent.

Lead scoring measures engagement intensity: how many high-value interactions has this lead taken, and how recently? It reflects implicit interest — inferred from behavior rather than stated directly.

Lead grading (used notably in Salesforce's Pardot / Account Engagement platform) measures profile fit: how closely does this company and person match your ideal customer? Grades are expressed as letters (A through F) and are based on explicit information — industry, job title, company size, geography — typically captured through form submissions or enrichment data.

A lead can score high (lots of activity) but grade low (wrong industry or seniority), or grade high (perfect ICP match) but score low (has never opened an email). The most actionable picture comes from a two-axis model that displays both: prioritize leads that are high-grade and high-score, nurture those that are high-grade but low-score, and deprioritize high-score but low-grade leads before they consume rep bandwidth that belongs elsewhere.

Lead scoring is the engine that powers the MQL/SQL funnel. A marketing-qualified lead (MQL) is a lead whose score has crossed the marketing team's agreed threshold — signaling that it is ready to route to sales. A sales-qualified lead (SQL) is a lead that sales has reviewed and accepted, confirming budget, authority, need, and timeline (BANT) in addition to the score.

The MQL threshold is where sales-marketing alignment lives or dies. If marketing sets it too low, sales gets flooded with poor-fit leads and stops trusting the queue. Too high and good leads stall in nurture and miss their buying window. Most B2B teams set MQL thresholds between 60 and 80 points; enterprise teams with longer cycles often push to 75–100. The right number comes from back-testing: find the score at which leads in your own CRM history closed at a rate worth a rep's time.

Lead scoring does not eliminate qualification — SQLs still require human judgment on timing, budget, and buying authority. What scoring does is make that qualification step faster and more consistent by pre-sorting the queue before a rep ever picks up the phone, so the human judgment gets applied to the right 20% of leads rather than the full 100%.

Signal-based selling depends on knowing not just who is a good fit, but who is a good fit right now. Lead scoring handles the "who" — it ranks accounts and contacts by ICP match and cumulative engagement. But the "right now" is where buying signals (job changes, funding rounds, intent spikes, hiring patterns) come in; they reveal that this particular week the timing is unusually good for a specific account.

Komo runs the plumbing that connects both layers. It monitors buying signals across your accounts, overlays them against each account's ICP fit score, and drafts the outreach when a high-scoring account also just crossed a signal threshold — so reps act on the intersection of fit and timing rather than on either dimension alone.

Because Komo keeps a human in the loop on every send that matters, the scoring logic feeds into a supervised workflow rather than autonomous bulk outreach. The result is the prioritization benefit of a well-tuned scoring model plus the relevance and timing advantage of signal-based selling — without the deliverability risk of firing at a scored list on autopilot.