How to score leads in CRM

Before you assign a single point, pull the last ninety days of leads and look at what you actually have. For each lead, check whether you captured job title, company size, industry, country, source, and at least one behavioral signal like a page view or form fill. If more than thirty percent of leads are missing any of these fields, scoring is premature. You will build a model that assigns zeros to half your pipeline and calls it low intent, when the truth is you never collected the signal. Fix the data collection problem first: shorten forms where possible, add progressive profiling, enrich from a third-party source at creation, and require a minimum field set before a lead can enter the scoring queue. Spend a week here and the model you build on top will survive its first quarter. Skip this step and you will spend the next three months explaining why reps keep calling leads the model flagged as hot and finding nobody home.

• Pull a ninety-day sample of leads and audit completeness of title, company size, industry, country, source, and one behavioral signal
• Flag fields with more than thirty percent missing data; fix collection before building the model
• Add firmographic enrichment at lead creation so company size and industry are never blank
• Define a minimum required field set that must be true before a lead enters the scoring queue

Every scoring signal falls into one of three buckets, and mixing them in a single bag is the fastest way to build a model nobody trusts. Demographic signals describe the person: job title, seniority, department, and function. Behavioral signals describe what they have done: visited the pricing page, requested a demo, opened five emails in two weeks, downloaded a comparison guide. Firmographic signals describe the company: employee count, revenue band, industry, geography, tech stack, funding stage. Each bucket predicts something different. Demographic answers the question of whether this person can say yes. Behavioral answers whether they are leaning in right now. Firmographic answers whether the company is a plausible customer at all. A lead who scores high on all three is a hot handoff. A lead who scores high only on behavior but low on firmographics is probably a competitor researching you or a student writing a paper. Keep the buckets visible on the lead record so reps can see at a glance which dimension drove the score, not just the total.

• Create three parent scoring fields on the lead record: demographic score, behavioral score, firmographic score
• List every candidate signal you currently capture and assign it to exactly one bucket
• Store the three sub-scores alongside the composite total so reps can diagnose any lead at a glance
• Reject any signal that could belong to two buckets; pick one and move on

Demographic scoring is where most teams overcomplicate things. You do not need twenty title variants. You need a short table that reflects how much authority this person has to influence or approve a purchase. Give the most points to titles that match your typical economic buyer: VP, Director, and Head-of titles in the function you sell to. Give moderate points to senior managers and specialists in that function because they often champion the deal internally. Give small positive points to adjacent roles like RevOps or Enablement when you sell to Sales, since they often participate in the evaluation. Give zero or small negative points to students, interns, and consultants, who rarely buy and consume disproportionate sales time. Keep the weights in a defined range such as zero to twenty-five so no single signal can dominate the composite. Store the title-to-points mapping in a reference table inside the CRM, not in a rep training deck that nobody reads. When you add a new title variant to the ICP, you update one place and the model follows.

Behavioral weights are the most dynamic part of the model and the most dangerous to overfit. Start with a short list of high-signal actions and assign them points proportional to how often they appear in closed-won journeys. The pricing page view, the demo request, the comparison download, and the hands-on trial signup usually sit at the top. Email opens, blog post reads, and generic resource downloads sit at the bottom. Resist the urge to score every click. If a signal does not meaningfully distinguish a buyer from a researcher, it should carry zero weight. Give especially high weight to actions that require explicit intent like booking a meeting, replying to a sales email, or starting a free trial. Give negative weight, or at minimum zero weight, to signals that look active but mean nothing in isolation: a single webinar registration, a bounced email, an unsubscribe. Keep the behavioral weight range two to three times wider than demographic because behavior changes week to week and demographic facts do not.

• Pull the behavioral journey of your last twenty closed-won deals and rank the actions by frequency
• Assign the top three actions high weight, the middle three moderate weight, and everything else zero
• Weight explicit intent signals like demo requests and reply-to-email higher than passive signals like page views
• Set a wider weight range for behavior than demographic to let real-time engagement move the score

Firmographic weighting is the step where most sales teams get sloppy and most marketing teams get precious. The job is simple: translate your ICP into a scorecard that gives the most points to companies that look exactly like your best customers. If your sweet spot is one hundred to five hundred employees in SaaS and professional services based in North America, that combination earns the maximum score. Companies within one standard deviation of that profile earn moderate points. Companies well outside the sweet spot but still plausible earn small points. Companies that fail an explicit disqualifier such as being in a regulated industry you cannot sell to, or being in a country you cannot invoice, earn zero and skip the pipeline entirely. Enrichment is the oxygen for this bucket. If you do not know a company size or industry at lead creation, the firmographic score is a guess, and the composite is poisoned. Automate enrichment at creation and nightly thereafter so firmographic facts stay fresh even when the person on the record changes jobs.

• Translate your ICP into a point table across employee count, revenue, industry, geography, and tech stack
• Assign zero to any lead that fails an explicit disqualifier so the model never surfaces it as hot
• Automate firmographic enrichment at lead creation and nightly thereafter
• Keep the firmographic weight range the widest of the three buckets since ICP fit matters most for long-term win rate

A lead who downloaded a buyers guide last Tuesday and a lead who downloaded the same guide fourteen months ago should not share a score. Time decay is the mechanism that drains behavioral points over time, and it is the single most overlooked lever in CRM scoring. Pick a half life that matches your sales cycle. For a thirty-day cycle, behavioral points should lose half their value in two to four weeks. For a six-month enterprise cycle, four to eight weeks is reasonable. Demographic and firmographic points do not decay because they describe stable facts: a VP of Sales is still a VP of Sales six months later, and a five-hundred-person SaaS company is still a five-hundred-person SaaS company unless it merges. Behavioral points must decay because intent is temporal. Implement decay with a recurring job that recalculates behavioral scores daily or weekly rather than letting a one-time point bump persist for a year. Reps who see a score climbing on recent behavior trust the model. Reps who see the same score on a cold lead and a hot lead learn to ignore it within a week.

The threshold is where theory meets routing. Pull the composite scores of leads that converted to opportunities over the last ninety days and plot the distribution. The point where a large majority of future opportunities land is your floor. Set the Marketing Qualified Lead threshold at that floor, publish it, and route anything at or above it to a dedicated inbound SDR queue with a service-level agreement attached. The SLA is the second half of the threshold decision. A lead that crosses the bar at four PM on a Friday and gets called at ten AM on Monday is not a hot lead by Monday, it is a cold lead with a decayed behavioral score. Industry research consistently shows that reaching out within the first five minutes dramatically improves conversion compared to even a one-hour delay. Pick a response SLA that matches your reality, instrument it with a timer on the CRM record, and alert managers when a lead ages past it. The threshold without an SLA is a vanity number. The SLA without a threshold is a workflow with no trigger. Both together are what makes scoring operational.

• Plot composite scores of recent converted leads and identify the floor the majority cleared
• Set the Marketing Qualified Lead threshold at that floor and publish it to sales and marketing
• Route anything at or above the threshold to a dedicated SDR queue, not the general inbound bucket
• Attach a response SLA to the queue and instrument a timer that alerts on breach

A score crossing a threshold is useless if the rep learns about it from a Monday morning report. The hot handoff must be instant, visible, and carry enough context for the rep to act in a single click. When a lead crosses the Sales Qualified Lead threshold, which is usually one tier above MQL and often requires a direct intent signal like a demo request, fire three actions at once. First, assign the lead to the correct owner using round-robin or territory logic, not by letting whoever opens the queue first grab it. Second, send that owner an immediate notification through the channel they actually watch, which for most teams is a Slack direct message and a mobile push, not an email. Third, surface a one-screen briefing on the lead record showing why they crossed the threshold: the three highest-weight signals they fired, their firmographic match, and any contextual notes like the content they consumed. The briefing is the difference between a cold-sounding call and a credible one. Reps who can open the record, read the briefing, and dial in under thirty seconds are the reps who close the leads scoring surfaced.

• Define the Sales Qualified Lead threshold one tier above MQL, usually requiring an explicit intent signal
• Automate owner assignment via round-robin or territory logic the moment the threshold trips
• Fire a real-time Slack or push notification to the assigned owner, not an email digest
• Surface a one-screen briefing on the record showing the top signals, firmographic match, and recent context

Lead scoring is not a set-and-forget artifact. Buyer behavior shifts, product changes, segments mature, and the signals that predicted a win last year drift. Every quarter, pull the leads that converted to closed-won in the last ninety days and the leads that scored above threshold but closed-lost or never converted. Compare the two populations across every signal in the model. Signals that appear strongly in winners and weakly in losers deserve more weight. Signals that appear equally in both deserve less weight or should be dropped. Look specifically for new signals that emerged in the data since you last calibrated, like a feature-page view that did not exist six months ago or a competitive-displacement search term you did not track. Also look for signals that have gone dead, like a webinar series you stopped running. Recalibration is a half-day exercise with the RevOps lead, the marketing operations lead, and one senior AE in the room. Ship the new weights in a staging environment, run them against last quarter in parallel, and only promote to production when the model would have surfaced the right winners within the expected volume. Document the delta in a short changelog so sales knows the model learned, which maintains trust.

• Pull closed-won and above-threshold-non-converted leads from the last ninety days
• Compare both populations across every signal; keep what differentiates, drop what does not
• Add new signals that emerged and retire signals that have gone dead
• Test new weights against last quarter in parallel before promoting to production
• Publish a changelog so sales sees the model learned, not changed arbitrarily