B2B Lead Scoring: How the Model Works, Thresholds, and MQL/SQL Handoff

79% of marketing-generated leads never convert into sales (MarketingSherpa). A big reason is trust: reps stop working lists labeled "qualified" after enough of them turn out to be students, competitors, and people who downloaded one whitepaper. Lead scoring is supposed to fix that. Most models make it worse by scoring activity instead of intent.

Here's how the scoring model actually works, where to set your thresholds, and the handoff step most teams skip.

---

TL;DR

• B2B lead scoring ranks leads on two axes: fit (who they are) and behavior (what they've done).
• Thresholds aren't industry averages — set them where your own closed-won deals cluster.
• An MQL crosses a score; an SQL is a lead sales has explicitly accepted. The acceptance is the handoff.
• Negative scoring and decay keep stale or bad-fit leads from inflating the pipeline.
• A score is only as good as its inputs — scoring real buying signals beats scoring email opens.

---

What Is B2B Lead Scoring?

B2B lead scoring is a system that assigns point values to each lead so sales can work the highest-probability accounts first. It ranks leads on two dimensions: fit — how closely the lead matches your ideal customer profile — and behavior — how strongly they're signaling active interest. When both are high, the lead is worth a call now. When either is low, the lead stays in nurture until the signals change.

Scoring ranks likelihood to convert. It doesn't create demand, and it doesn't replace a sales conversation. Its one job is to stop reps from guessing which leads to call — turning a volume game into a prioritized queue. The failure mode, and the reason so many models get quietly ignored, is confusing activity with intent: a prospect who downloads five ebooks may be a researcher with no budget, while one who visits your pricing page once may be a VP ready to buy.

---

The Two Scoring Dimensions

Every score combines two independent inputs. Score them separately, then add them — a high-fit lead with no engagement and a low-fit lead with lots of engagement are different problems, and blending them into one number hides that.

Fit scoring is mostly static. It answers "does this lead look like a customer?" using firmographic data (company size, industry, revenue, tech stack) and demographic data (job title, seniority, decision authority). A VP of Sales at a 500-person SaaS company scores far higher than a coordinator at a company outside your ICP.

Behavioral scoring is dynamic. It answers "is this lead buying now?" by weighting actions according to how close they sit to a purchase decision.

Signal	Typical points	Why
Demo request	+15 to +20	Highest-intent action a lead can take
Pricing page visit	+10 to +15	Evaluating cost = active consideration
Webinar attendance	+10	Invested time, not just a click
Content download	+5	Interest, but often top-of-funnel
Email click	+2 to +3	Weak signal on its own
Return visit	+2	Reinforces other signals

The exact numbers matter less than the ranking: intent-proximate actions must outweigh passive ones, or the model rewards busywork.

---

Setting MQL and SQL Thresholds

A threshold is the score at which a lead changes status. The single most common mistake is copying an "industry average" threshold. Thresholds aren't benchmarks — they're the score where your sales team should start working a lead, and that's a number only your own data can tell you.

Set it by scoring your last 12–24 months of closed-won and closed-lost deals, then finding the score above which conversion rates rise materially. That inflection point is your MQL line.

Band	Typical range (100-pt scale)	What it should trigger
Warm	40–64	Stays in nurture; not sales-ready
MQL	65–79	Routed to sales for review
SQL trigger	80+	Immediate follow-up, real-time alert

Treat the ranges above as a starting point to calibrate against, not a prescription. Run the new model against historical deals before going live: if your highest-scoring leads don't correspond to your actual closed-won deals, the weights are wrong and no threshold will save the model.

---

The MQL → SQL Handoff

The SQL threshold is not just "a higher score." An SQL is a different event: a salesperson reviews the lead and explicitly accepts it as worth pursuing. That acceptance — a deliberate action, not an automatic status flip — is what creates accountability on both sides. Marketing owns the MQL; sales owns the SQL.

Three things make the handoff work:

• Speed. Route MQLs to a named rep within minutes, not hours. The odds of qualifying a lead drop sharply once response time passes the first few minutes.
• An SLA. Give the rep a defined window — commonly 24 hours for high-intent signals, 48–72 for warmer ones — to accept or reject, enforced through CRM workflow rather than left to individual judgment.
• A feedback loop. Every rejection should capture a structured reason (wrong fit, bad timing, already a customer, competitor). That feedback is the single most valuable input for recalibrating the model — it's how scoring stops drifting.

Without these, the handoff fails in the same predictable way: marketing fires the alert, sales ignores it, the lead goes cold, and both teams blame the other.

---

Negative Scoring and Decay

Positive points alone inflate every list. Negative scoring and decay are what keep a score honest.

Attribute	Typical adjustment
Competitor email domain	−30
Personal / free email (gmail, yahoo) for a B2B product	−20
Careers-page visit (job seeker, not buyer)	−10
Unsubscribe	−10
Invalid or bounced email	Disqualify
Inactivity 30–60 days	Decay −10

Decay matters as much as the deductions. A lead that was hot in March isn't hot in June without renewed engagement — without decay, your MQL list slowly fills with stale contacts that drag down conversion and waste follow-up time. (This is also where list hygiene overlaps scoring: an invalid address should never accrue a score at all, which is why email verification sits upstream of the model.)

---

How to Build the Model

Building a working score is six concrete steps, in order:

1. Validate what predicts close rate. Analyze closed-won vs. closed-lost from the last 12–24 months. The attributes your gut says matter and the ones that actually correlate with revenue are usually different lists.
2. Pick a scale. A 100-point scale is standard and easy to reason about.
3. Assign points by bucket. Allocate across fit and behavior, reserving a band for negative signals.
4. Set MQL and SQL thresholds from the historical inflection point, not a round number.
5. Backtest before going live. Score last quarter's leads and confirm the ranking predicts close rate.
6. Review quarterly. Recalibrate when MQL-to-SQL conversion drops, rejection reasons cluster, or your ICP shifts.

---

Where Most Scoring Models Break: Activity Isn't Intent

A score is only as predictive as the signals feeding it. This is the part most guides skip, and it's where the majority of models quietly fail.

Score email opens and content downloads, and you're measuring engagement with your marketing — which correlates weakly with readiness to buy. The signals that actually predict a purchase happen outside your funnel: a company raising a funding round, hiring a new VP into a department you sell to, adopting a technology that pairs with yours, or posting ten roles that signal a budgeted initiative. A model that can't see those is scoring noise with a point total attached.

This is the difference between a lead that engaged with you and a lead that's entering a buying window. Flinter, an AI-native cold email personalization platform, is built around the second one — it identifies real company and contact signals and generates outreach from them, so the same event that should raise a lead's score is the one worth reaching out about. Scoring tells you which leads to work; the signal tells you why now, which is what makes the outreach land. (For the catalog of signals worth scoring and acting on, see cold email buying signals and examples.)

---

Score the Right Things, Then Act Fast

A lead scoring model isn't a set-and-forget project — it's a system that's only as good as its inputs and its handoff. Build it from your own closed-won data, weight intent over activity, add negative scoring so stale leads don't inflate the queue, and route the top of that queue to sales the moment it crosses the line.

The teams that win outbound in 2026 aren't scoring more activity. They're scoring the signals that mean a company is about to buy — and reaching out first. Book a 30-minute walkthrough to see how signal-based personalization turns a scored lead into a booked meeting.