Use this skill when defining ideal customer profiles, building scoring models, identifying intent signals, or qualifying leads. Triggers on lead scoring, ICP definition, scoring models, intent signals, MQL, SQL, lead qualification, BANT, and any task requiring lead prioritization or qualification framework design.
Use this skill when defining ideal customer profiles, building scoring models, identifying intent signals, or qualifying leads. Triggers on lead scoring, ICP definition, scoring models, intent signals, MQL, SQL, lead qualification, BANT, and any task requiring lead prioritization or qualification framework design.
When this skill is activated, always start your first response with the 🧢 emoji.
Lead Scoring
Lead scoring is the discipline of quantifying how likely a prospect is to become a
paying customer so sales teams spend time on the right people. A good scoring model
combines profile fit (does the company match your ICP?) with behavioral intent (are
they actively signaling purchase readiness?). This skill equips an agent to define
ICPs, build point-based or predictive scoring models, weight intent signals, set
MQL/SQL thresholds, implement score decay, and create a shared sales-marketing
framework that drives consistent, measurable pipeline qualification.
When to use this skill
Trigger this skill when the user:
Needs to define or refine an Ideal Customer Profile (ICP)
Wants to build or overhaul a lead scoring model with point values
Asks how to identify, classify, or weight intent signals (first-party or third-party)
Needs to set MQL, SQL, or PQL thresholds and handoff criteria
Wants to implement score decay for aging or disengaged leads
Asks about BANT, MEDDIC, CHAMP, or any qualification framework
Needs to validate whether a scoring model is actually predicting conversions
Wants to align sales and marketing on lead definitions and SLA terms
Do NOT trigger this skill for:
CRM technical implementation or integration wiring - use a CRM/engineering skill
General demand generation strategy unrelated to lead qualification
Key principles
Fit + intent = score - Profile fit answers "should we ever sell to this company?"
Intent answers "should we reach out right now?" Neither alone is sufficient. A
perfect-fit company with zero intent is a nurture candidate. High intent from a
poor-fit company wastes sales cycles. Weight both dimensions and require minimum
thresholds on each, not just a combined total.
Decay scores over time - A lead who downloaded a whitepaper six months ago and
has not engaged since is not still a hot prospect. Apply time-based decay to
behavioral scores so inactivity reduces urgency. Fit scores (firmographic,
technographic) typically do not decay; behavioral scores should decay 10-25% per
month of inactivity.
Align sales and marketing on definitions - "Marketing Qualified Lead" means
nothing if sales uses a different threshold to decide whether to work it. Define
MQL, SQL, and PQL in a shared document, tie them to specific score thresholds, and
measure SLA compliance. Misalignment here is the single largest source of pipeline
leakage.
Start simple, iterate often - Begin with a manual point model covering 8-12
attributes. Get sales and marketing to validate it on historical closed-won data
before layering in predictive ML. Complexity that has not been validated destroys
trust faster than simplicity.
Validate with closed-won data - Build the model, then score your last 100
closed-won deals and 100 lost/no-decision deals. If the model does not clearly
separate the two populations, the attribute weights are wrong. Recalibrate before
deploying to live pipeline.
Core concepts
Demographic vs. behavioral scoring are the two axes of every lead score.
Demographic (also called profile or fit) scoring assigns points based on static
attributes: company size, industry, job title, tech stack, geography, funding stage.
These attributes describe who the prospect is. Behavioral scoring assigns points based
on actions: page visits, content downloads, email opens, webinar attendance, free trial
sign-ups. These attributes describe what the prospect is doing right now. Most models
maintain separate fit and behavioral sub-scores and require a minimum threshold on each
before routing to sales.
MQL / SQL / PQL definitions are the thresholds that gate handoffs between teams.
A Marketing Qualified Lead (MQL) has crossed a score threshold indicating marketing
believes it warrants sales attention. A Sales Qualified Lead (SQL) is an MQL that sales
has accepted as worthy of active pursuit, typically after a discovery call confirms fit,
budget, and timeline. A Product Qualified Lead (PQL) is specific to PLG motions - it
is a user (not just a lead) who has reached a product activation milestone that
predicts conversion, such as inviting a second user, creating three projects, or
integrating with a key tool.
Intent signals taxonomy classifies signals by source and strength. First-party
signals come from your own properties (website visits, docs engagement, trial usage,
email clicks) and are the highest-confidence because you own the data. Second-party
signals come from partner ecosystems (co-marketing events, integration marketplace
installs, referral partner activity). Third-party intent signals come from vendors
like Bombora, G2, TechTarget, or 6sense - they aggregate content consumption across
publisher networks to surface companies researching your category. Rank signals from
strongest (pricing page visit, free trial start) to weakest (single blog visit,
newsletter open).
Score decay is the mechanism that reduces a lead's behavioral score over time
without fresh engagement. Without decay, a lead's score only ever increases, making
old engagement permanently inflate priority. Implement decay as a scheduled job
(daily or weekly) that multiplies behavioral sub-scores by a decay factor (e.g., 0.9
per week of inactivity). Reset the decay clock when a new qualifying action occurs.
Fit scores are not decayed because firmographic attributes do not change frequently.
Common tasks
Define an Ideal Customer Profile (ICP)
An ICP is a description of the company type (not individual) most likely to buy,
retain, and expand. Build it from closed-won analysis, not intuition.
Firmographic criteria:
Industry verticals: e.g., FinTech, HealthTech, B2B SaaS
Company size (employees): e.g., 50-500
ARR / Revenue range: e.g., $5M-$50M ARR
Geography: e.g., North America, EMEA
Funding stage: e.g., Series A - Series C
Technographic criteria:
Tech stack signals: e.g., uses Salesforce + Slack (integrates well)
Competitor usage: e.g., currently on legacy tool X (displacement motion)
Infrastructure: e.g., AWS/GCP (cloud-native, not on-prem only)
Negative ICP (disqualifiers):
Explicitly list company types to reject: e.g., solo-founder, pre-revenue, regulated
industries you cannot serve, or geographies you do not support. These should
auto-fail leads regardless of behavioral score.
Pull your last 50 closed-won deals and cluster them by firmographic attributes.
The cluster with the shortest sales cycle and highest NRR is your ICP. Do not
define ICP by who you want to sell to - define it by who actually bought and stayed.
Build a scoring model - point system template
A point-based model assigns values to attributes. Sum the points to produce a score
from 0 to 100. Divide into a fit sub-score (0-50) and a behavioral sub-score (0-50).
Fit scoring template:
Attribute | Match | Points
-----------------------------|-------------------|-------
Industry match | Exact ICP | +15
| Adjacent | +8
| Outside ICP | 0
Company size | ICP range | +12
| One tier off | +6
Job title / seniority | Economic buyer | +10
| Champion / user | +7
| Unrelated | 0
Technographic signal | Key tech match | +8
Funding stage | ICP stage | +5
Geography | Target region | +0 (neutral)
| Excluded region | -20 (hard block)
Tier 3 - Early research (low weight, standard nurture):
Single blog post visit
Newsletter subscription
Podcast listen or video view
Social media follow or engagement
Third-party intent signals should boost a score but never alone qualify a lead.
They confirm category interest, not vendor selection. Combine with first-party
engagement before routing to sales.
Set MQL and SQL thresholds
Thresholds must be agreed by both sales and marketing before launch.
Threshold-setting process:
Score your last 100 closed-won deals with the proposed model
Score your last 100 lost/no-decision deals
Find the score that best separates the two populations (ROC curve / F1 score)
Set the MQL threshold at the point with acceptable false-positive rate for sales
Document the threshold in a shared definition document
Review and recalibrate quarterly
Recommended SLA after MQL:
MQL created → Sales accepts or rejects within 24 business hours
SQL created → First meaningful outreach within 4 business hours
Demo completed → Follow-up sent within 2 business hours
If sales rejects more than 25% of MQLs, the threshold is too low or the fit
criteria are wrong. Track MQL rejection reasons - they are the most actionable
feedback for recalibrating the model.
Implement score decay
Score decay prevents stale behavioral scores from inflating lead priority.
Decay implementation:
-- Pseudocode for weekly decay job
FOR EACH lead WHERE last_behavioral_action > 7 days ago:
behavioral_score = behavioral_score * 0.85 -- 15% weekly decay
IF behavioral_score < 5:
behavioral_score = 0 -- floor to avoid ghost scores
total_score = fit_score + behavioral_score
UPDATE lead record
Decay rate guidance:
Signal type | Decay rate | Rationale
---------------------|-----------------|----------------------------------
Single content click | -20%/week | Low-intent, fades fast
Webinar attendance | -10%/week | Higher effort, slower decay
Trial inactivity | -15%/week | Active usage is what matters
Demo no-show | -30% immediate | Strong disqualification signal
No engagement 90d | Reset to fit | Behavioral slate wiped clean
Validate scoring model against outcomes
Before going live, back-test the model against historical data.
Validation checklist:
Score last 6 months of closed-won deals - average score should be >60
Score same period of closed-lost deals - average score should be <40
Calculate separation ratio: (avg won score - avg lost score) / std dev
Run precision and recall: what % of deals above MQL threshold actually closed?
Identify attributes that are over- or under-weighted by inspecting outliers
Validate with sales: show them the top 20 scored leads, ask if they agree
Misalignment on definitions and handoff procedures is the most common reason lead
scoring fails to improve pipeline. Build a shared definition document covering:
Shared definition document structure:
1. ICP definition (firmographic + technographic criteria)
2. Negative ICP / auto-disqualify criteria
3. Lead lifecycle stages: Raw → Engaged → MQL → SAL → SQL → Opportunity
4. Score thresholds for each stage transition
5. Handoff SLA: who does what and within how long
6. Rejection protocol: how sales rejects an MQL and what reason codes to use
7. Recycling protocol: how rejected/lost leads re-enter nurture
8. Review cadence: monthly score review, quarterly model recalibration
Anti-patterns
Anti-pattern
Why it's wrong
What to do instead
Scoring only behavioral signals
A highly engaged person at a wrong-fit company wastes sales time
Require minimum fit sub-score before behavioral score can trigger MQL
Never decaying scores
Old engagement permanently inflates score; leads from 6 months ago stay "hot"
Apply weekly behavioral decay; reset to fit-only score after 90 days of inactivity
Back-test on closed-won vs. closed-lost before launching; set threshold at the empirical separation point
Treating all page visits equally
A pricing page visit is 10x stronger than a blog visit
Tier signals by purchase intent; assign points proportionally
Defining MQL without sales buy-in
Marketing routes leads sales won't work; both teams disengage
Co-define MQL criteria with sales leadership; make sales sign off on thresholds
Ignoring negative signals
Leads who unsubscribe or use competitor emails stay "qualified"
Apply score penalties or hard blocks for disqualifying actions
Building a complex ML model first
Black-box models are hard to debug and lose sales trust
Start with a transparent point model; add ML only after validating the manual model
Gotchas
Combined score threshold hides bad fits - A lead with a perfect behavioral score (50/50) but a zero fit score (wrong industry, wrong company size) can still hit the MQL threshold if you only gate on the total. Always require a minimum sub-score on both fit AND behavioral independently, not just the combined total.
Score decay runs on stale data when jobs fail silently - Decay jobs that error without alerting leave behavioral scores frozen at their peak indefinitely. The symptom is sales complaining that the pipeline is full of "ghost" hot leads who never respond. Add failure alerting and a last-ran timestamp check to your decay job.
Technographic signals are often stale - Data from providers like Clearbit or ZoomInfo can be 6-18 months out of date. A company that "uses Salesforce" in the data may have migrated off it. Treat technographic data as a weak fit signal, not a strong one, and never make it the sole reason for a high fit score.
PQL thresholds need per-product calibration - A "product qualified lead" in a tool with a 5-minute time-to-value (e.g., Loom) needs a completely different activation milestone than one with a 30-day ramp (e.g., a data platform). Copying PQL definitions from other companies' case studies without calibrating against your own activation data produces high MQL-rejection rates.
Sales rejection reasons are unstructured by default - Most CRMs let reps type free text when rejecting an MQL. Without a fixed reason code dropdown ("wrong company size", "no budget", "competitor", "timing"), you cannot aggregate feedback to recalibrate the model. Enforce a picklist from day one.
References
For detailed content on specific sub-domains, read the relevant file from references/:
references/scoring-models.md - Example scoring models for SaaS B2B, PLG, and
enterprise motions with full attribute tables and threshold recommendations. Load
when building or comparing scoring model templates for a specific GTM motion.
Only load a references file if the current task requires deep detail on that topic.
Companion check
On first activation of this skill in a conversation: check which companion skills are installed by running ls ~/.claude/skills/ ~/.agent/skills/ ~/.agents/skills/ .claude/skills/ .agent/skills/ .agents/skills/ 2>/dev/null. Compare the results against the recommended_skills field in this file's frontmatter. For any that are missing, mention them once and offer to install:
