Lead Scoring
“Which leads will convert to paid?”
Book a demo and get a free trial of the full platform: research agent, fine-tune capabilities, and forward-deployed engineer support.
By submitting, you accept the Terms and Privacy Policy.
Loved by data scientists, ML engineers & CXOs at
A real-world example
Which leads will convert to paid?
SDR teams spend 65% of their time on leads that will never convert. A SaaS company generating 10,000 MQLs per month with a 3% conversion rate wastes $2.4M annually in SDR labor on dead leads. Traditional lead scoring based on form fills and page views misses the buying signals hidden in multi-contact engagement patterns, firmographic fit, and the temporal sequence of interactions that distinguish real buyers from researchers.
Quick answer
The best B2B lead scoring models connect multi-contact engagement patterns, firmographic fit, content consumption sequences, and buying committee behavior in a relational model. When 3+ contacts from the same account view pricing pages, download security whitepapers, and attend a webinar within 14 days, that account converts at 15x the base rate. SAP SALT benchmark shows relational graph ML achieves 91% accuracy vs 75% for XGBoost on flat tables.
Approaches compared
4 ways to solve this problem
1. Point-based scoring (Marketo/HubSpot)
Assign points for form fills, page views, email opens, and demographic attributes. Threshold triggers MQL status.
Best for
Easy to set up in existing marketing automation. Provides a directional signal that is better than no scoring at all.
Watch out for
Scores individual contacts, not buying committees. A VP who downloads one whitepaper scores lower than an intern who fills out five forms. The model rewards activity volume over buying intent.
2. Logistic regression on lead attributes
Train a conversion model on lead source, job title, company size, and engagement count to predict closed-won probability.
Best for
More accurate than point-based scoring because it learns which attributes actually correlate with conversion from historical data.
Watch out for
Still treats each lead as an individual. B2B purchases involve 6-10 stakeholders. When two VPs and a CTO from the same account are all engaging, that is a buying signal no single-contact model can detect.
3. Account-based scoring (ABM platforms)
Aggregate engagement signals across all contacts at an account and score at the account level using firmographic and intent data.
Best for
Captures multi-threaded engagement at the account level. Good fit for enterprise sales with long buying cycles.
Watch out for
Typically uses simple aggregation (sum of touches) rather than learning the engagement sequences that predict conversion. Three blog views from one person score the same as a pricing page view, case study download, and demo request from three different stakeholders.
4. KumoRFM (relational graph ML)
Connect leads, contacts, activities, content views, and firmographics into a graph. The GNN learns multi-threaded buying committee patterns, engagement velocity, and firmographic fit signals across the entire lead-to-revenue funnel.
Best for
Highest accuracy. Detects buying committee coordination, engagement sequence quality, and firmographic similarity to recent closers in a single model.
Watch out for
Requires multi-contact tracking at the account level. If your CRM only tracks individual leads without account linkage, the buying committee signal is unavailable.
Key metric: SAP SALT benchmark: relational graph ML achieves 91% accuracy vs 75% for XGBoost on flat tables in lead conversion prediction tasks.
Why relational data changes the answer
B2B buying is a committee decision that unfolds across multiple contacts and multiple interaction types. The signal lives in contact engagement (who is engaging, at what seniority, and from which departments), content consumption (pricing page vs blog post carries very different intent), firmographic data (company size, industry, tech stack compatibility), and temporal patterns (engagement accelerating over 14 days vs spread over 6 months). A flat lead score that counts total page views and form fills misses the difference between a researcher collecting content and a buying committee converging on a purchase decision.
Relational models connect contacts to accounts to activities to content and learn patterns like 'VP of Engineering and CTO from the same account both viewed the pricing page, the VP downloaded the security whitepaper, and they attended the same webinar within 14 days.' SAP SALT benchmark shows relational graph ML achieves 91% accuracy vs 75% for XGBoost on flat tables. For a SaaS company generating 10,000 MQLs per month, that accuracy gap is the difference between SDRs spending time on the top-20% of leads that will actually convert and wasting 65% of their effort on leads that were never going to buy.
Scoring leads by individual activity is like evaluating a restaurant reservation by counting the number of phone calls. One call from a party planner booking 30 seats is worth more than ten calls from someone checking if you have outdoor seating. What matters is who is calling, what they are asking about, and whether multiple people from the same group are reaching out. Graph ML reads the buying signal in the committee pattern, not the activity count.
How KumoRFM solves this
Graph-learned product intelligence across your entire account base
Kumo connects leads, contacts, activities, content views, and firmographic data into a graph where buying intent propagates through the company network. It learns that when 3+ contacts from the same account view pricing pages, download the security whitepaper, and attend a webinar within 14 days, that account converts at 15x the base rate. The model captures multi-threaded buying committee behavior, firmographic similarity to recent closers, and engagement velocity that single-contact scoring cannot detect.
From data to predictions
See the full pipeline in action
Connect your tables, write a PQL query, and get predictions with built-in explainability — all in minutes, not months.
Your data
The relational tables Kumo learns from
LEADS
| lead_id | company | source | created_date | status |
|---|---|---|---|---|
| LD001 | Acme Corp | Webinar | 2025-02-15 | MQL |
| LD002 | DataTech | Google Ads | 2025-02-20 | MQL |
| LD003 | MegaRetail | Organic | 2025-03-01 | MQL |
CONTACTS
| contact_id | lead_id | title | department | seniority |
|---|---|---|---|---|
| CT01 | LD001 | VP Engineering | Engineering | VP |
| CT02 | LD001 | CTO | Engineering | C-Level |
| CT03 | LD002 | Data Analyst | Analytics | IC |
ACTIVITIES
| activity_id | contact_id | type | timestamp |
|---|---|---|---|
| ACT01 | CT01 | Demo request | 2025-02-18 |
| ACT02 | CT02 | Pricing page view | 2025-02-19 |
| ACT03 | CT03 | Blog view | 2025-02-22 |
CONTENT_VIEWS
| view_id | contact_id | content_type | title | timestamp |
|---|---|---|---|---|
| CV01 | CT01 | Whitepaper | Security & Compliance | 2025-02-16 |
| CV02 | CT02 | Case study | Enterprise deployment | 2025-02-17 |
| CV03 | CT03 | Blog | Getting started guide | 2025-02-22 |
FIRMOGRAPHICS
| firm_id | lead_id | industry | employees | revenue | tech_stack |
|---|---|---|---|---|---|
| FG01 | LD001 | Technology | 2500 | $500M | Snowflake, AWS |
| FG02 | LD002 | Analytics | 45 | $5M | PostgreSQL |
| FG03 | LD003 | Retail | 12000 | $2B | Azure, Databricks |
Write your PQL query
Describe what to predict in 2–3 lines — Kumo handles the rest
PREDICT BOOL(LEADS.STATUS = 'Closed Won', 0, 90, days) FOR EACH LEADS.LEAD_ID WHERE LEADS.STATUS = 'MQL'
Prediction output
Every entity gets a score, updated continuously
| LEAD_ID | COMPANY | CONTACTS | CONVERSION_PROB_90D |
|---|---|---|---|
| LD001 | Acme Corp | 2 (VP + CTO) | 0.76 |
| LD002 | DataTech | 1 (IC) | 0.09 |
| LD003 | MegaRetail | 1 (Dir) | 0.34 |
Understand why
Every prediction includes feature attributions — no black boxes
Lead LD001 -- Acme Corp, Technology, 2,500 employees
Predicted: 76% conversion probability within 90 days
Top contributing features
Multi-contact engagement
2 contacts, VP + C-Level
31% attribution
High-intent content viewed
Security + Case study
24% attribution
Firmographic fit score
92% match to ICP
19% attribution
Demo request within 3 days
Yes
14% attribution
Tech stack compatibility
Snowflake (key integration)
12% attribution
Feature attributions are computed automatically for every prediction. No separate tooling required. Learn more about Kumo explainability
PQL Documentation
Learn the Predictive Query Language — SQL-like syntax for defining any prediction task in 2–3 lines.
Python SDK
Integrate Kumo predictions into your pipelines. Train, evaluate, and deploy models programmatically.
Explainability Docs
Understand feature attributions, model evaluation metrics, and how to build trust with stakeholders.
Frequently asked questions
Common questions about lead scoring
What is predictive lead scoring for B2B SaaS?
Predictive lead scoring uses ML to estimate the probability that a lead will convert to a paying customer. The best models connect multi-contact engagement, firmographic fit, and content consumption patterns across the entire account. Unlike point-based scoring, predictive models learn which signals actually correlate with conversion from historical data.
How do you detect buying committee signals?
Track engagement at the contact level and connect contacts to their account. When 3+ contacts from the same account engage within a short window (14 days), especially from different departments and seniority levels, it signals a buying committee in motion. This multi-threaded pattern converts at 15x the rate of single-contact engagement.
What data improves B2B lead scoring accuracy?
Three data sources provide the biggest lift: multi-contact engagement data (who at the account is engaging, not just how much), firmographic data (industry, company size, tech stack), and content-level detail (pricing page views carry more intent than blog views). Adding tech-stack compatibility data (do they use tools your product integrates with?) further improves firmographic fit scoring.
How much can lead scoring improve SDR efficiency?
SDR teams spend 65% of their time on leads that will never convert. Routing SDRs to the top-20% of model-scored leads doubles conversion rates from 3% to 6%, adding $4.8M in new ARR annually for a company generating 10,000 MQLs per month. SDR productivity doubles because they are working higher-quality leads.
Should you use account-level or contact-level lead scoring?
Both, connected. Contact-level engagement tells you who in the buying committee is active and what they care about. Account-level signals tell you firmographic fit and buying committee breadth. The best models connect both levels: a single VP engaging is interesting, but a VP plus CTO plus a technical evaluator engaging simultaneously is a strong buy signal.
Bottom line: A SaaS company generating 10,000 MQLs per month that routes SDRs to the top-20% leads doubles conversion rates from 3% to 6%, adding $4.8M in new ARR annually. Kumo detects multi-threaded buying committee engagement and firmographic fit signals that single-contact lead scores miss entirely.
Related use cases
Explore more B2B SaaS use cases
Topics covered
One Platform. One Model. Infinite Predictions.
KumoRFM
Relational Foundation Model
Turn structured relational data into predictions in seconds. KumoRFM delivers zero-shot predictions that rival months of traditional data science. No training, feature engineering, or infrastructure required. Just connect your data and start predicting.
For critical use cases, fine-tune KumoRFM on your data using the Kumo platform and Research Agent for 30%+ higher accuracy than traditional models.
Book a demo and get a free trial of the full platform: research agent, fine-tune capabilities, and forward-deployed engineer support.
