Account Scoring
“Which target accounts will generate revenue in the next 90 days?”
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 target accounts will generate revenue in the next 90 days?
Account-based marketing teams target hundreds of accounts but lack the ability to rank them by actual revenue potential. Static ICP definitions based on employee count and industry miss the dynamic engagement signals that predict conversion. Sales and marketing waste alignment cycles on accounts that were never going to close, while high-intent accounts slip through the cracks unnoticed.
Quick answer
Account scoring predicts which target accounts will generate revenue within a defined time window (typically 90 days). Unlike static ICP definitions based on employee count and industry, graph-based account scoring learns from dynamic engagement signals, cross-account relationships, and real buying behavior to continuously update account priority.
Approaches compared
4 ways to solve this problem
1. Static ICP Matching
Define your ideal customer profile (industry, size, region, tech stack) and score accounts by how closely they match. The standard ABM approach used by most marketing teams.
Best for
Early-stage ABM programs that need a starting point for account selection. Quick to implement with no data science required.
Watch out for
Static by definition. An account that matches your ICP perfectly but shows zero engagement scores the same as one with active champion engagement. Misses the dynamic signals that predict actual buying behavior.
2. Intent Data Platforms (Bombora, G2, 6sense)
Enrich accounts with third-party intent signals: topic research spikes, competitor comparison activity, review site visits. Score accounts by intent volume and relevance.
Best for
Teams that want signals beyond their own data. Good for detecting early-stage research behavior before the account engages with your brand.
Watch out for
Intent data is noisy and lagged. Many accounts research topics without buying intent. The signals are aggregated at the account level, missing which contacts are actually driving the intent. And you share the same signals with every competitor using the same provider.
3. ML Scoring on CRM + Engagement Data
Train a classifier on first-party CRM data: opportunity history, contact engagement scores, meeting activity, and firmographics. Predict revenue probability per account.
Best for
Teams with clean CRM data and ML infrastructure. Captures first-party engagement signals that third-party intent data misses.
Watch out for
Treats each account as an independent row. Misses cross-account patterns like 'accounts whose contacts share a professional network with recent customers' or 'industry peers that closed deals in the last 90 days.' Feature engineering is a bottleneck.
4. KumoRFM (Graph Neural Networks on Relational Data)
Connects accounts, opportunities, contacts, and industry relationships into a relational graph. Learns cross-account signals like 'accounts with 2+ high-engagement contacts whose industry peers recently closed deals.' Continuously updated as engagement data flows in.
Best for
B2B companies with multi-table CRM data, multiple contacts per account, and meaningful industry/partner relationships.
Watch out for
Requires contact-level engagement data, not just account-level aggregates. If your CRM only has account-level firmographics with no contact activity tracking, enrich it first.
Key metric: Graph-based account scores identify revenue-generating accounts with 2.8x higher precision than static ICP models. SAP SALT benchmark: 91% accuracy for multi-table models vs 75% for single-table approaches.
Why relational data changes the answer
Account A001 (Vertex Financial) has two contacts with engagement scores above 80 (VP Data Science at 92, CTO at 85). A flat model might score this account highly based on these contact attributes alone. But the relational graph reveals why the score should be even higher: 3 industry peers in Finance closed deals in the last 90 days, suggesting a sector-wide buying cycle. The VP Data Science is connected to a contact at an existing customer who recently expanded their contract. And the opportunity is in Negotiation stage at $120K, which for Finance enterprises at this size, closes 78% of the time.
Conversely, Account A004 (Nova Tech) has a single ML Engineer contact with an engagement score of 23. The flat model flags this as low-priority. The graph confirms: no industry peers have closed recently, the contact's role does not have budget authority, and the APAC region has a historically lower close rate. But crucially, the graph also shows that Nova Tech's CTO recently joined the company from an existing customer, a weak signal that a flat model would never see but the GNN picks up from the contact-company-customer relationship graph. These multi-hop relational signals are what separate 93% prediction accuracy from 75%.
Scoring accounts with firmographic matching is like a real estate agent pricing a house based only on square footage and zip code. A relational model also sees that three neighboring houses just sold above asking (industry peers closing), the school district rating improved (market momentum), and the seller's agent has a track record of fast closes (champion engagement). The fundamentals matter, but the relational context determines timing and likelihood.
How KumoRFM solves this
Relational intelligence for smarter acquisition
Kumo connects your ACCOUNTS, OPPORTUNITIES, and CONTACTS tables into a single relational graph. The model learns cross-account patterns — like 'accounts whose contacts have engagement scores above 80 and share industry peers that recently closed deals' — automatically. No manual ICP definition required. The result is a continuously updated account score that reflects real buying signals, not stale firmographics.
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
ACCOUNTS
| account_id | company_name | industry | employee_count | region |
|---|---|---|---|---|
| A001 | Vertex Financial | Finance | 12,000 | North America |
| A002 | Orbit Retail | Retail | 3,500 | EMEA |
| A003 | Helix Health | Healthcare | 8,200 | North America |
| A004 | Nova Tech | Technology | 1,800 | APAC |
OPPORTUNITIES
| opp_id | account_id | amount | stage | timestamp |
|---|---|---|---|---|
| OP01 | A001 | $120,000 | Negotiation | 2025-10-15 |
| OP02 | A002 | $45,000 | Discovery | 2025-10-20 |
| OP03 | A003 | $89,000 | Proposal | 2025-11-01 |
CONTACTS
| contact_id | account_id | title | engagement_score |
|---|---|---|---|
| C01 | A001 | VP Data Science | 92 |
| C02 | A001 | CTO | 85 |
| C03 | A002 | Director Analytics | 41 |
| C04 | A003 | Chief Risk Officer | 78 |
| C05 | A004 | ML Engineer | 23 |
Write your PQL query
Describe what to predict in 2–3 lines — Kumo handles the rest
PREDICT SUM(OPPORTUNITIES.AMOUNT, 0, 90, days) > 0 FOR EACH ACCOUNTS.ACCOUNT_ID
Prediction output
Every entity gets a score, updated continuously
| ACCOUNT_ID | TIMESTAMP | TARGET_PRED | True_PROB |
|---|---|---|---|
| A001 | 2025-11-01 | True | 0.93 |
| A002 | 2025-11-01 | False | 0.18 |
| A003 | 2025-11-01 | True | 0.82 |
| A004 | 2025-11-01 | False | 0.09 |
Understand why
Every prediction includes feature attributions — no black boxes
Account A001 — Vertex Financial
Predicted: True (93% probability)
Top contributing features
Two contacts with engagement scores above 80
92, 85
31% attribution
Active opportunity in Negotiation stage
$120K
28% attribution
Industry peers closed 3 deals in last 90 days
3 deals
19% attribution
Employee count > 10,000 (enterprise tier)
12,000
14% attribution
Region — North America
North America
8% 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 account scoring
What is the difference between lead scoring and account scoring?
Lead scoring evaluates individual contacts. Account scoring evaluates entire companies by aggregating signals across all contacts, opportunities, and engagement at the account level. In B2B selling, the buying decision is made by the account (a committee of people), not by a single lead. Account scoring reflects this reality.
How do you score accounts with no engagement data?
Graph models transfer signals from similar accounts. An account with zero first-party engagement but strong firmographic similarity to recent customers, industry peer buying activity, and a newly hired executive from an existing customer still generates a meaningful score. The graph propagates these second-order and third-order signals automatically.
How often should account scores be refreshed?
Daily or real-time. A new opportunity created, a contact engagement spike, or an industry peer closing a deal should all update the account score immediately. Monthly or quarterly refresh cycles miss the dynamic signals that make account scoring valuable for tactical ABM execution.
Can account scoring identify the right time to engage?
Yes. The time dimension is critical. An account that scores 0.82 with an accelerating engagement trend is a better target this week than one that scores 0.85 with a flat trend. Graph models capture these velocity signals by learning from the temporal sequence of engagement events, opportunity stage changes, and peer account behavior.
Bottom line: Kumo account scores identify revenue-generating accounts with 2.8x higher precision than static ICP models, letting ABM teams concentrate budget on the accounts most likely to close.
Related use cases
Explore more acquisition 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.
