Kumo Co-Founder Hema Raghavan Named to Inc.’s 2026 Female Founders 500
Fraud & Financial Crime
From account takeover to crypto mixer detection — Kumo's graph transformers predict fraud across banking, crypto, and compliance before it happens. No feature engineering. No rules to maintain.
Book a demo and get a free trial of the full platform: data science agent, fine-tune capabilities, and forward-deployed engineer support.
By submitting, you accept the Terms and Privacy Policy.
$485B
Global fraud losses
Nasdaq 2024
23
Fraud scenarios covered
Banking + Crypto
15–25%
Net loss reduction
Average improvement
<1 hr
Time to deploy
No ML pipeline needed
From account takeover to money mule detection — Kumo predicts every major banking fraud scenario using the relational signals in your existing data.
From rug pulls to sanctions evasion — Kumo detects crypto crime patterns across on-chain and cross-domain data.
See how graph-learning AI compares to rules engines and traditional ML.
Detection approach
Traditional
Rules + isolated features
With Kumo
Graph transformers on relational data
Fraud ring detection
Traditional
Manual investigation
With Kumo
Automatic link prediction
Time to detect
Traditional
After fraud occurs
With Kumo
Days before fraud happens
False positive rate
Traditional
90–95%
With Kumo
30–50%
# of prediction types
Traditional
1 model per use case
With Kumo
23 from one platform
Setup time
Traditional
6–12 months
With Kumo
Under 1 hour
Simply connect your data, start asking predictions, and get results.Want more control? Fine-tune the model for your specific use case.
Integrates directly with your warehouse, no additional pipeline setup.
Ask questions in plain English and let Kumo do the modeling for you.
Get clear predictions and push them instantly into your workflows.
PREDICT COUNT(transactions.*, 0, 90, days) = 0
FOR EACH customers.customer_id
WHERE COUNT(transactions.*, -60, 0, days) > 03 lines. No feature engineering. No pipeline code.
For developers
Kumo's Predictive Query Language (PQL) replaces months of feature engineering, model training, and pipeline work with a few lines of SQL-like syntax. Describe what you want to predict — Kumo handles the rest.
Get accurate predictions on relational data instantly—no training or ML setup required.
Read the KumoRFM announcement
Vanja Josifovski
CEO and Co-Founder
Former CTO at Airbnb and Pinterest
Jure Leskovec
Co-Founder & Chief Scientist
Stanford Professor · Co-creator of RDL and GNN
Hema Raghavan
Co-Founder & Head of Engineering
Former Sr. Director of Engineering at LinkedIn
Loved by data scientists, ML engineers & CXOs at
Peer-reviewed
Kumo is built on 40+ peer-reviewed papers at NeurIPS, ICML, and KDD. The methodology is public and reproducible.
K. Huang, M. Fey, J. Leskovec et al.
A foundation model for relational data — pre-trained across schemas, it delivers accurate predictions out of the box and improves with fine-tuning.
Read paperM. Fey, W. Hu, K. Huang, J. Leskovec et al.
Learning predictive models directly on relational databases, eliminating the feature engineering pipeline.
Read paperJ. Robinson, R. Miao, K. Huang et al.
An open benchmark for evaluating relational prediction methods across 11 databases and 30 tasks.
Read paperSee what Kumo can predict from your existing relational database.
