Kumo Co-Founder Hema Raghavan Named to Inc.’s 2026 Female Founders 500
Loyalty & Retention
Churn doesn't happen overnight. Kumo detects the compound signals — declining usage, shrinking baskets, unresolved tickets — weeks before customers leave. Predict, intervene, and retain.
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.
7
Prediction types
Churn to reactivation
5x
Cheaper to retain
Than to acquire
<1 hr
To production
Per use case
0
Feature engineering
Fully automated
Loved by data scientists, ML engineers & CXOs at
7 retention use cases
From churn prediction to loyalty optimization — one relational foundation model powers every retention surface.
Detect compound signals - declining usage, shrinking order size, unresolved tickets - weeks before customers leave.
Identify which churned customers are most likely to return and which re-engagement strategy will work for each.
Predict engagement depth from product usage, support interactions, community activity, and feature adoption.
Predict which rewards, tiers, and incentives drive retention for each customer segment.
Forecast subscription renewals by learning from usage patterns, support history, and account relationships.
Combine product usage, support interactions, billing status, and engagement trends into a single predictive health score per account.
Predict which dormant users are most likely to return - and what offer or message will bring them back.
See how graph-learning AI compares to rule-based churn models and manual health scores.
Churn signals
Traditional
Lagging indicators
With Kumo
Compound relational signals
Intervention timing
Traditional
After churn
With Kumo
Weeks before
Feature engineering
Traditional
Months of manual work
With Kumo
Fully automated
Customer health scoring
Traditional
Manual rules
With Kumo
Learned from graph
Win-back targeting
Traditional
Batch campaigns
With Kumo
Individual predictions
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.
