Kumo Co-Founder Hema Raghavan Named to Inc.’s 2026 Female Founders 500
Industry
Inaccurate recommendations, demand forecasts, and churn models cost retailers millions every year. Traditional ML makes it worse by flattening relational data into feature tables, destroying the signals that actually predict purchase propensity, demand patterns, and churn risk. KumoRFM learns directly from the relationships in your data and is pre-trained on tens of thousands of datasets, delivering higher accuracy than any internally-built model, in hours, not months.
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.
Why retailers choose Kumo
Here's how Kumo transforms retail and e-commerce with relational AI.
$100M+
GMV from better recommendations
At DoorDash, Kumo-powered recommendations drive hundreds of millions in revenue. KumoRFM sees the full graph: customer, product, category, location, purchase history, and browsing behavior, not just item-to-item similarity.
10-50%
More accurate demand forecasts
Pre-trained on thousands of relational datasets, KumoRFM understands demand patterns across SKU-store-week granularity. It captures promotional effects, cannibalization, and seasonal shifts that flat models miss.
Hours
From idea to production model
Recommendations, demand forecasting, churn, LTV, dynamic pricing, inventory optimization. Deploy all from one platform without building separate feature pipelines for each use case.
Loved by data scientists, ML engineers & CXOs at
One platform, every prediction
Retail prediction problems share the same underlying data. Kumo learns from all of it simultaneously — no per-model engineering, no per-use-case pipelines.
Kumo learns from the full relational graph — purchases, browsing, returns, reviews — producing recommendations that reflect genuine preference, not just popularity.
Explore →
Predict demand at the SKU-store-week level. Kumo connects promotions, seasonal patterns, pricing changes, and supplier constraints into a single model.
Explore →
Declining visit frequency, shrinking basket size, unresolved tickets — Kumo detects compound signals and identifies at-risk customers weeks earlier than flat models.
Explore →
Price elasticity depends on segment, competitive context, and inventory. Kumo learns these interactions jointly — identifying optimal pricing per customer.
Explore →
LTV from transaction history, product affinity, support experience, and loyalty behavior. Kumo predicts individual-level LTV from all signals simultaneously.
Connect supplier tables, logistics events, and demand forecasts into a unified model. Kumo identifies disruption risk and optimizes allocation across your network.
Explore →
Recommend newly launched items based on purchase history and relational context — even with zero interaction data on the new products.
Explore →
High-quality embeddings for complementary categories, increasing attach rate and average order value. The model learns co-purchase patterns automatically.
Explore →
Predict staple depletion and trigger timely reminders. Kumo learns individual purchase cadences and surfaces restocking prompts before customers search.
The retail data advantage
Even if you have a world-class data science team, feature engineering fundamentally limits their accuracy. When you flatten 10, 20, or 50 relational tables into feature vectors, you discard the nuanced relationships between entities. A customer's relationship to products, transactions, returns, browsing sessions, and support tickets contains rich predictive signal. Flattening it into a single row loses most of that. This isn't a team quality problem — it's a structural limitation of the traditional ML approach.
Even the best internal model is trained on one retailer's data. KumoRFM is pre-trained on thousands of relational schemas across industries. It has already learned what “churn looks like” and “purchase intent looks like” across hundreds of different data structures. Your team, no matter how talented, cannot replicate this breadth. This is the same advantage GPT has over a custom NLP model — foundation model scale.
KumoRFM doesn't replace your data science team — it 10x's them. Instead of spending months on feature engineering and pipeline work, they define predictions in a simple query language. They go from shipping 3–5 models per year to 50+ per quarter. The tedious work disappears; the interesting work — defining what to predict, interpreting results, driving business impact — remains.
One platform powers recommendations, churn, demand forecasting, pricing, and every other retail prediction — with higher accuracy, in a fraction of the time, from the same connected data.
95%
Less data preparation
Automated feature engineering
10–50%
Accuracy improvement
Over traditional ML baselines
20x
Faster to production
From months to hours
$2.4M
Average annual savings
Reduced pipeline and engineering costs
KumoRFM isn't limited to your data alone. Pre-trained on billions of relational patterns across diverse datasets and fine-tuned to your schema, it sees what no in-house model can. As per the SAP SALT benchmark.
LLM
GPT4 + AutoML
PhD Data Scientist
Feature eng. + XGBoost
KumoRFM
Relational Foundation Model
40%
lift in recommendation conversion
“Beating internal XGBoost model on key metrics with far less data/features - on Kumo pre-trained. We replaced six months of pipeline work with a single afternoon.”
Matt Loskamp
GTM Data Science Leader, Enterprise Retail Customer
From startups to enterprises, leading organizations rely on Kumo to deliver predictive insights at scale.
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 on your specific data.
Read paperM. Fey, W. Hu, K. Huang, J. Leskovec et al.
Introduces learning predictive models directly on relational databases, eliminating the feature engineering pipeline that has historically bottlenecked enterprise ML.
Read paperJ. Robinson, R. Miao, K. Huang et al.
An open benchmark for evaluating relational prediction methods across 11 databases and 30 tasks. Kumo consistently outperforms traditional ML baselines.
Read paper
