Tecton
Tecton provides enterprise feature management for production machine learning systems.
Tecton is an enterprise feature store that centralizes the management, serving, and governance of machine learning features used in production models. It solves the critical problem of training-serving skew by maintaining consistency between offline (batch) and online (real-time) feature environments, with online latencies under 5ms at 100K+ queries per second. The platform enables data teams to discover, reuse, monitor, and govern ML feature pipelines, reducing weeks from deployment cycles. Tecton was acquired by Databricks in August 2025.
Problem solved
Data scientists and engineers spend excessive time engineering, managing, and maintaining ML features across training and production environments, creating bottlenecks that delay model deployment by weeks or months.
Target customer
Machine learning teams at mid-market to enterprise companies (fintech, e-commerce, marketplaces) building real-time prediction systems; companies with complex feature engineering workloads requiring collaboration between data engineers and ML engineers.
Founders
M
Mike Del Balso
CEO & Co-Founder
Senior PM of ML platform infrastructure at Uber (2015-2018); previously led Search ads ML teams at Google; one of the architects of Uber's Michelangelo ML platform.
K
Kevin Stumpf
CTO & Co-Founder
Co-architect of Uber's Michelangelo ML platform; deep expertise in ML infrastructure and feature engineering systems.
J
Jeremy Hermann
Co-Founder (departed 2020)
Head of data infrastructure at Uber (2014-2015); founded and led Uber's Michelangelo ML platform team; held VP/director roles at Hearsay Systems, ZangZing, and Vontu.
Funding history
Series A
$20M
April 2020
Led by Sequoia Capital, Andreessen Horowitz
· Sequoia Capital, Andreessen Horowitz
Series B
$35M
December 2020
Led by Andreessen Horowitz, Sequoia Capital
· Andreessen Horowitz, Sequoia Capital
Series C
$100M
July 2022
Led by Kleiner Perkins
· Databricks, Snowflake, Andreessen Horowitz, Sequoia Capital, Bain Capital Ventures, Tiger Global
Total raised:
$160M
Pricing
Usage-based pricing model. Average annual contract value of $150k. Gross margins exceed 80%. Estimated 2024 ARR of $24.6M based on unverified sources.
Notable customers
PayPal, Atlassian, Doordash, Roblox, Plaid, Hello Fresh, Tide
Integrations
Snowflake, Amazon S3, Apache Kafka, Databricks, data warehouses, data lakes, Spark, Postgres
Website
Competitors
Feast
Open-source feature store with lower barrier to entry but less enterprise support and governance capabilities than Tecton's managed platform.
Databricks Feature Store
Integrated into Databricks lakehouse platform; tightly coupled to Databricks ecosystem, while Tecton offers broader data warehouse/lake integrations.
AWS SageMaker Feature Store
AWS-native solution with AWS ecosystem lock-in; Tecton provides cloud-agnostic approach with support for multiple data platforms.
Hopsworks
Open-source with managed cloud option; focuses on both feature engineering and model management, broader scope than Tecton's feature-store-focused approach.
Why this matters: Tecton addressed a critical pain point in production ML—the feature management gap that forces teams to rebuild and maintain features across training and serving environments. With $160M in funding and customers like PayPal and Atlassian achieving measurable improvements (50% increase in credit approvals, 4x fraud detection lift), Tecton established the commercial viability of feature stores as essential infrastructure. The Databricks acquisition validates the strategic importance of feature management to modern data platforms.
Best for: Enterprise ML teams building real-time prediction systems who need reliable feature serving, training-serving consistency, and cross-team feature discovery and governance.
Use cases
Real-time fraud detection
Financial services companies use Tecton to combine streaming transaction data with batch historical patterns to detect fraud in milliseconds. Tide achieved a 4x increase in fraud detection accuracy by using Tecton to manage features from both Snowflake and Kafka.
Dynamic pricing engines
E-commerce and marketplace platforms leverage Tecton to serve features with sub-second freshness for real-time price optimization. Features combine user behavior streams with product inventory and competitor data.
Personalized recommendations
Companies like PayPal use Tecton to power hybrid batch-and-streaming recommendation systems. The platform manages user interaction features from clickstreams alongside batch features from user profiles and product catalogs.
Credit decisioning
Fintech platforms use Tecton to manage features for lending models. Tide saw a ~50% increase in credit approval rates while decreasing losses by 5% by using Tecton to consolidate batch and streaming features.
Alternatives
Feast
Open-source and free to self-host, but requires internal infrastructure investment and lacks enterprise governance, monitoring, and support.
Databricks Feature Store
Choose this if you're already deeply invested in the Databricks ecosystem; Tecton offers more flexibility across data platforms and better offline-to-online consistency guarantees.
AWS SageMaker Feature Store
Better for teams already standardized on AWS; Tecton provides stronger multi-cloud support and superior low-latency online serving capabilities.
FAQ
What does Tecton do? +
Tecton is an enterprise feature store that manages machine learning features across training and production environments. It provides a centralized platform for storing features in offline stores (data warehouses/lakes), serving them in online stores with sub-5ms latency, and maintaining consistency between the two. The platform helps teams discover, reuse, monitor, and govern ML feature pipelines.
How much does Tecton cost? +
Tecton uses a usage-based pricing model with an average annual contract value of approximately $150k. Enterprise pricing is customized based on feature serving volume and storage needs. Contact Tecton for specific pricing.
What are alternatives to Tecton? +
Feast (open-source feature store), Databricks Feature Store (lakehouse-integrated), AWS SageMaker Feature Store (AWS-native), and Hopsworks (open-source with managed option).
Who uses Tecton? +
Enterprise and mid-market companies building production ML systems, particularly in fintech (Tide, PayPal), e-commerce/marketplaces (Doordash, Roblox, Hello Fresh), and SaaS (Atlassian, Plaid). Target customers are ML teams with significant feature engineering complexity and need for real-time serving.
Was Tecton acquired? +
Yes, Tecton was acquired by Databricks on August 22, 2025. Tecton is no longer an independent company and is now part of the Databricks platform.
What problems does Tecton solve? +
Training-serving skew (inconsistency between features used in model training vs. production), time-consuming feature engineering and pipeline management, difficulty discovering and reusing features across teams, lack of governance and monitoring for ML feature systems, and slow time-to-production for ML models due to feature engineering bottlenecks.
Tags
feature store
machine learning
MLOps
ML infrastructure
feature engineering
real-time ML
data engineering
training-serving skew