Overview
Tecton built a feature platform for machine learning teams. A feature is a piece of data a model uses to make a prediction, such as a customer's recent transaction count. Tecton helped teams prepare those features and serve them consistently to models in production. The company grew out of the founders' experience building Uber's Michelangelo machine learning platform. In 2025, Databricks acquired Tecton's assets to add real-time data capabilities to its own platform. [1][2][3]
Founding Story
Mike Del Balso, Kevin Stumpf and Jeremy Hermann worked together at Uber, where they helped create Michelangelo. They encountered a recurring gap between a model that worked in an experiment and a system that could make reliable predictions for real customers. A model needed the same definitions and timely data both when it learned from historical examples and when it was used live. [1][2]
The founders started Tecton to make that infrastructure available to other companies. Andreessen Horowitz says it invested in the seed round at the end of 2018 and later joined Sequoia in the Series A. Tecton publicly emerged from stealth in April 2020 with $25 million in seed and Series A funding and paying beta customers. [1][4]
Timeline
- 2018: Andreessen Horowitz made its seed investment, according to the firm. [1]
- 2020: Tecton launched from stealth and later made its feature platform generally available alongside a $35 million Series B. [4][5]
- 2021: It introduced low-latency streaming pipelines and became a core contributor to the open-source Feast feature store. [6]
- 2022: A $100 million funding round included Databricks and Snowflake as strategic investors. Tecton announced a Databricks partnership that year. [6][7]
- 2025: Databricks announced the deal; its counsel described it as an agreement to acquire Tecton's assets. Databricks later described the acquisition as completed. [3][8][9]
What They Built
Machine learning teams often transform raw events into signals before a model can use them. For fraud detection, a useful signal might be the number of attempted transactions in the past hour. Tecton's platform helped define those transformations, process historical and streaming data, and serve feature values to models. It aimed to reduce inconsistencies between the features used for training and those supplied during a live prediction. [1][5]
The company expanded from a feature store into a broader feature platform. In a 2024 interview, Del Balso said the founders saw data access, transformation and production integration as the bottleneck. Tecton also worked on delivering contextual data for generative AI applications. [10][11]
Market Position
Tecton sold infrastructure to teams that had moved beyond experimentation and needed to run models reliably. Its competition included internal systems, open-source tools such as Feast and services from larger data platforms. Tecton's contribution to Feast gave it a relationship with an open-source community as well as a commercial product. Its partnerships with Databricks and Snowflake placed it alongside platforms that could also provide adjacent functionality. [6][7]
Business Model
Tecton offered an enterprise feature platform. A 2021 customer case study describes a six-week evaluation with Tide, a financial technology company, before Tide adopted it for production models. The sources do not establish a complete public pricing schedule or Tecton's profitability. [12]
Traction
Tecton said its customer base grew more than fivefold in the year before July 2022, naming Convoy, HelloFresh, Plaid and Tide as customers. It also said annual recurring revenue nearly tripled from one fiscal year to the next, without disclosing the revenue base. A Tecton case study reported that Tide reduced model deployment time by more than half. These are company or customer claims, not independent audits. [6][12]
Outcome and evolution
Databricks announced in August 2025 that Tecton would join it. Databricks' legal adviser described a definitive agreement to acquire assets of Tecton; a later Databricks explanation says it completed the acquisition. The record cited here does not disclose the purchase price or establish that every Tecton asset, customer contract or legal entity was acquired. Databricks said it intended to combine real-time data serving with its AI tools. [3][8][9]
Key Lessons
A useful model depends on the data supplied to it after deployment. Tecton's product focused on maintaining that supply, not just training the model. Its path from a 2022 partnership with Databricks to the 2025 asset deal also illustrates how an infrastructure vendor can become part of a larger platform; the sources do not establish the financial return or all causes of the transaction. [7][8]
Sources
- Andreessen Horowitz: Investing in Tecton
- Tecton launch announcement
- Databricks: What is a feature store?
- Tecton launch announcement and funding
- TechCrunch: Tecton's Series B and launch
- Tecton: $100 million funding announcement
- Tecton: Databricks partnership
- Fenwick: acquisition of Tecton's assets
- Databricks: Tecton is joining Databricks
- Cerebral Valley: interview with Mike Del Balso
- CRN: Tecton expands into generative AI
- Tecton: Tide case study