Keebo autonomously tunes Snowflake and Databricks in real time - cutting cost without slowing a single query. Founded 2019 · Ann Arbor, Michigan.
Cloud data warehouses made it trivial to start querying enormous datasets - and just as trivial to overspend. Snowflake and Databricks lowered the barrier to entry, then quietly raised the recurring bill. For a growing number of companies, data spend now climbs faster than revenue. Keebo, a 41-person startup out of Ann Arbor, Michigan, was built to close that gap without asking anyone to babysit warehouse settings.
Founded in 2019 by Barzan Mozafari and Yongjoo Park, Keebo grew out of roughly fifteen years of database and machine-learning research at the University of Michigan, MIT and UCLA. Mozafari, an associate professor turned founder, had spent his academic career at the seam between machine learning and database systems. The company's premise was that conventional, human-driven tuning could not keep pace with exponential data growth - so the tuning itself had to be automated.
The name is a tell. Keebo derives from the Japanese word kibo - "hope." The product it sells is considerably less poetic: patented Data Learning technology that watches production workloads and continuously optimizes them, adjusting warehouse size, routing queries and scaling compute in real time. It does this, notably, with metadata-only access - Keebo reads how your warehouse behaves, never the data inside it.
That combination - academic rigor, an autonomous product, and a security posture enterprises can accept - is what turned a research idea into paying deployments at companies like Costco Travel, Dr. Squatch, PayJoy and Barstool Sports.
Above: the Keebo wordmark. The company translated a SIGMOD research paper into a production product that pays for itself in savings.
Figures reported by Keebo from customer deployments. Individual results vary by workload.
Modern data pipelines are easy to build and hard to optimize. More users, more queries, more sources - and every inefficiency compounds into wasted compute. The manual fix is a data engineer hand-tuning warehouse settings, a chore that never finishes and is obsolete the moment the workload shifts.
Cloud data costs grow faster than the business does. Warehouses are provisioned for peak, run at idle, and nobody has time to right-size them continuously.
Reinforcement-learning algorithms study real production workloads and act on them - rightsizing warehouses, routing queries and auto-scaling - continuously, not on a schedule.
Five layers of performance protection keep the savings honest. The system will not trade a cheaper bill for slower queries the team can't trust.
A $1M annual Snowflake spend at the ~27% average implies roughly $720/day back in the budget.
The flagship: autonomous, real-time tuning of Snowflake and Databricks warehouses - rightsizing, auto-scaling and query routing, wrapped in multi-layer SLAs.
Cost attribution and analytics that show data teams where spend comes from and where the optimization opportunities are across the data cloud.
Continuous tuning paired with cost transparency and forecasting - and success-based pricing that ties Keebo's fee to the savings it actually delivers.
"We've seen 50% to 70% savings - without Keebo the cost of this product would be unsustainable."
Nick Booth, Head of Backend Engineering
Keebo sells B2B SaaS, but the pricing is the unusual part. It's success-based: the fee is aligned to the verified savings delivered, so if the bill doesn't drop, neither does what Keebo charges. That inverts the normal software incentive, where a vendor is paid whether or not the product moves a number.
Onboarding leans on that same trust argument. Keebo connects with metadata-only access - it optimizes the warehouse without reading the underlying data. For regulated buyers in finance, healthcare and retail, that distinction is often the difference between a pilot and a hard no.
The competitive field is crowding. Rivals include focused data-cloud cost tools such as Bluesky, Espresso AI, Sundeck and SELECT, plus Snowflake and Databricks' own native controls and broad FinOps platforms like Vantage and CloudHealth. Keebo's differentiation is autonomy backed by academic research: not a dashboard that tells you what to do, but a system that does it - and publishes its methods at ACM SIGMOD.
Where it fits: the data-cloud FinOps layer, sitting between the warehouse and the finance team, converting query metadata into continuous cost decisions.
"Keebo's automated optimizations gave us a 50% reduction while empowering our team to focus on delivering value."
"With Keebo, I log in for a few minutes. Before Keebo, I spent hours on manual optimizations."
"We've seen cost savings while maintaining peak performance - without having to continuously tweak settings."
Barzan Mozafari and Yongjoo Park commercialize years of database research into a data-learning platform.
A $10.5M Series A led by True Ventures brings total funding to ~$15M; the flagship product launches.
The production optimization system is published as peer-reviewed research at ACM SIGMOD.
Keebo is named an AI-Based Snowflake Cost Optimization Solution of the Year.
Eric Shoemaker, former Device42 CRO, is appointed CEO to scale go-to-market; Mozafari remains engaged.