Ibotta operates the largest digital promotions network in North America, reaching over 200 million consumers through a network of publishers called the Ibotta Performance Network. As more teams built applications that demanded millisecond-level response times, querying Delta tables through SQL warehouses introduced unpredictable latency and costly always-on compute. By adopting Lakebase on the Databricks Platform, Ibotta now serves curated data at sub-second speeds, with dramatically lower costs and zero outages since migration.
How Ibotta Serves Data to 200M+ Consumers with Lakebase
Ibotta connects more than 200 million consumers with digital promotions from thousands of brands and retailers through its mobile app and the Ibotta Performance Network. Behind that experience, machine learning models can rank and personalize offers that shoppers see, deliver campaign performance to clients via a large-scale analytics API, and support internal teams’ fraud detection and financial monitoring workflows. All of these systems depend on fast, reliable access to data.
As Ibotta expanded the number of production applications built on top of its data platform, serving curated data to those applications created friction and increased the need for predictable performance. Teams faced an uncomfortable trade-off between keeping a warehouse running around the clock and paying for idle compute overnight, or accepting latency spikes when clusters needed to warm up. As more teams began building applications on top of the data platform, neither option held up.
The engineering team considered an alternative path that would extract curated datasets and load them into a managed database on AWS. That approach would have introduced a new ETL layer and the governance burden of keeping data synchronized across two environments. "The second you start pulling data into other places, you have a lot more governance to deal with," said Quinlin McNatt, Staff Platform Engineer at Ibotta. "We were on the brink of building that ETL layer when we learned about Lakebase and realized it gave us a way to avoid that complexity altogether."
What Is Lakebase and How Did Ibotta Deploy It?
What Is Lakebase?
Lakebase is Databricks’ managed transactional database, built natively on the lakehouse. It exposes data as fully managed Postgres tables with built-in scaling, governed by Unity Catalog — meaning the same access controls that apply to Delta Lake extend directly into Lakebase without a separate sync or ETL layer.
How Did Ibotta Deploy Lakebase?
Instead of building a separate serving layer on AWS, Ibotta adopted Lakebase as a Databricks-native transactional store. Teams prepare and aggregate data where it already lives, then use Lakeflow syncs to move it into Lakebase, where it becomes available as fully managed Postgres tables with built-in scaling. Unity Catalog governs permissions and roles, so the same access controls applied to the data lake extend seamlessly into Lakebase.
The first production use case targeted the company's ML-powered recommendation engines. "Data scientists can focus on shaping the right features and data, without worrying about how it will be served. They publish a table, set it to snapshot mode and it becomes a cost-effective, production-ready asset. It removes the extra hurdle of getting data into production after the table is built," said Dennis Johnson, Staff Software Engineer on the machine learning platform team. Two recommenders now run on Lakebase, each delivering measurable improvements in both latency and cost.
Lakebase Use Cases at Ibotta
Adoption spread quickly to other teams. Platform engineers use Lakebase as the backend for Databricks Apps in areas like fraud detection and financial monitoring, with Unity Catalog permissions mapped directly into Lakebase roles. The largest use case is Ibotta's analytics API, which now returns campaign performance data to clients and internal operations teams with an average response time of 700 milliseconds across large datasets. When a new dataset needs to be served, the process no longer requires a multi-system project. “With the old approach, it could take three to five seconds to get data, which is too slow for customer-facing or even internal apps,” said Kenny Nguyen, Senior Data Engineer at Ibotta. “Lakebase gave us a fast path to sub-second performance, and we’re now seeing that pattern spread across teams.”
Ibotta's current Lakebase use cases include:
ML model serving: Two recommendation engines serve personalized offers with 3–10x lower latency and 90% lower compute cost than the prior system.
Customer-facing analytics API: Campaign performance data returned to brand clients and internal teams at an average response time of 700 milliseconds.
Fraud detection and financial monitoring: Databricks Apps backed by Lakebase, with Unity Catalog permissions mapped into database roles.
Results: Lakebase Results: 10x Lower Latency, 90% Lower Cost, Zero Outages
After migrating to Lakebase, Ibotta achieved 3–10x lower latency on ML model serving, reduced serving compute costs by 90% and now delivers analytics API responses averaging 700 milliseconds across large client datasets — compared to 3–5 seconds under the previous architecture.
Two recommendation systems achieved a three-to-ten-times decrease in latency after moving to Lakebase, and on serverless infrastructure, that speed translated directly into cost savings. One serving workload saw a 10x reduction in compute cost after moving to Lakebase. Reliability improved just as sharply. Outages that occurred roughly every other week under the legacy system dropped to zero incidents after migration. "We used to have an outage every other week where we had to restart everything," Dennis said. "We haven't had any incidents for a few months now."
The shift has also changed how teams spend their time. The platform engineering group, which previously invested significant effort in building ETL pipelines and synchronizing data across environments, now focuses on feature development and delivering higher-quality data to clients. New engineers ramp up faster, too, by learning a single platform rather than navigating separate systems. "Our engineering lift, the amount of work we used to put into building ETL and advanced database schemas, all of that is gone now," Quinlin said. "We've completely shifted our mindset from engineering solutions to how we can put the most useful data in front of our customers."
Why Ibotta Chose Lakebase Over a Separate AWS Database
Lakebase replaced the need for Ibotta to provision and maintain a separate AWS Relational Database Service (RDS) instance for ML Feature Serving, eliminating cross-environment ETL, synchronization overhead and the governance complexity of managing two data stores.
Ibotta is also running AI/BI Dashboards in production on financial datasets, enabling leadership to ask questions in natural language, spot spend anomalies and support forecasting and auditing. Looking ahead, the team is exploring Lakebase as a replacement for its AWS-based RDS store used in ML feature gathering, part of a longer-term push toward a unified platform. “Over time, we expect to move toward a unified platform,” said Kenny. “That means we won’t need to rely on RDS or other external systems in the same way. Databricks is becoming the standard.”