Every sales organization wants AI agents that find the right buyers, write the right message, and flag the deal that is about to slip. Few have the data foundation those agents need. Aurasell was built around that gap.
The company provides an AI-native go-to-market platform that replaces a stack of sales tools with agents that prospect, manage pipeline, generate quotes, and coach reps to close deals. Aurasell is part of the Built on Databricks program, with Databricks serving as the data foundation that powers its platform. With 950 million+ contacts and 84 million+ accounts out of the box, Aurasell's agents run the full revenue cycle automatically. As agent workloads scaled to hundreds of queries per second, the team replaced its traditional data warehouse serving layer with Lakehouse//RT — cutting costs 7x, reaching millisecond latency, and keeping every agent querying one governed copy of data.
AI agents need real-time data — and traditional data warehouses couldn’t keep up
Aurasell’s agents and dashboards generate roughly 300 to 400 queries per second around the clock. On a traditional data warehouse, queries averaged one to two seconds. Dashboards loaded slowly, agent conversations stalled, and powering that concurrency meant running several warehouse clusters at prohibitive cost.
The team evaluated copying hot data to external serving layers like ClickHouse, but replicating data to a separate system would create silos, break cross-table joins agents depend on, and introduce consistency risks.
“The moment you start copying data across different systems — inconsistency,” said Ejaz Veljee, Founding Data Engineer at Aurasell. “Agents querying one system versus another will get different responses. And these external systems restrict certain query patterns, so you end up with data your agents can’t even reach. That creates a silo.”
Lakehouse//RT: real-time performance without leaving the lakehouse
All of Aurasell's data was in the Databricks Lakehouse. Once Lakehouse//RT launched, the team could test immediately with no data migration and no new infrastructure.
80% of analytics and data discovery queries now route to Lakehouse//RT. A single Lakehouse//RT cluster handles 300 concurrent queries that previously required multiple warehouse clusters. Because the data stays in one place, governed by Unity Catalog, agents can join any table they need without the query-pattern restrictions a specialized serving store imposes.
"Earlier, to power 300 concurrent queries, I would need to spin up multiple clusters," Veljee said. "Now a single Lakehouse//RT cluster can power that same workload.
7x lower cost, millisecond latency, and a faster path to AI-first
The financial impact exceeded expectations. Aurasell reduced real-time serving costs by 7x, consolidating multiple warehouse clusters into a single Lakehouse//RT cluster.
Queries now return in 50 to 300 milliseconds at p90 depending on workload complexity, compared to a one-to-two-second average on a traditional data warehouse. The entire agent experience became noticeably faster, and in-product dashboards load without lag.
The biggest win was eliminating the separate real-time serving layer entirely. Every workload — data ingestion, AI model routing via Unity Gateway, analytics, and now real-time serving — runs on a single platform with a single copy of the data.
“To be an AI-first company, you have to be a data-first company,” said Srinivas Bandi, Co-founder and CTO of Aurasell. “If you’re not good at data, you can never be a good AI company.”
Explore more
Frequently asked questions about Lakehouse//RT
Aurasell routes 80 percent of its analytics and data discovery queries to Lakehouse//RT. Agents query data directly in the lakehouse at 50 to 300 millisecond latency across hundreds of concurrent queries.