How Deel Powers Real-Time Customer-Facing Analytics with Estuary

How Deel moved its database CDC to Estuary, cut reporting latency from hours to minutes, and reduced integration costs by 50 percent.

When batch ETL economics stop making sense

Before Estuary, Deel ingested data from production databases every two to six hours, but as volumes grew, a few problems started to compound.

The first was cost. Change data capture at Deel's scale was becoming expensive, and going real-time would have made it even worse. "If we went for real-time on the database side, our bill would have significantly increased," Xi recalls. "When there's a comparable product on the market, why not go for the more efficient one?"

The second was support: "Estuary brought us the proximity we need for the criticality of our deployments," Juan says. "Having a Slack channel and direct communication with the engineering team makes things easier and more transparent."

The decision came down to capability and economics: Estuary delivered robust CDC streaming in real time, with a cost structure that scaled. "The most important thing was enabling low-latency use cases at an economics that made sense," Juan says.

Migrating the core first

The team moved deliberately. Database ingestion, the highest-volume and highest-value workload, went first. Today, Deel's database CDC runs on Estuary, streaming from dozens of Aurora PostgreSQL databases across the company's services. SaaS sources followed and continue to migrate, with a few critical legacy workloads scheduled to move last.

The economics improved immediately. More importantly, the cost structure supported real-time streaming. "Real-time data unlocked use cases we couldn't consider before," Xi says. "For customer reporting, that's the difference between stale dashboards and live insights."

One platform, many front doors

What started as an ingestion migration turned into something broader. Estuary now works as Deel's data backbone, with data entering and leaving through whatever interface fits the job.

Streaming CDC into Snowflake

The main flow captures changes from production databases and materializes them into Snowflake using Snowpipe Streaming, so data lands continuously. This turned out to be a second source of savings: Snowpipe Streaming costs far less than keeping a warehouse running around the clock, so Deel gets live views of production state at a predictable, reduced cost. The landing layer preserves the full CDC history, then a dbt pipeline carries it into the analytical layer, internal APIs, and applications. SaaS sources sync hourly into the same flow.

Kafka-compatible streams for the lakehouse

For customer analytics, the team needed point-in-time state building on Apache Iceberg with low latency and low cost. Rather than deploying and operating a Kafka cluster, they used Dekaf, Estuary's Kafka-compatible API, and built a custom Spark Structured Streaming connector that writes to Iceberg. The result is a lakehouse fed in near real time, with no brokers to run.

ETL, CDC, streaming interfaces, and partner integrations all run through one platform, one contract, and one support channel.

The big win: customer analytics on live data

The most critical workload on the backbone is customer-facing analytics. Deel's customers see reporting built on data that flows through Estuary into Snowflake, with an Iceberg-based serving layer in development for even lower latency at lower cost.

Under the old batch cadence, that data was hours stale. "For customer reporting, people want the data instantly, rather than waiting hours," Xi says. "That enabled a whole set of use cases we couldn't imagine before."

Because these pipelines sit in the customer path, reliability is a necessity. Every record at Deel carries legal and financial significance, so our data governance ensures zero-tolerance for discrepancies. Juan's operational threshold: these pipelines cannot be down for hours.

The architecture that supports that bar has two layers:

• Landing holds the full CDC history, every change to every record.
• RAW holds the reduced state, one row per record at its latest value, mirroring the source. A reconciliation job collapses landing into RAW every 30 minutes.

Keeping both means the team validates downstream state against the change log rather than against the source database. That matters because permissions on the operational databases are restricted, so querying the source to verify a pipeline is not always available. The change history becomes the reference copy.

The full CDC history pays a second dividend: data quality. "Estuary tells us the story of the data," Juan explains. "We can validate data quality across our downstream layers independently, which is critical given compliance requirements around our customer data.”

Moving fast without a platform tax

Beyond any single pipeline, the team credits Estuary with changing how quickly Deel can test ideas. New sources come online in a few clicks, which means hypotheses get validated or killed in days. "This flexibility helps us validate hypotheses, fail fast, and try again," Juan says.

That agility compounds across a company where every team, from finance to product to the groups measuring how AI is improving operational workflows, pulls on the same data foundation. And it takes surprisingly few people to run: roughly 10 engineers operate all 150 connectors, spanning every production database and a long tail of SaaS tools, ad platforms, and support systems.

Results: hours to minutes, at half the cost

The move to streaming CDC gave Deel real-time customer-facing analytics, a 50% reduction in integration costs, and reduced operational overhead. Snowpipe Streaming cut warehouse costs on top of that. New integrations that once meant months of vendor onboarding now ship quickly.

"Estuary enabled building use cases we wouldn't have considered before," Juan says. "And it did it without exploding the cost."