Zembula fast-tracks its roadmap with Tinybird

Unfortunately, the technology they were using to deliver these real-time insights to customers wasn‘t working.

Slow queries and an ultimatum

Zembula initially chose Rockset to run their real-time analytics workloads. In June 2024, however, Rockset announced its acquisition and subsequent shutdown. Zembula was on the clock to migrate, but even before this news, the team was struggling to get the performance they needed from Rockset.

For some of their more complex use cases, Rockset simply couldn‘t handle the load. Queries were timing out, taking minutes to complete. Zembula was already at the compute limit of their existing Rockset instance, so they were facing a complex refactor on top of a massive increase in cost to scale the cluster.

Such a cost hike wouldn‘t just impact their own bottom line; it would undermine their core competitive value. Zembula‘s customers rely on them to boost Return on Ad Spend (ROAS). If costs go up, ROAS goes down, creating a churn risk for Zembula.

The founders knew this feature would be extremely valuable to their customers, but with Rockset it would simply be too expensive to deliver. They could build it, but the ROI wasn‘t there.

Shifting to structured data

Initially, Zembula had chosen Rockset in part because of its support for unstructured data. They thought that they could avoid having to worry about strictly defined schemas, giving them more flexibility to build analytics without needing to set up indexes, typed columns, and sorting keys.

As they‘ve progressed in their data maturity by building out these features, they‘ve shifted their mindset. Now, they see structured data as the key to achieving the performance and ROI that they need to give their customers a useful solution.

In the past, Zembula had evaluated different data tools including Redshift, BigQuery, AWS Timestream, Elasticsearch, ChaosSearch, Snowflake, and various other AWS solutions. At every turn, they ran into scalability issues and concerns about tooling.

Once they set their minds on more structured data formats, ClickHouse became an attractive option. Its columnar, structured storage format showed the potential to give them the cost-performance ratio they needed to deliver a valuable solution.

But, while ClickHouse as a foundational technology seemed promising, they were concerned about having the necessary tooling around the database to be able to be effective with it.

Tinybird fast-tracked their adoption of ClickHouse

The engineering team at Zembula decided to use Tinybird as its real-time analytics platform. It offered the underlying performance benefits of ClickHouse‘s structured data model, with the necessary tooling and support they would need to be able to get to production as quickly as possible.

One thing that helped immensely was Tinybird‘s DynamoDB Connector, which simplified the process of syncing data from their DynamoDB tables into Tinybird. With the connector, they simply had to map their DynamoDB items to columns in Tinybird, and Tinybird automatically handled backfilling and continually syncing the DynamoDB table using change-data streaming upserts.

Once the data was in Tinybird, Zembula found a bunch of efficiency in Tinybird‘s hosted API solution. With Tinybird, they can immediately publish any query as a dynamic, scalable, and fully-documented REST API Endpoint. Integrating the data back into the product became much simpler, as they didn‘t need to worry about setting up and scaling a backend or building out their own API.

They relied heavily on Tinybird‘s tokens implementation, using custom JWTs to create row-level access policies for their multi-tenant application.

They also learned to love the Tinybird CLI and its ability to represent a data project as code. With these tools they could easily integrate Tinybird into their CI/CD pipelines, which made code delivery for their data analytics features much easier to manage and much safer.

Tinybird‘s support got Zembula to production in a month

Zembula‘s customers include massive, multinational retail companies that send a bunch of emails. Staring down a complex migration with massive amounts of data - and with a tight deadline to complete it before the Rockset shutdown - Zembula leaned on Tinybird‘s support team to help them move billions of rows of highly granular email data from their old environment onto Tinybird‘s platform.

Data migrations are rarely easy, but with Tinybird’s support they become manageable. Zembula was included in the beta program for the DynamoDB Connector and became partners with Tinybird support as they gathered feedback and iterated together to support the migration while building a product that could scale and support future customers.

Tinybird‘s performance eclipsed Rockset

Any developer moving from an unstructured, NoSQL data model to a structured, schema-dependent model will have some added anxiety about the added friction of defining schemas upfront. But the engineers at Zembula learned that the payoff was worth it. After moving data into Tinybird and setting up table schemas, they were able to write faster queries, faster.

Before making these changes, Zembula was generating attribution and revenue decisioning models, but they couldn‘t expose them to the customers because the performance wasn‘t there. Zembula‘s team could provide quarterly business reviews based on these long-running analytics queries, but they couldn‘t actually build features within their application based on that data.

By moving the data to Tinybird, they achieved the performance they needed to put data directly into the hands of their customers in real time, at a cost that didn‘t undermine their ROI.

With Tinybird, Zembula changes its roadmap

When a designer creates a piece of content for an email campaign, they have to make choices about what components to use in their designs. Every component can impact the performance of that image, that email, and that campaign.

With Tinybird, Zembula has achieved a nearly instant feedback loop for designers. Instead of reviewing prior campaigns and trying to extract insight into the performance of various creatives weeks after the completion of a campaign, designers now see which components and creative elements will have the most impact as they select those components and add them to their designs.

Without Zembula, a designer could glean the overall performance of an email or campaign after the campaign ends. But with Zembula and the features they‘ve built with Tinybird, a designer now has deeply granular information about the performance of individual design elements as they‘re building.

Emboldened by what Tinybird gives them, Zembula has changed its roadmap. Previously unanswerable questions - “Should I use a star rating or the review text?“ “Should I choose a percentage discount or fixed value code?“ “Should I use an animation or static image?“ - are just a query away, answered with real-time, quantifiable proof.

With Tinybird, Zembula has the tooling, performance, and cost-effectiveness they need from a real-time data platform to build these massively differentiating features coveted by marketers at the world‘s largest retailers.