Phantom gives crypto traders faster buy/sell decisions with Tinybird

Phantom is a secure, all-in-one cryptocurrency wallet that lets you buy, hold, send, and receive digital assets while building your portfolio of cryptocurrencies and NFTs. With millions of users, Phantom is known for its ease of use for both beginners and crypto-enthusiasts.

Stale data, quality issues, and a brittle stack

The crypto market moves fast. Crypto traders need to know the exact price of a coin at any given moment to make buy/sell decisions. There is a lot of information to keep track of - Are the coins going up? Down? Are they trending? With thousands of coins being launched every day, it is important to be able to process a lot of data in real time.

Phantom wanted to expose their wallet holders to more and new coins to increase adoption and revenue. They set out to build a real-time discovery feature to allow users could explore and identify trending tokens. They knew this was a data-driven feature that demanded scalable real-time infrastructure - something their current stack could not support.

Phantom had engineered a piecemeal real-time engine with Postgres, RudderStack, Snowflake, and Redis, but it had three main problems:

• Delayed data: The data that Phantom was serving to users was delayed by an hour or more. Their current infrastructure for ingestion, Rudderstack, was inefficient with streaming ingestion into Snowflake. Once the data arrived in Snowflake, they relied on in-warehouse transformation models, which took time to run, resulting in stale data for end-users.
• Bad data quality: Phantom was using third-party aggregated crypto data instead of the massive amounts of proprietary data they had gathered, simply because they lacked the infrastructure and tooling to process terabytes of data in real time. Relying on third-party data sources exposed them to spam and fake data, leading to a poor UX.
• Technically brittle: The piecemeal solution resulted in technical brittleness that concerned Phantom's engineers. For example, one Phantom data engineer was responsible for managing the data ingestion pipeline, and their existing workflow required running a script locally and copying the queries into Snowflake. If the engineer took vacation (or ended up leaving the company), data wouldn't make it into their pipeline without a manual fallback system.

Finding real-time infra and tooling

In search of better real-time data infrastructure, Phantom began evaluating stream processing platforms that would be both powerful enough to handle their real-time data requirements and user-friendly enough for their entire engineering team to use.

Phantom prioritized two things: managed, integrated infrastructure and collaborative tooling with a low barrier to entry for their entire engineering team. They considered self-managed Flink/Spark, Arroyo, and Materialize, but none of these tools would reduce their infra burden, and few offered the collaborative, code-first tooling they sought.

Phantom ultimately chose Tinybird, a real-time analytics backend, because it eliminated infrastructure concerns and allowed them to build and deploy their real-time data pipelines as code using git and GitOps. With Tinybird, they built and tested a functional proof of concept within a day, something that might have taken weeks (or more) had they chosen to build out their own analytics backend.

Faster user experiences, more revenue

Phantom chose Tinybird as the analytics backend for their discovery feature, providing their users with a fast, real-time experience buying and selling cryptocurrencies. After implementing the feature, they saw an 8% increase in swaps (buys), a 4% increase in overall revenue, and a 2x increase in users interacting with coins they didn't yet own.

After their success with the discovery feature, Phantom quickly added two more use cases with Tinybird. First, they wanted to track in real time if a token had graduated or was graduating (that is, the token has become mature and is available to the world to be purchased), which could help influence their millions of users to make buy decisions. They were able to quickly build this feature by connecting a Kafka stream with blockchain data to Tinybird, writing SQL to analyze the chain for recent graduations, and exposing the transformation as a real-time, RESTful API.

Second, they created a feature to show a more complete view of a user's event history. Like the validating graduated status feature, this historical feature needed to be fast; a user could open the app and see their entire activity history instantly.

When looking for a new tool, Phantom's key objective was to minimize friction, enable their engineers, and ship fast. They had wanted a tool that anyone on the product and engineering team could use and that would allow them to integrate various data sources with zero infrastructure setup and minimal overhead. In Tinybird, they found a platform that cut their infrastructure costs, eliminated multiple tools from their stack, and allowed them to quickly deliver new features that have subsequently increased their revenue.