Austin Spiegel, Sift CEO on Hardware Data, Telemetry and Engineering

Sift Builds the Data Layer for Hardware Testing

Sift builds the data layer for hardware testing, a problem software observability tools cannot serve. A rocket engine's thousands of sensors emit data tens of thousands of times per second, and every point matters, so Sift unifies test telemetry into one source of truth that can trace an in-orbit failure back to a test. Spiegel locates the moat in the pace of execution and learning.

Takeaways

• Sift unifies hardware test telemetry into one source. A rocket engine carries thousands of sensors emitting data tens of thousands of times per second during a 60-second test, and every point matters because an anomaly can happen in a nanosecond. Pre-Sift teams stitched together Jupyter Notebook, LabVIEW and PLC data in different places.
• Telemetry economics hinge on tiered object storage. Less data comes from operations than from testing, and upwards of 90% of queries are for fresh data, yet customers refuse to delete history. Sift stores telemetry in S3 or Azure with tiering. Customers who once generated megabytes now produce hundreds of terabytes a month.
• Hardware firms treat software as a one-off project. Spiegel says hardware companies underestimate the scale of their data and treat software as a project with an end date rather than an ongoing investment. Build-versus-buy decisions often come down to wanting to own their destiny or keep the team busy.
• Execution pace is Sift's real moat. Spiegel calls Datadog a decent comparison and says infrastructure plus the pace of execution and learning are the moat. The company bought Salesforce rather than keep an account executive's home-built CRM because Salesforce's data model and infrastructure are best in class.
• Sift hires senior operators from hardware and autonomy. Mike scaled a startup's engineering team from 20 to 80 before its acquisition by Meta. Ben was head of North America sales at Applied Intuition, and Sim ran its application engineering organization, bringing commercial rigor to Sift's forward deployed engineers.
• A Raspberry Pi edge app reaches earlier-stage customers. Sift runs an edge application on a Raspberry Pi, which brings in earlier-stage customers, and Spiegel says it prices to meet startups where they are. Automated analysis runs low-code or Python checks as telemetry streams in, aborting a test on an unexpected signal.

How Sift lands hardware customers

1. The first demo is a generic walkthrough.
2. The second runs on the customer's own data.
3. Engineers sign off bottom-up.
4. Executives sponsor the deal top-down at primes.

What Spiegel expects next