April 71 hour 35 mins
Mike and Ian sit down with David Buonasera, CTO of Magnite, to unpack what it really takes to build and scale complex systems in ad tech.
Before becoming the CTO of Magnite, David co-founded SpringServe--a bootstrapped, video-native ad server that grew from an internal tool into an acquisition target. He started his career in finance before joining AppNexus as one of its first 18 employees.
The conversation covers two modes of building: greenfield at SpringServe, where David explains the opportunistic founding moment when Facebook shut down LiveRail, and unification of teams at Magnite. The dive into a technical breakdown of budget pacing in distributed ad systems: why enforcing a spend cap becomes one of the hardest problems in ad tech (along with the emotional reality of going through an acquisition process that doesn't always close).
Key Takeaways
- "Just stop spending at $100" is not a simple problem. In a distributed ad system, no single machine knows the true spend at any moment. Impression data arrives late, routes through different servers, and the feedback loop lags by minutes or hours. Budget overruns are almost never one bug--they're several compounding. The fix is less about clever algorithms and more about instrumenting the full data pipeline so you know exactly where things break down.
- When a critical system is failing, resist the urge to start over. Everyone's first instinct is to declare it beyond repair and rewrite it. That rewrite will take longer than you think, clients will leave before it ships, and you'll introduce new bugs in place of the old ones. Get it stable first, then have the rewrite conversation.
- A/B tests in ad systems can lie to you. Make your ad server faster, run a split test, and the faster side looks like it's printing money. Roll it out globally and the gain evaporates; budgets are finite and pacing algorithms just redistribute spend. Any test where both sides share a budget constraint isn't truly isolated.
- If teams are still on separate Slacks and VPNs, they're still separate companies. Tooling fragmentation is easy to deprioritize because it doesn't feel urgent. But every time an engineer has to switch VPNs to collaborate, it's a reminder that they don't actually work together yet.
- The number on the term sheet isn't the whole story. If the acquirer doesn't understand why they're buying you, the product gets deprioritized regardless of what the check said. And once you hire a banker and go to market, everyone knows you're for sale--which changes the dynamic in ways that don't always help you.
Chapters
- 00:00:00 - Meet Dave Buonasera
- 00:03:30 - Goldman to Ad Tech
- 00:07:30 - Inside Early AppNexus
- 00:13:00 - Democratizing Data Access
- 00:16:30 - Budget Pacing Explained
- 00:25:00 - Fixing Distributed Systems
- 00:29:00 - Building SpringServe
- 00:44:00 - Real-Time Data as a Feature
- 00:53:00 - Merging Cultures at Magnite
- 01:05:00 - Migration Without Client Disruption
- 01:12:00 - A/B Testing Pitfalls
- 01:16:00 - Navigating Acquisitions
The Build with Mike O'Sullivan is a podcast produced by The Current which is owned and operated by The Trade Desk, Inc. Learn more: https://www.thecurrent.com/
Sincera was acquired by The Trade Desk, Inc.