Snowball: Google Cloud PLG
Product Manager · Google · Dec 2025–Sep 2026
A cross-sell platform for Google Cloud, rebuilt around removing the engineering bottleneck instead of adding more targeting.
The Question
How do you get customers already using part of a platform to discover more of it, without engineering having to build every campaign by hand?
Context
Snowball recommends Google Cloud products to existing customers. It was running on 2 active surfaces, and every new campaign required an engineering ticket, a real ceiling on how fast marketing could actually move.
The Constraint
Every new campaign needed an engineering ticket. Marketing could design a good cross-sell motion in a day and then wait weeks for someone to ship it.
My Hypothesis
The obvious next move looked like better targeting: a smarter model recommending the right product to the right customer. I bet the real constraint was upstream of that: nobody could ship a campaign fast enough to test whether better targeting even mattered.
What I Tried
Shipped a self-serve documentation layer for campaign creation, aimed squarely at removing that engineering dependency. Shipped a flexible, multi-campaign Top Banner component to production, replacing multiple narrower legacy components that couldn't be reused across campaigns. Also proposed an early framework for how autonomous AI agents should reason about which cloud products to recommend, ahead of the platform's next phase.
The Decision
Prioritized unblocking marketing teams over improving targeting sophistication first. The bet was that removing the bottleneck would compound faster than any single smarter recommendation would.
The Build
A self-serve documentation layer for campaign creation, and a flexible multi-campaign Top Banner component that replaced several narrower one-off components.
The Evidence
$200M+ in incremental revenue and 70k+ influenced meaningful adoptions year to date. Active surfaces expanded from 2 to 5, driving a 3x increase in monthly impressions to 30.6M against 80M eligible impressions. Campaigns launched increased 68%, from 64 to 108, with the majority of that growth from teams who no longer needed an engineer to ship one.
What Changed
Campaigns launched jumped 68%, from 64 to 108. Most of that growth came from marketing teams shipping directly, with no engineering ticket in between.
What I Learned
The highest-leverage fix wasn't a better algorithm. It was removing the team that had to ask engineering for permission to try something.