Brand Metrics had hit an architectural ceiling. I rebuilt the platform for the AI era, and it shipped as an industry first.

The situation

Brand Metrics measures whether advertising actually builds brands, and sells that service to publishers and advertisers across four global markets. When I joined, the company was targeting 30%+ year-on-year growth, and the platform wasn't keeping up.

The underlying system was one large codebase that slowed under heavy data loads, couldn't connect to third-party tools, and got harder to change with every release.

Customers across four markets needed integrations, faster setup, and better reporting, but the platform couldn't reliably deliver any of it, and it had been designed around publishers, the segment it grew up serving. Networks, media buyers, and other parts of the ad ecosystem were operating around the edges without a product designed for them.

What I found

I ran 20+ customer interviews across US, EMEA, APAC, and LATAM while Engineering dug into what was technically wrong. The same complaints came up everywhere: setup took too long, connecting to other systems wasn't possible, and the reporting wasn't detailed enough.

"The feature requests weren't about missing features. Customers were describing what the platform couldn't do."

"Setup takes too long" described a slow system, not a bad interface, and "we need integrations" meant the platform had never been built to connect to anything external. Together they pointed to structural problems, not surface-level ones, which changed everything we needed to build.

The approach

Working directly with Engineering and Data Science, we decided to rebuild the foundations while continuing to ship, replacing each part of the old system one at a time and proving the new approach worked before moving to the next.

We rebuilt the data layer to handle ten times the volume, opened the API to third-party connections for the first time, and broke the platform into independent pieces so teams could work without blocking each other.

For major decisions about how the new system should work, I built working prototypes instead of writing specifications. When we needed to decide how campaigns should be structured in the new system, I had a working version in two hours using Claude Code and Lovable, walked Engineering through it, and we reached a decision in two days instead of the three weeks of debate that had preceded it.

The outcomes

Opening up the platform to third-party integrations unlocked two customer segments that had been structurally out of reach: mid-market buyers and in-house teams. Customer count doubled from 40 to 80 during the rebuild. The AI Survey Question Generator shipped as a paid add-on with fast adoption, which justified the rebuild commercially and added a new revenue stream.

Engineers could ship faster again, working in parallel without blocking each other, and the platform could now handle the data volumes and integration demands that had previously been out of reach.

"The product trio now makes feature and architecture decisions together, connected to real customer evidence. That operating model didn't exist before."

What this taught me

When the platform stops delivering customer value, the PM can't hand it to engineering and move on, because every release that skips it makes the system harder to change. I treated architecture decisions as product decisions, and it changed how quickly we could move.

Showing a working prototype instead of writing a specification changed how fast we made decisions: engineers could see the idea, spot problems, and commit to an approach in days rather than the weeks of debate that preceded it.