Watchtower & Magnifying Glass.
A mixed-methods study that uncovered a shift in how cloud teams want to monitor their systems, and reshaped two product roadmaps.
TL;DR
Google Cloud Monitoring is the tool teams use to keep their online systems healthy. Product managers assumed top customers were satisfied with its core jobs. I led a 4-person team pairing a large-scale survey with in-depth interviews across major companies. Segmenting the results by role surfaced two very different users, infrastructure operators (the Watchtower) and application developers (the Magnifying Glass), and a deeper shift beneath both: people no longer want to actively watch dashboards, they want to be alerted when something needs attention. The study drove two roadmap changes: shift investment from dashboards toward alerting, and unify two tools that had lived apart, metrics and logs.
Systems map
How a sentiment study became a strategy reset
On the surface, the data said customers were “mostly happy.” But looking only at the aggregate sentiment would have missed the real story.
By segmenting the data by role, I uncovered two distinct users with entirely different goals:
- Infrastructure operators: Needed a wide, system-wide view to keep everything healthy (“Watchtower”).
- Application developers: Needed to pinpoint problems in their own code fast (“Magnifying Glass”).
Underneath both groups was a shared shift: people didn't want to actively monitor anymore. This insight triggered a major product strategy reset, bringing two separate product teams together to build one unified experience.
In troubleshooting, the tools are strong at telling people something is wrong, then fall away exactly where the work gets hard. Confidence fell off sharply between spotting an issue and actually finding the root cause and fixing it. Both groups wanted one click from a spike on a chart to the logs that explain it.
What I almost missed
What I almost missed
The first cut of the analysis ran the survey data in aggregate (the standard move), and it told a story stakeholders would have accepted: sentiment slightly positive, no major issues. I almost shipped it that way. What stopped me was a falsification habit: before any readout, I check whether the same data, segmented differently, tells a different story. Splitting by role wasn't an obvious cut; it required treating role as a primary axis rather than a demographic. When I did it, the aggregate dissolved. Operators and developers weren't using the same product. The “users are happy” story would have been technically true and strategically useless. Segmentation is where the senior judgment lives in this kind of study.
“I don't want to stare at this dashboard. I want it to tell me when I need to look.”