Cutting Wasted Ad Spend 48% with New-User Research

The problem: healthy dashboards, quiet users

In 2020–2021 I was Growth Marketing Manager at Upsun (formerly Platform.sh), a managed cloud application platform for developers. On the dashboards, acquisition worked: traffic arrived, accounts got created, and the paid channels hit their volume goals. Then too many of those new accounts went quiet.

Two signals disagreed. The traffic data said the ads were winning. The behavior of the people those ads brought in said we were buying the wrong visitors. Both traced to a single root cause. Nobody had asked the arrivals who they were or what they came to do.

Ask the people who just arrived

I scoped a chat cohort in Drift to users in their first two weeks, when intent and confusion are freshest, and gave them a direct line to ask us anything.

Their questions did the diagnosing. The jobs new arrivals wanted done didn't match what the platform does. The ads lacked the clarity to filter out an audience that would never get value from the platform.

The loop, start to finish

Every growth problem I take on runs the same loop. Here is how it ran for this one:

Define the metric. What mattered was the share of new signups matching our ICP and the ad spend buying everyone else.

Read real user signal. The two-week cohort surfaced who was actually arriving and what they expected the platform to do.

Form the hypothesis. Paid targeting and messaging were acquiring people whose jobs the platform was wrong for.

Brief the owning team. The in-house ads managers got an evidence-backed brief on how targeting and messaging needed to change. They implemented, and they kept ownership of their channel.

Report the result. The same cohorts and funnels that found the problem measured the fix.

What changed in two weeks

The brief's instruction fit in one line: stop paying to acquire people the product can't help. The ads team rewrote the campaigns, and two weeks later the numbers were in.

Wasted ad spend fell 48%. The savings came from clearer targeting and messaging, not from cutting budget.

Non-ICP signups fell 45%. The quality check on the spend number. The same budget was now buying arrivals whose jobs matched the platform.

The motion outlived the project. The cohort research and the reporting became standing programs, adopted by the marketing team.

Built hands-on

I was the analytics operator as well as the marketer. The cohorts, funnels, and readouts were mine, built in Pendo, Amplitude, FullStory, PostHog, Google Analytics, and BigQuery (SQL), with Drift running the in-product research.