Making post-launch sentiment something teams could act on

Context

In September 2025, Goodnotes shipped a major update. One of the key goals was to grow our desktop user base.

The launch did not land as expected, and public backlash followed. App Store ratings across all devices fell from 4.7 to 4.2, and desktop did not grow. The team did not have a clear account of why the launch had gone so poorly, and leadership needed to know what to fix.

While product squads focused on their own areas, the CEO pulled a cross-functional group to audit the app as a whole. QA took bugs and stability, Design took how features connected to each other, and I focused on why users were losing trust.

Outcomes and impact

The audit helped set the picture for company-wide quality work through Q4 2025 and Q1 2026. It uncovered five key focus areas. Product teams used that picture to prioritise stability, improvements, and raising the quality bar, which set direction for several months.

In the months of quality work that followed, App Store ratings across all devices climbed from 4.2 to 4.65 within three months, and within six months they were tracking at around 4.74, above the rating before the September launch.

The desktop user base had not grown with the launch. After that quality work, DAU increased more than 20% from launch, with double-figure growth in subscriptions.

Approach

There was a lot of pressure to put something usable in front of the team quickly. The backlash was hitting business metrics, and the Q1 2026 roadmap was already being planned. A tight timeline meant research would primarily use existing data.

Build a system to audit sentiment

On user sentiment, we had tens of thousands of data points across sources, too many for individual synthesis. I built a custom workflow to make that signal usable.

Using AI, the workflow categorised incoming sources (support tickets, App Store reviews, social posts) and clustered similar items to surface potential themes. Those themes were the basis for AI-assisted human synthesis, which got us to a first pass.

We validated that pass with human spot-checks on the key themes, and by checking the model's labels against how Support actually replied. CS keeps pre-written replies tied to known issues; if a reply used a given template, the ticket was probably that issue.

I then ran a workshop with user-facing teams to stress-test the themes. Framing it around prioritisation gave us another lens on the analysis: we cut lesser issues and built a prioritisation table for the top ones.

Follow new Mac users from first visit to first note

The pre-audit desktop user base was too small for the sentiment work to explain why desktop had not grown, so I looked at our telemetry data instead. First-time drop-off was far higher on desktop than on tablet, which meant most people never got far enough for the product to matter.

The working hypothesis became that first-time use was the adoption blocker: if that conversion improved, broader desktop adoption would follow.

I ran eight moderated sessions with professional users trying Goodnotes Mac for the first time, from first discovering the app to making their first notes. From those sessions I built an emotional journey map: where curiosity turned into confusion, where confidence dropped, and where trust was lost before they had started.

The sharpest drops sat around signup, the empty state, first feature discovery, and the paywall.

Bringing the findings together

The last job was to turn three streams into one audit the company could actually use. We framed the report around the user-centred research findings, and tied QA’s bug prioritisation and Design’s work to that picture, so bugs and broken connections sat against why people were losing trust.

We shared the findings at two altitudes: a company-wide session for the high-level story, then tighter sessions in each product team’s weekly stand-up, so each team heard the part they could act on.

Learnings

Align the lens before kickoff

The de facto plan for this audit was QA-only. Before kickoff, I worked with Design to reframe what leadership wanted from the project toward a user lens. Without that, the audit would have been a defect list. That kept why users were losing trust in the brief, and made the outputs usable beyond bugs.

Put AI on the critical path

This was the first time I built a for-purpose, reusable synthesis tool and used it as a core method, rather than an assistant. The work was the guardrails: human spot-checks on the analysis, independent auditing loops on the model, and AI as a step in the synthesis, not the final output.