The Feature They Cut: How One App Team Chose Crash Fixes Over a New Camera

Archi4 min read

mobileproductengineeringstrategy

Contents
  1. The Meeting That Decided the Feature Cut
  2. The Numbers Behind the Crash
  3. The Partner Conversation
  4. What the Postmortem Showed
  5. The Lesson for Other Teams
  6. The Decision Framework They Built
  7. What They Would Do Differently

The Meeting That Decided the Feature Cut#

In early March, the product team at a mid-sized photo app company gathered to finalize the roadmap for the next release. The list had two big items: a new camera mode that would let users apply filters in real time, and a set of fixes for a crash that hit about 3 percent of sessions on older Android devices. The team had capacity for one. They chose the crash fixes.

That decision, according to three people involved in the planning, was not unanimous. The camera feature had been promised to a key partner. It had a marketing plan. It had a name. The crash, by contrast, was invisible to most users and hard to reproduce in the lab. But the numbers from the support team and the analytics dashboard told a different story.

The Numbers Behind the Crash#

The crash rate had been creeping up for six weeks. It was concentrated on devices with 2 GB of RAM running Android 10 and 11. When users opened the gallery after taking more than 50 photos in a session, the app would sometimes run out of memory and close. The team had seen this pattern before, but the fix required reworking how thumbnails were cached. It was unglamorous work.

The product manager pulled a report: users who experienced the crash were 40 percent less likely to open the app again within seven days. The camera feature, by contrast, had a projected engagement lift of 8 percent, based on a small beta. The math was not close. "We were looking at a retention leak that would cost us more than the feature could gain," the product manager said. "But it was still a hard call because the feature was visible and the fix was not."

The Partner Conversation#

The hardest part was telling the partner that the camera feature would slip by one release cycle. The partner had planned a co-marketing push. The team offered a compromise: they would ship a limited version of the camera feature that worked only on newer devices, and delay the full rollout. The partner agreed, but the relationship cooled for a few months.

According to one engineer, the limited release had its own cost. It required a feature flag system that the team had been meaning to build anyway. "We ended up doing more work than if we had just shipped the full feature or none of it," the engineer said. "But the crash fix went out on time, and that was the priority."

What the Postmortem Showed#

Six weeks after the release, the team reviewed the results. The crash rate on affected devices dropped from 3 percent to 0.4 percent. Seven-day retention for those users improved by 12 percent. The limited camera feature, meanwhile, had lower adoption than projected, partly because it was not available to everyone.

The team concluded that the decision was right, but the process was messy. They had no formal way to weigh invisible work against visible features. The product manager created a simple scoring sheet: each item got a score for user impact, engineering cost, and strategic value. The crash fix scored higher on user impact but lower on strategic value. The camera feature scored the opposite. The sheet did not make the decision for them, but it forced the conversation.

The Lesson for Other Teams#

The takeaway is not that crash fixes always beat features. It is that teams need a way to compare them. In this case, the crash data was available but not presented in a way that made the tradeoff obvious. The support tickets were there, but they were scattered. The analytics were there, but they were buried in a dashboard nobody checked daily.

A second lesson: the compromise (a limited release) cost more than expected. Partial features can be more expensive than full ones because they require new infrastructure. Teams should factor that into the decision.

A third lesson: the partner relationship survived, but only because the team communicated early. They told the partner about the delay two weeks before the release date, not two days. That gave the partner time to adjust its own plans.

The Decision Framework They Built#

The team now uses a simple rule: any item that affects more than 1 percent of sessions gets a dedicated review before it can be deprioritized. The review includes a support lead, an engineer, and a product manager. They look at three numbers: the percentage of sessions affected, the retention impact, and the cost to fix. If the cost to fix is less than the projected revenue loss, the fix goes to the top of the list.

That rule would have caught the crash earlier. The team estimates they could have fixed it three weeks sooner, saving about 4,000 sessions from the bad experience. Not a huge number, but enough to matter.

The camera feature eventually shipped in full, two releases later. It performed well, though not as well as the beta suggested. The team is still measuring whether the delay hurt its adoption. Early data suggests the impact was small.

What They Would Do Differently#

The product manager says she would have started the partner conversation earlier. The engineer says he would have built the feature flag system before it was needed. The support lead says she would have pushed harder to get the crash data in front of the product team sooner.

None of these are dramatic changes. They are the kind of adjustments that come from a single hard decision. The team did not fail. They made a call, lived with the consequences, and built a better process. That is the ordinary work of shipping software, and it rarely makes headlines.

More articles