How Class Technologies slashed their observability costs by 80% by consolidating on Sentry

Class is the virtual classroom built on Zoom and Microsoft Teams. Class adds tools like engagement tracking, analytics, and enhanced breakout rooms to provide learners with a better and more active experience for virtual training and learning.

As Michal Marek, Staff Software Engineer explains, “We can’t afford for the Class app to not work on any device. Many of our users are students with low-end Chromebooks provided by their schools. If it doesn’t work for them, they can’t just go buy better devices.”

For a company serving educators and students on everything from state-of-the-art laptops to decade-old Chromebooks, reliability and fast performance is mission-critical. Teachers don’t have the time or resources to debug mystery issues — so when students say something isn’t working, the only option becomes to move the class elsewhere. Class has to know about issues before users are reporting them.

The challenge: Unconstrained observability costs and complex tools

From the start, Class prioritized observability to ensure great user experiences. But as they grew—launching new platforms and acquiring two companies—they inherited a patchwork of tools, including Datadog, New Relic, and CloudWatch.

This fragmented setup led to skyrocketing costs. Log ingestion alone accounted for 80% of their spend. Worse, surprise overages tripled invoices due to unexpected log spikes.

“After a few acquisitions, we were supporting five platforms and monitoring costs were spiraling. It was hard to trace issues across all these platforms. At one point, 80% of our bill came from log ingestion we weren’t even using.” said Brook Ballard, Enterprise architect

Key challenges included:

• Soaring costs: Monitoring expenses exceeded $1M annually.
• Tool overlap: Five monitoring tools created redundancy and inefficiency.
• Inconsistent data: Each tool had its own quirks, making cross-platform monitoring frustrating and manual to find relevant issues

“The other observability tools we were using made it difficult—especially for less technical users—to surface relevant errors quickly. The high bills coupled with an unintuitive platform had us looking for opportunities to divest and consolidate.”

The Solution: Consolidation on Sentry– getting to answers with breadcrumbs, not bulk logs

Class chose to consolidate on Sentry for its easy setup across their core platforms, including JavaScript and React for browser frontends and web apps, .NET for Windows, iOS, and Android.

Right out of the box, they saw a decrease in noise. Sentry surfaced critical issues with actionable debugging context (such as the affected user, device, and app version) enabling the team to actually see problems they could take action on without unnecessary manual work.

“Sentry was one of the few tools that worked across all platforms from day one. Plus, its UI made it easy to find the information we needed,” said Brook

Additionally, Sentry offered clear, predictable pricing—a major relief after struggling with unexpected APM vendor bills.

“Sentry helps us control spend with spike protection and quota alerts, so we can scale usage up or down as needed. If we exceed our quota, we can figure out exactly what went wrong and work with the Sentry team to adjust accordingly. We didn’t have that experience with other tools, where overages can really add up quickly without an easy ability to understand why” said Brook.

Class reduced their log ingestion significantly by shifting their approach to focus more on errors. Instead of ingesting millions of logs across platforms, which ballooned costs without clear insights, they used Sentry’s breadcrumbs to capture the last 100–200 log statements tied to an issue. This created self-contained error reports with all the necessary context to debug efficiently—without overwhelming their system with unnecessary data.

To further optimize, Class customized sampling rates with Sentry to focus on the data that helped them most. Known low-impact errors are sampled at 10%, for example, while critical errors and unknown errors are logged at 100% to help them eliminate big/new issues as quickly as possible.

“We’ve tailored our sampling rates to prioritize high-impact errors. That way, we can focus on problems that actually affect users,” shared Michal Marek, Staff Software Engineer.

Brook highlighted how using Sentry changed their workflow: “Sentry gives us the context we need without the overhead of ingesting every single log. It’s the relevant part of the log we care about anyway.”

This shift helped Class’s engineering team triage faster, lower noise, and avoid the spiraling costs that came from Datadog’s bulk log ingestion.

With Sentry, Class’s engineering team:

• Reduced noise with issue grouping to find and prioritize the issues that mattered—without manual searching.
• Resolved issues faster with clear stack traces, breadcrumbs, and custom tags.
• Optimized costs with predictable pricing, spike protection, and quota alerts—cutting log ingestion by nearly 90%.

Slow is broken—staying ahead of performance issues

For Class, even a single bad user experience can lead to churn. In classrooms, technical issues don’t just affect individual students—they can derail entire lessons.

Teachers don’t have the time or resources to debug issues; if a handful of students suddenly complain that the platform is “broken” without any clear cause, the teacher’s only real option is to move the class somewhere else.

As Brook explained, “If one or two students can’t connect, teachers often move the whole class back to Zoom. If that happens enough, we risk losing the entire account.”

The challenge is made even tougher by the wide range of devices their app runs on—from high-end Macs to outdated Chromebooks. Michal Marek emphasized: “Performance is our number one priority because we can’t just tell students to upgrade their hardware.”

For Class, Performance monitoring isn’t just about latency—it’s about fixing any perceived issue impacting their users.

The team sends extra tags and data alongside performance events, such as GPU acceleration status and CPU architecture, to capture diagnostics that make debugging easier — especially when trying to reproduce arcane issues. As Michal explained: “We collect custom tags to understand our users’ audio and visual experiences, but that sometimes requires case-by-case investigations when only a few users encounter an issue.”

With Sentry’s custom tags and aggregated insights, they can easily correlate common factors among affected users—whether it’s external headsets, outdated hardware, or GPU configurations—to pinpoint the root cause.

“Sentry’s aggregated insights also let us quickly see if an issue affects 30% of users or just a handful,” says Brook, “with tags that reveal whether GPU acceleration or another factor is involved—making it easier to resolve even the most unique edge cases.”

It also helps Class’ dev team better understand when performance issues are outside of their control, so they can better prioritize their time and attention:

“If someone reports that Class is slow, we can dive into their transactions and show that, for example, nine out of ten seconds were due to Zoom’s connection process—not our app.”

Reducing time to triage by 40% with custom context

Last, having all this relevant context in one tool helps developers zero in on the root cause quickly, dramatically reducing their average time to resolution.

For example, after deploying a release that introduced a new integration, Sentry immediately alerted them of an issue that only rears its head the very first time a user signs in — the kind of thing that’s very easy to miss when you’re testing in a sandbox instead of looking at real world data. By reviewing the stack trace and breadcrumbs, Brook’s team pinpointed the exact line of code causing the issue. Brook explained, “The stack trace showed exactly where it failed, so we were able to fix it before users even noticed.” This saved hours of manual investigation and prevented escalations.

Between Sentry’s issue grouping, context including custom tags, and GitHub integration that connects to source maps and CODEOWNERS file to tell you who pushed the broken code and where it broke — Class has reduced their time to triage issues by 40% on average.

The Results

By consolidating on Sentry, Class has been able to proactively fix issues and take action faster, delivering a more reliable user- and developer- experience.

Class plans to deepen their use of Sentry by implementing features like Session Replay and AI-assisted debugging with Autofix.

“Sentry has made us proactive instead of reactive. We catch issues early, save money, and deliver a better product.”

The results:

1. Cost Savings:
   • Cut spend on other observability platforms by 80%
   • Overall Monitoring Savings: Hundreds of thousands annually.
2. Improved User Experience:
   • 40% decrease in time to resolution: Developers quickly identified and fixed issues like first-time Google sign-in failures, avoiding hundreds of potential user complaints.
   • Improved User Experience: Ensured stability on all devices, from high-end laptops to 10-year-old Chromebooks.
3. Simplified Workflows:
   • A single tool for monitoring reduced redundancy and frustration.
   • Non-technical teams easily adopted Sentry, enhancing triage efficiency.