How CustomInk cut their CloudWatch bill by 85% without impacting Dev or Ops

CustomInk: Where Custom Merchandise Meets Scale

CustomInk is the leading online platform for custom apparel and merchandise, serving individuals, teams, and organizations across the United States. Their engineering infrastructure handles the volume and complexity that comes with operating one of the most recognized consumer platforms in the custom merchandise space, spanning multiple services, high transaction throughput, and the kind of log density that makes observability both essential and expensive.

The Challenge: Cut Costs Without Cutting the Trail

Most observability cost problems have a straightforward solution in theory: send less data. CustomInk's problem was that "less data" carried a specific risk that made the obvious answer unacceptable.

CustomInk's security team depended on CloudWatch Log Insights to trace individual transactions through noisy log streams. When something went wrong, the ability to isolate a specific Request ID across thousands of log lines was not a nice-to-have. It was the investigation. Any approach that sampled or dropped logs aggressively enough to move the cost needle could quietly eliminate exactly the kind of granular transaction trail that security investigations required.

At the same time, the cost pressure was real. CloudWatch log spend had begun exceeding $100K annually, and New Relic costs for metrics and traces were running several times higher. As a best practice, the team had always been focused on reducing infrastructure expenditures, but they had determined that the status quo was unsustainable.

The team needed something that could distinguish between a health check that had fired ten thousand times and a unique transaction event that might matter in a security investigation three weeks from now. And it needed to do that automatically, because doing it by hand at CustomInk's log volume was not a realistic option.

According to Vít Šesták, who led the initial evaluation, "We knew the cost problem was real, but we weren't willing to trade visibility for savings. We needed something that could tell the difference between a log that matters and one that doesn't, and do it automatically. That's not something we wanted to build and maintain in-house at our volume."

The Solution: Grepr's Signal Processing Engine

CustomInk partnered with Grepr to deploy Grepr's Autonomous Telemetry Pipeline between their log sources and CloudWatch. The implementation, led on the technical side by Ondřej Trojan, was designed to reduce what reaches CloudWatch without changing how developers interact with it.

The core of how Grepr works for CustomInk:

Intelligent summarization rather than sampling. Grepr's signal processing engine identifies high-frequency, repetitive log messages, things like health checks, routine debug traces, and repeated error strings, and condenses them into unified summaries. Instead of ingesting 100 identical lines, CloudWatch receives a single summary noting that a specific message appeared 100 times. Unique, novel signals pass through without modification.

Dual-destination architecture. Compressed log summaries are forwarded to CustomInk's existing CloudWatch endpoints, preserving the dashboards and workflows developers already use. Simultaneously, full-fidelity raw logs are written to CustomInk's own S3-based data lake, queryable via Grepr. Engineers retain complete access to raw data when they need it for deep investigations, without paying CloudWatch prices to store everything there.

Granular transaction trail preservation. To address CustomInk's security investigation requirements, the pipeline was configured to capture unique Request IDs within log summaries. This preserved the ability to search and isolate individual user impacts across noisy log streams, exactly the kind of trace capability the security team depended on.

Flexible log routing for sensitive data. For log types requiring special handling, Grepr routed those streams to a separate destination, allowing CustomInk to meet their data handling requirements without any changes to how those logs were instrumented or generated.

The technical implementation included SSO integration with CustomInk's internal identity systems, enabling automated role mapping and secure production access without requiring a separate credential management process.

As Ondřej Trojan noted after the production rollout, "The configuration process was more straightforward than I expected. We had the pipeline running in our dev environment quickly, and moving to production was a matter of validating the output and adjusting a few thresholds. The SSO integration worked cleanly with our existing setup, which removed a lot of the friction we anticipated around access management."

The Results

CustomInk achieved an 85% reduction in both log volume and CloudWatch log costs, cutting CloudWatch spend from roughly $100K to $15K at standard AWS list pricing.

Developer workflows remained unchanged. CloudWatch Log Insights continued working exactly as before. Dashboards kept populating. Security investigations retained the transaction-level granularity the team required. The cost came down without the operational disruption that had made the team cautious about previous cost-reduction approaches.

As Vít put it, "We were skeptical that something could reduce our log volume that aggressively without us feeling it somewhere. The fact that our developers haven't noticed a change in how they work is the result we were hoping for. We knew the savings potential was real. That part wasn't hard to reason through. The workflow stability was what we needed to prove."

What's Next

With CloudWatch log costs under control, CustomInk is continuing to evaluate how Grepr can be applied to additional telemetry streams, including the New Relic metrics and traces spend that remains a significant cost driver. They have also signed up as a design partner to shape our AI-powered anomaly detection capability.

Want to see what Grepr can do for your observability bill? Get started free and see results in under 30 minutes.

‍