b.well Debugs AI-Scale Healthcare Workloads Without Sampling or Surprise Bills

About b.well

b.well Connected Health is the most data-rich digital health platform for AI-powered consumer experiences, personalized care, and better outcomes. Built on a scalable, FHIR-based foundation, b.well unifies fragmented healthcare data into longitudinal health records, enabling real-time engagement, proactive insights, and regulatory compliance.

Headquartered in Baltimore, Maryland, with 100 plus employees, b.well powers large-scale, privacy-sensitive healthcare workloads and serves as the healthcare backbone for a leading frontier AI company’s flagship product .

The situation engineers were actually in

In the weeks leading up to the AI Health backbone launch, b.well was running large-scale, multi-region performance tests for workloads they had never operated at before. They needed to answer basic but high-stakes questions fast:

• Why are requests stalling in one region but not another?
• Is this a Kubernetes scaling issue or an application bottleneck?
• Are we dropping signals right when we need them most?

At the time, their observability stack made those questions harder to answer, not easier. Datadog costs forced heavy sampling, while logs and infrastructure metrics lived in a separate Grafana setup. Critical context was fragmented, and engineers were spending time managing telemetry instead of using it.

Why they replaced Datadog and Grafana

Under Datadog, b.well sampled production traces down to 25 percent and non-production to 5 percent. Logs and infrastructure data were not sent at all. Cost controls drove observability decisions.

After moving to groundcover, the team disabled trace sampling entirely and began sending all telemetry by default.

That shift changed how engineers worked. Instead of debating what data to drop, teams could follow a request end to end across services, pods, and nodes. During performance testing, they identified application-level behavior adding a 30-second overhead and Kubernetes autoscaling issues that were not responding quickly enough to traffic spikes.

Those issues were visible only because traces, metrics, and logs were correlated in one place.

One system, full context

groundcover gave b.well a unified view across Kubernetes, infrastructure, and application layers. Even when pods no longer existed, engineers could still inspect their metrics and logs. This made post-incident forensics and performance tuning deterministic instead of hypothesis-driven.

Instead of jumping between tools, engineers could start from a slow request and drill down to the exact infrastructure behavior causing it.

“I don’t think we ever would have found that issue if we weren’t able to visualize the traces the way we were with groundcover.” 

Cost predictability changed behavior, not just budgets

With Datadog, observability spend fluctuated between $15k and $30k per month, which led to constant monitoring of ingestion and metric usage. With groundcover’s BYOC model, b.well stabilized observability costs under $10k per month while ingesting roughly 10x more application data.

For developers, this removed an entire class of friction. Logging more data was encouraged instead of discouraged. Sampling strategy discussions disappeared. Engineers stopped thinking about cost impact mid-debug session.

“With groundcover, we don’t really have cost concerns.”

Observability became a shared tool, not an infra-only one

During the launch, groundcover usage spread beyond a small infrastructure group. Engineers and product teams shared links to traces and dashboards directly when reporting issues. Alerts were no longer ignored. When something fired, people looked at it, trusted it, and acted on it. groundcover became the default place engineers opened when something felt off.

Built for healthcare privacy constraints

Because groundcover runs inside b.well’s own cloud account, sensitive telemetry never leaves their environment. This simplified compliance audits and reduced third-party risk, while still allowing full-fidelity observability.

Results engineers care about

• Zero trace sampling, full-fidelity logs, metrics, and traces
• One observability system instead of Datadog plus Grafana
• Faster performance tuning and incident debugging under real load
• Predictable costs that removed instrumentation anxiety
• Organization-wide adoption during a critical product launch

groundcover shifted observability at b.well from something engineers worked around into something they relied on when it mattered most.

Try it yourself for free

Want to see how teams like b.well achieve full observability without unpredictable costs?
Try groundcover for free today.