A Digital Health Platform Removed Sampling and Added MCP to Change its Engineering Culture

The Stakes

This company is a leader in the patient-focused health data space, widely recognized by industry analysts as one of the top platforms in its category. It is unique because every feature the company builds to gather and disseminate health data starts with patient consent in mind. That's critical whether patient data manifests in the company's relationship with wearable device manufacturers or in its services to frontier AI labs. The company needed an observability partner that could meet its scale and privacy requirements, so it chose groundcover.

"I don't believe we could be executing on some initiatives - or at least executing as efficiently - without access to the groundcover MCP" - SRE Lead, Digital Health Platform

The Challenge

In 2025, when this company's site reliability lead looked at the existing stack - Datadog for metrics and APM, and self-hosted Grafana for logs management - he realized there was a critical gap in his ability to deliver on his vision for the SRE team and the observability ecosystem they owned. He wanted the wider technical staff to be fluent in observability and for it to be economically feasible to share access directly to the telemetry data each member used to derive insight. "We are a company for health data, we culturally are built on data. And having that (data) available is one of the most important things that makes us effective."

On top of that, when trouble arose, the SRE lead didn't want the team to share screenshots of the patterns they were seeing in their observability data - he wanted them to be able to share access to the dashboards themselves. "There's a certain amount of evangelism required where I had to say to people 'please don't send a screenshot, send a link.'"

Problematically, on Datadog, the company was already sampling down to 10% of DEV data and 25% of PROD data. They were missing the telemetry data they needed if they were going to enjoy the value from that sharing pattern, because the math for SaaS observability didn't work.

Were they to consider OSS Grafana or ClickHouse, they had the painful choice to try to manage their own stack: "ClickHouse can obviously do the scale, but I can only imagine the headache of doing it myself." Besides building everything themselves, the team didn't have a sustainable answer for how to manage compliance requirements like HIPAA.

The company also has its own MCP service, which is the foundation of the AI SDK for its health application consumers. Its customers, therefore, require it to serve data for both traditional and agentic applications. The implication is the company can expect its consumption to grow, and its observability requirements to scale, as more agentic health applications come to market.

The Implementation

The company moved logs, traces, and metrics off its other vendors to groundcover, hosted on its own AWS instance via the bring your own cloud (BYOC) deployment model. Nearly all of its workloads include sensitive PII and PHI, so that private cloud implementation gave the team the benefits of a SaaS-like experience with none of the privacy gaps: "We have to be HIPAA-compliant and to be very, very, very strict about who we send data to and why. With the BYOC model, we're not sending data to you, we're sending it to ourselves. And so the data stays within our borders. Our provenance." The team also appreciated that the implementation didn't require a completely bespoke solution. They ruled out other platforms which they considered "almost too white glove."

With their observability implementation no longer dependent on a vendor that charged on data volume, the company could remove its sampling requirements. Additionally, every time the team deploys a new application, they get immediate visibility into at least a portion of their ecosystem with the groundcover eBPF sensor. When describing the installation and subsequent continuous use, the SRE lead said, "Ease of implementation was a part of the reason we chose groundcover. The eBPF sensor helps. You get a pretty good signal just by deploying the sensor and without having to tell your services to start forwarding OTel."

What was once a burdensome process of instrumenting before getting any value now automatically and immediately benefits the SRE team with the groundcover implementation.

The Delta & The Proof

Full fidelity becomes the default, not the exception

The delta: Once this company's observability spend was no longer tied to data volume, the team removed sampling entirely. Every log, metric, and trace, including every trace span, now gets ingested, with no tier or threshold governing what gets kept.

The proof: "We don't sample anything. We ingest 100% of the standards — logs, metrics, traces. All of our trace spans. We do not sample," the SRE lead says. "For a given signal or trace, there's no question of whether it's going to be present. We know it's going to be present. If it's not, it didn't happen." The team also leverages extended retention: the company keeps roughly 13 months of full-fidelity data in groundcover.

MCP becomes assumed infrastructure

The delta: Access to groundcover data through MCP has stopped being a specialized tool that the SRE team reaches for — it's become infrastructure every developer at the company is assumed to have. Internally, the team has shifted away from static runbooks toward a library of Claude skills, and a growing share of them route directly into groundcover for whatever data they need.

The proof: "I don't believe we could be executing on some initiatives — or at least executing as efficiently — without access to the groundcover MCP," the SRE lead says. "It's assumed every developer is hooked into the MCP and that it's just available. It's become a ubiquitous tool that everyone has access to." That shift showed up concretely when the company debated whether to grant developers direct read access to production Kubernetes clusters. The answer was no: "What can they pull there that they can't pull out of groundcover?"

A link-driven culture, without the price tag

The delta: Sharing observability data at this company used to mean sending a screenshot — a single frozen view that couldn't be explored further. With full-fidelity data now available to everyone, sharing a live link to a dashboard or query became the default, letting anyone pick up an investigation and dig deeper rather than take someone else's word for it.

The proof: "We are certainly far more observability-focused than we were under our Datadog and Grafana setup," the SRE lead says. "Several months after the deployment, I started seeing people passing around links to groundcover — dashboards, queries they'd written — instead of just saying 'I noticed this.' Everyone can see exactly what I did, the exact same view." That habit now extends into how the company makes decisions. Ongoing discussions about user-specific rate limiting, and plans to run structured before-and-after comparisons for infrastructure changes, are both grounded in data pulled directly from groundcover. It has elevated observability's strategic role in the broader organization.

The complete telemetry data ingestion, the expanded MCP usage, and the shift to a link-driven culture weren't feasible for this company when it was still sampling to stay under a limit for its Datadog bill. In an earlier phase of this partnership, moving to a usage model built around BYOC and eBPF let the team stabilize observability spend under $10,000 a month (down from $15,000–$30,000) while ingesting roughly 10x more application data than before. That earlier cost transformation is what makes everything above possible without a second look at the bill.

Why This Digital Health Platform Chose groundcover

The company originally evaluated other observability solutions, but struggled to find a suitable replacement that could both satisfy its cost and privacy requirements and was easy to deploy and maintain. "Cost was a very big driver," the SRE lead says. "It was also, to an extent, ease of implementation." For example, after looking at Grafana Cloud and one other vendor, the second vendor stood out for the wrong reason: "That whole thing was a turn-off," the SRE lead says. "I think they could do OpenTelemetry, but it was a bit more (complex) - especially for the instrumentation. It might have been a custom SDK that we would have had to put in."

The sales process itself became a signal. The team found some vendors rigid and others swung too far the other way, layering on so much hands-on support that it read as a warning sign. "They almost seemed a little too handholdy, too white glove, which actually signals to us that your product is too complicated to understand," the SRE lead says. "We're a very technical, hands-on team. We prefer to do it ourselves rather than being white-gloved through something." That preference shaped how the evaluation played out internally. The SRE lead ran the initial technical assessment himself before looping in the wider engineering org for a formal POC.

"groundcover has consistently shown how they're willing to work with, partner, and grow with us as a business. I don't think our feature velocity and time to market would be as fast as it is without groundcover." - SRE Lead, Digital Health Platform

The ClickHouse question came up again during evaluation, this time framed by a direct comparison: the company also looked at another ClickHouse-backed platform. It wasn't close. "I never wanted a self-managed ClickHouse — ever, ever, ever," the SRE lead says. "It's a wonderful product, but it's complicated as hell to manage and run. And I never want to do that. You all do that for us. You have that expertise." He frames the decision as a matter of engineering discipline: "Don't build what you can buy. Focus your effort on where you provide value, and pay someone for the rest of it. I want to build dashboards and enable developers to get the data they need - not run ClickHouse clusters." That tradeoff extends to retention: the company keeps roughly 13 months of full-fidelity data on groundcover's ClickHouse backend, without operating any of the infrastructure that makes that retention window possible. The SRE lead says, "As you scale and you're going to consider ClickHouse, it would never make sense to go ahead without looking at groundcover."

Ultimately, the clearest proof of groundcover's value came when the company used it to solve a scalability problem of its own. Late last year, the team was preparing to onboard a large new customer, and scalability suddenly became a first-order concern for an organization that hadn't put much strategic weight on it before. The SRE lead led a three-to-four-week push to get the company's own customer-facing MCP endpoint ready for the load - identifying metrics, building dashboards, and closing infrastructure gaps as they surfaced: ingress controllers that wouldn't scale, then the application layer, then a MongoDB backend that started falling over under pressure. Each fix was driven directly by groundcover data. "That would not have been possible under the Datadog/Grafana setup," the SRE lead says. "I can say 100%." The company hit its scalability targets and onboarded the customer with confidence - proof that groundcover wasn't just cheaper or easier to deploy, but a tool the team could lean on to solve a real, time-boxed business problem under pressure.

When asked to describe what groundcover has helped the company accomplish in its time as a customer, the SRE lead closed with, "groundcover has consistently shown how they're willing to work with, partner, and grow with us as a business. I don't think our feature velocity and our time to market would be as fast as it is without groundcover."

Are you ready to see in as few as 30 minutes how full-stack and 0% sampled observability can help your team get to market more quickly?

Book a demo to see how groundcover delivers observability in your own cloud - without per-query or spiking unpredictable pricing.