We cut an enterprise Datadog bill from €2.1M to €340K: architecture and what broke

<p>The headline numbers are real in aggregate; the customer is anonymized. <strong>€2.1M/year</strong> commercial observability down to roughly <strong>€340K/year</strong> all-in on AWS and OSS is not magic. It is architecture plus organizational work, and the second part is where posts usually lie by omission.</p> <h2>What the architecture actually looked like</h2> <p>Metrics landed on <strong>Grafana Mimir</strong> on <strong>EKS</strong>, traces on <strong>Tempo</strong>, logs on <strong>Loki</strong>, with <strong>OpenTelemetry</strong> as the ingestion spine. S3-backed blocks and lifecycle rules mattered as much as any chart choice. The CFO-facing win was not "we installed Grafana"; it was retiring duplicate indexing paths and enforcing cardinality budgets.</p> <h2>What broke (so you can plan for it)</h2> <ul> <li><strong>Ownership gaps:</strong> teams that treated vendor agents as free suddenly owned collector configs and sampling policies.</li> <li><strong>Cardinality migration:</strong> custom metrics that were "free" in SaaS became visible in self-hosted capacity planning.</li> <li><strong>Query habits:</strong> power users who relied on vendor-specific query languages needed time in PromQL and LogQL.</li> </ul> <h2>Where to go next</h2> <p>We will publish the long-form version with diagrams, failure timelines, and the negotiation playbook for internal stakeholders. If you are modeling a similar exit, <a href="https://etalon.systems">Etalon</a> focuses on cost and operability, not slide decks.</p>