Dubber cuts observability costs by 25% with Grafana Cloud

Dubber has overhauled its observability setup by moving to Grafana Cloud, cutting run costs by 25%.

Sai Haritha Rampe, head of platform engineering at Dubber, said the company replaced a fragmented mix of self-hosted and managed tools with a single managed service for metrics, logs and monitoring.

Dubber, a Melbourne-headquartered call capture company with operations across several regions, records more than one million calls a day through a network of 215 partners and resellers. It serves sectors including healthcare, retail, finance and government, and also offers conversation intelligence tools based on its own AI systems.

The previous setup combined self-hosted Grafana and Kibana with AWS OpenSearch and AWS Managed Grafana. For a lean platform team that also oversees service operations, cyber and security practices, delivery tooling and first-line incident response, that mix created operational strain.

The older approach also brought heavy maintenance work, scaling challenges, security patching demands, alert fatigue and frequent context switching between consoles. As data volumes grew, the observability estate became harder to manage.

Why it changed

A key factor in choosing Grafana Cloud was familiarity within the engineering team. Staff were already used to the Grafana interface, reducing the training and internal change management needed during migration.

The managed service model was another draw. It removed the need for Dubber's platform team to spend time on patching, vulnerability management and capacity expansion for observability systems.

The ability to manage the environment through infrastructure as code also influenced the decision. Dubber uses Terraform widely, and available modules helped keep observability practices aligned with the rest of its engineering standards.

The platform's multi-stack design also suited a company with global operations and data sovereignty requirements. Role-based access controls were another factor, allowing access to be assigned by job function.

Migration path

Dubber began by reviewing available features against its existing estate, then moved to what Rampe described as a proof-of-value exercise. From there, it started migration work before shifting into optimisation.

The rollout began with metrics and logs. Dubber initially ingested more than one million metric series, making cost control and data filtering increasingly important as volumes rose.

Adaptive Metrics became central to that effort. Rampe said it helped Dubber reduce metric noise, manage cardinality and cut the manual work needed to determine which data should be retained.

"By moving into Grafana Cloud, we've pretty much reduced our run costs on observability by 25%," Rampe said.

Operational view

Dubber uses Kubernetes extensively across Azure and AWS, and Kubernetes Monitoring has become one of the main operational dashboards for the platform team. It provides an overview of clusters, nodes, pods and containers, and supports alerts for events such as node failures and crash loops.

Those dashboards also inform cloud cost management. Performance data from Kubernetes clusters helps guide capacity planning and cloud spending decisions.

One example involved CoreDNS monitoring. A prebuilt dashboard helped the company identify a production issue linked to limits on DNS hits across the platform, allowing the team to adjust internal DNS queries.

Dubber has also added custom integrations, including AWS CloudWatch for monitoring managed Kafka on AWS and a Temporal integration for workflow visibility. Both were adopted through prebuilt dashboards and community integrations rather than bespoke development.

Synthetic checks

Dubber uses Synthetic Monitoring for browser-based checks on its web portals. Probes run every five minutes to confirm customer-facing sites remain available, a priority for a business where portal outages can trigger high-severity incidents.

The addition of secrets management also improved how Dubber handled usernames and passwords in synthetic scripts, replacing hard-coded credentials with a more controlled method.

Beyond infrastructure and site checks, Dubber has built dashboards to monitor AI token usage across models and regions. The data is used to compare consumption across providers and support planning around usage and supplier costs.

The company also tracks call capture data through dashboards that flag blank calls, which may indicate recording issues or unwanted noise in the system.

Next steps

Dubber is also migrating to Grafana IRM as part of a broader update to its incident response tooling. Rampe said the observability programme continues to expand across more of the platform and application stack.

"What I loved about Adaptive Metrics is that you can undo it," Rampe said. "For example, you might apply an adaptive metric, then after a month or two someone from the engineering team says, 'Hey, Sai, I need that metric because I want to start creating a dashboard for it.' It was so easy to reverse, and now there's a Terraform module for it as well, where you can put all the exemptions for an adaptive metric into Terraform and apply them there. I think that's the beauty of it."