Decide what to measure, which tools to keep and what reliability to aim for
A target state for observability your teams share: which telemetry each service emits, the pipelines and tools that carry it, the service level objectives that say how reliable each service must be, and a cost policy that keeps what you must keep and stops paying to store data that is never read.
Telemetry decisions get made by default
Nobody decides to run several monitoring tools or to keep every debug log for a year. It happens because each team chose for itself, retention was set once and never revisited, and the bill sat in an infrastructure line without an owner. The result is coverage that follows history rather than risk, and alerting set on thresholds someone guessed. A strategy makes those choices on purpose: what each service must emit, who owns each signal, how reliable each service needs to be, and what the telemetry is allowed to cost.
Signs it is time for a strategy
If two of these are true, the next tool decision needs a target state behind it.
Each team chose its own tools
Operations, development and security each run their own monitoring, and nobody owns the whole picture.
A renewal or migration is due
A contract ends or a platform moves, and the choice of tool will set the next few years of telemetry.
Moving to cloud-native platforms
Containers and managed services arrive faster than the instrumentation standards that should come with them.
AI is heading to production
Models and agents are about to answer customers, and nobody has decided how they will be traced.
No agreed reliability targets
Nobody can say whether a service is good enough, so reliability work goes to whoever shouts loudest.
Telemetry spend has no owner
The bill grows with every service, and no one decides what data is worth keeping.
What the strategy decides
Architecture
- A target architecture for collecting, storing and analysing logs, metrics and traces
- One collection layer on open standards, so each server runs one agent and the telemetry stays portable
- Pipelines that filter, sample and route data before it reaches paid storage
Tooling
- Keep, consolidate or retire, decided per tool from what it covers and what it costs
- Instrumentation standards new services must meet before they go live
- The order in which tools are consolidated
Reliability
- Service level objectives for the services that matter most, agreed with the business owners
- Alerting on how fast the error budget is being spent, so people are paged for what customers feel
- Ownership: who answers for each service, each signal and each alert
Cost and retention
- Retention set by type of data and by obligation, with hot and cold storage tiers
- A floor set by your obligations, such as the ISM's twelve months of searchable logs
- A named owner for telemetry spend and a way to show each team what its data costs
Security telemetry
- The signals your security operations centre needs, and where each comes from
- One collection path shared by operations and security instead of the same data collected twice
- Retention for security logs set to what your obligations require
AI workloads
- Tracing for prompts, retrieval, model calls and tool calls
- Latency, token cost and answer quality measured per request
- Prompt capture off by default, switched on deliberately where privacy and purpose allow
How the strategy is set
Three steps, with your engineering, operations, security and finance leads.
- 01
Start from evidence
We work from the assessment's findings on coverage, tools, alerting, reliability and spend, or confirm the picture you already hold.
- 02
Decide
We work through architecture, tooling, objectives, retention and cost with the people who will run the result.
- 03
Write it down
Each decision is recorded with its reasons, so the implementation can be scoped and funded from it.
Set the target before the next tool decision
Scope and fee are fixed at a free scoping conversation.
What you leave with
Decisions
- A target architecture for telemetry
- Tooling decisions, with the reasons behind each one
- Service level objectives for your most important services
- A retention and cost policy that protects what must be kept
Standards
- Instrumentation standards for new and existing services
- Ownership and alerting principles
- An approach for tracing AI models and agents
- Decisions written so the implementation can be scoped from them
Reliability targets the business can agree to
A service level objective is a target such as how many requests may fail, or how slow they may be, over a period. The gap between the target and perfection is the error budget. When the budget is being spent quickly, someone is paged; when it is healthy, the team ships. That replaces alerts on CPU and disk with alerts on what customers feel, and it gives the business and engineering one number to argue about. For a regulated entity, objectives on critical operations are also how CPS 230 tolerance levels become something you can monitor.
AI observability is governance evidence
The Guidance for AI Adoption asks organisations to monitor how their AI systems behave, and APRA has told regulated entities it expects that monitoring to be continuous. Traces of each prompt, model call and tool call, with latency, token cost and quality checks, are that evidence, produced as the system runs. The strategy sets how AI workloads are traced alongside everything else, so they share the same pipeline, owners and alerting. The open standard's conventions for AI telemetry are still marked as in development, and the strategy notes which parts may still change.
Start with a conversation
A partner replies within one business day.
Before you enquire
The strategy is built on the assessment's findings: coverage, tools, alerting, reliability and spend. If you already hold that picture, we confirm it and start from there. If not, the assessment comes first, because a target state drawn without it tends to repeat the tool choices that caused the problem.
It recommends what fits the evidence. The tooling decisions weigh what you already pay for against what each option would add. Building collection on an open standard keeps the decision reversible.
Service level objectives, error budgets and incident review are site reliability engineering practices, and the strategy sets them for your services. We help your teams adopt them and set up the alerting that follows them.
Usually, because much of the cost sits in data nobody queries. Filtering and sampling in the pipeline, retention by type of data and consolidating overlapping tools reduce the bill while keeping the signals incidents need. The cost policy sets a floor from your obligations so savings never go below it.