Problem
Latency is 340ms. Whether that is fine depends on a number that currently lives in someone’s head, in a slide from last quarter, or in a threshold field on a panel nobody has opened since it was created.
Solution
Make the target a first-class object. Something with a value, an owner, a review date and a definition of the metric it applies to, referenced by every panel and alert that needs it rather than retyped into each.
The distinction from a threshold is where most of the value is. A threshold is a number on a panel. A target is a commitment with provenance: who set it, when, against what, and what happens when it is missed. Panels reference it; they do not own it.
The SRE framing is the most developed version and worth borrowing wholesale. An SLO is a target for a service level indicator, and the derived quantity—error budget, the amount of failure the objective permits—turns out to be far more useful than the target itself. A budget converts a binary into a rate: not “are we meeting it” but “how fast are we spending the allowance, and will it last the quarter”. That reframing is what makes a target actionable on a Tuesday rather than only at a quarter boundary.
Three properties separate a target that works from a number in a config file.
Ownership. A target with no name attached cannot be revised, only ignored.
A review date. Traffic grows, systems change, and an unreviewed target is wrong within a year and still colouring panels.
Reachability. The target should be visible where the metric is, and its definition reachable from there. A goal nobody can find the reasoning for gets treated as arbitrary, and arbitrary targets get gamed or dismissed.
Few’s bullet graph is the display counterpart: it exists precisely to show a measure against a target and qualitative ranges in one compact strip, which is what a target-aware panel should look like.
Use when
Someone has genuinely committed to a number, the commitment has consequences, and multiple surfaces need to agree about it.
Don’t use when
The number is aspirational or nobody owns it. A target nobody is accountable for teaches viewers that targets on this dashboard are decorative, and that lesson transfers to the ones that matter.
Trade-offs
Targets create incentive, and incentive creates gaming: the metric improves and the thing it was standing in for does not. They also age badly and rarely have a process for revision, so a display can be confidently red against a goal set for a system that no longer exists. Making targets first-class is real product work—storage, permissions, history, versioning—which is why most tools leave them as per-panel thresholds. And a single target flattens a distribution: hitting a p95 target while the p99 doubles is a pass that hides a real deterioration.
Checklist
- Who owns this target, and do they know?
- When was it last reviewed, and when is it next due?
- Is the metric definition it applies to unambiguous and reachable?
- Is the target stored once and referenced, or retyped per panel?
- Does the display show distance to target, not only current value?
- Is there a budget or rate framing, or only a pass/fail?
- What happens when it is missed—does anything at all change?
- Could this target be met while the underlying goal is not?
- Does the alert that fires use the same number the panel draws?
- Is the history of the target visible, so a change is not silent?
Compare
Honeycomb treats SLOs as objects with budget burn as the primary display, which is the fullest expression of the pattern and the one that answers “should I act now” rather than only “am I over”. Grafana has no target concept: goals live as per-panel thresholds, so the same objective is retyped across dashboards and drifts independently in each. Datadog offers SLOs as first-class entities with their own status and history, so a target has a page rather than a field. Looker and BI tools generally push targets into the modelling layer, which makes them consistent everywhere and makes changing one an engineering task rather than a product decision.
Related
Target and progress is the display half, and covers the bullet graph properly. Threshold line is what a target usually degrades into when nobody owns it. Alert rule is what should reference the target rather than carry its own copy of the number. Explain this metric matters more here than anywhere, since a target on an ambiguous metric is an argument waiting to happen. KPI tile is where a target most often should appear and usually doesn’t.