Day 45 of 100
The Dashboard Nobody Uses
π₯ Problem Statement
The team has dozens of dashboards, built up one at a time over a couple of years by whoever needed one at the time. During an incident, instead of pulling up one clear starting point, people scatter across several tabs, hunting for whichever dashboard happens to have the relevant graph, half-remembering which one showed something useful last time.
Nobody would say the team lacks observability data β there's plenty of it. Finding the right view of it, fast, under pressure, is the actual problem.
ποΈ Environment
- Dashboards: dozens, built up organically over time by different engineers
- During incidents: team struggles to find the right dashboard quickly
- No standard structure or naming convention across dashboards
- No dashboard tied explicitly to an SLO or golden signal set
π€ Your Challenge
What should a genuinely useful production dashboard contain, and why isn't 'more dashboards' the answer?
- What's the difference between having data available and having it organized around the decisions someone needs to make?
- If an engineer opens one dashboard during an incident, what should it immediately tell them?
- Why would golden signals (latency, traffic, errors, saturation) be a reasonable default structure?
- How should a dashboard connect a technical metric to actual business or customer impact?
Solution Hidden
Think through the problem yourself before looking at the answer.
π‘Solution
Step 1 β Understand the symptoms
Dozens of dashboards, each built for a specific need at the time it was created, tend to reflect whatever question its author was answering that day β not a coherent structure for "what does someone need to see during an incident, right now." The result is data abundance without organization: everything technically exists somewhere, but nothing points a stressed responder to the right place quickly, which is exactly the moment that speed matters most.
Step 2 β Identify the likely bottleneck
The gap isn't data volume, it's structure and purpose. Without a standard β a common starting point tied to golden signals and SLO status, a consistent naming and organization convention, explicit links from a top-level view down into deeper, more specific dashboards β every incident starts with a search problem before the actual investigation can even begin. That search cost, paid repeatedly, adds real minutes to every incident's response time, minutes that have nothing to do with the technical complexity of the actual issue.
Step 3 β Investigation
Rather than investigating a root cause, this calls for auditing what exists against what's actually needed:
- Interview a few engineers about which dashboards they actually reach for during a real incident β this usually surfaces both the few genuinely useful ones and the majority that go unused, informing what to keep, consolidate, or retire.
- Check whether any dashboard currently shows golden signals (latency, traffic, error rate, saturation) for the most critical services at a glance β if not, that's the single most impactful gap to close first.
- Check whether SLO status is visible anywhere as a first-glance signal β "how much error budget is left" is often the fastest way to gauge severity at the very start of an incident.
- Check whether dashboards link to each other β a top-level overview that drills down into more detailed, service-specific views is far faster to navigate under pressure than a flat list of dozens of independently-built dashboards with no relationship to each other.
Step 4 β Recommended action
Build (or designate) one clear top-level incident dashboard per critical service or system, structured around golden signals and current SLO/error-budget status β the thing anyone should open first, regardless of what the specific incident turns out to be. From there, link explicitly into more detailed, specialized dashboards for deeper investigation, rather than expecting people to already know which of dozens of independently-built dashboards has the relevant graph.
This doesn't require deleting every existing dashboard β some of the deep, specific ones are genuinely useful for their narrow purpose. It requires establishing which one is the starting point, making that obvious and consistent, and organizing the rest around it rather than leaving them as an undifferentiated pile.
Step 5 β Engineering lesson
A dashboard's value isn't measured by how much data it displays β it's measured by how quickly it helps someone make the decision they actually need to make in the moment they're looking at it. Dozens of dashboards built for narrow, one-off purposes can add up to worse incident response than a handful of well-structured ones, because the real cost during an incident isn't "the data doesn't exist," it's "finding the data takes too long."
π§ Todayβs Takeaway
A dashboard should help you make a decision, not just display data.
π Donβt Miss Tomorrowβs Challenge
A new real-world engineering challenge is released every day.
100 Days β 100 Challenges β 100 New Things Learned.
Get the useful stuff, not the noise.
Occasional notes on engineering, Platform Engineering, AI, cloud and things Iβm learning along the way.

