Walk into any data team, and you'll find dashboards. Hundreds of them. Revenue by region. Sessions by day. Bounce rate. Average order value. Churn percentage. All of it beautifully rendered in Tableau or Looker or Power BI, automatically refreshing, always available.
And yet — almost none of it moves decisions.
That's a provocation, and I mean it seriously. Most metrics in most dashboards are there because someone once asked for them, they were easy to track, and they stuck around. The question nobody asks enough is: if this number changed, would anyone do anything differently?
The decision test
Every KPI I work with now gets put through what I call the decision test. It's simple: state the metric, then finish this sentence — "If this number went up/down by X%, the team would ___."
If you can't finish that sentence with a concrete action, the metric isn't a KPI. It's a vanity metric wearing a KPI's badge. Impressions, total sign-ups, page views — these are outputs, not decision triggers. They feel good when they go up, but they don't tell you what to do when they go down.
A real KPI creates a decision tree. Customer acquisition cost (CAC) crossing a threshold means you pause the current campaign and run experiments. Week-3 retention dropping below a threshold means you ship a re-engagement flow and review onboarding. The metric has a next step embedded in it.
Leading vs. lagging — and why most teams get this backwards
Revenue is the ultimate lagging indicator. By the time it drops, a dozen upstream decisions have already failed. Smart analytics teams obsess over leading indicators — the early signals that predict the lagging outcomes before they materialize.
Here's a pattern I see everywhere: a team tracks monthly revenue (lagging), weekly active users (slightly less lagging), and maybe NPS (a survey that takes weeks to collect). But they don't track the behaviors that reliably predict whether those numbers will move. Days-to-activation. Feature adoption depth in week 1. Support ticket topics trending upward. The ratio of users who complete the core workflow on day 1 versus day 7.
The framework I use: for every lagging metric your organization cares about, map at least two measurable leading indicators with a documented lag time. "When X happens, Y tends to follow in approximately Z weeks." That's a hypothesis. Over time, it becomes infrastructure.
The hierarchy of metrics
Not all metrics deserve equal attention. I think about them in three tiers:
North Star Metric: One number that captures the core value the product delivers to users. For Spotify, it's probably something like time spent listening. For a B2B SaaS, it might be workflows completed per user per week. This metric should be — if the product is working — a reliable proxy for business health. Every team should know it. Every sprint goal should connect to it.
Driver Metrics: Four to six metrics that demonstrably affect the North Star. These are the levers. When they change, the North Star moves. These live in executive dashboards and weekly reviews.
Diagnostic Metrics: The deep-cut numbers you pull when something breaks. Cohort retention curves. Funnel conversion by traffic source. These aren't on dashboards — they live in notebooks and queries, pulled on demand when you're debugging a problem.
The failure mode I see: organizations promote diagnostic metrics to driver status. Suddenly the team is optimizing for things that correlate with the real problem instead of addressing it directly. Conversion rate by page becomes the objective instead of understanding why users aren't converting.
The denominator problem
One of the subtlest sources of misleading KPIs is the denominator. "Monthly active users" sounds great — but active defined how? Anyone who logged in? Anyone who completed a meaningful action? If your definition of active is a login, you can inflate MAU by triggering more login prompts. The number goes up. Nothing valuable changed.
I've started requiring explicit denominator documentation for every metric: what counts, what doesn't, and what edge cases are excluded. A metric without a clear denominator definition is a metric waiting to be gamed — intentionally or accidentally.
Building the right measurement culture
Technical rigor is only half the problem. The other half is organizational. Even with perfect metrics, decision-making fails if the team doesn't trust the data, doesn't review it regularly, or doesn't have authority to act on what they see.
The best analytics setups I've studied share a common pattern: a small set of agreed-upon metrics, reviewed on a fixed cadence, with clear ownership. Not fifty metrics reviewed whenever someone has time — five metrics, every Monday, with someone accountable for each.
Measurement without accountability is theater. When every metric is everyone's problem, it's really no one's problem. Assign a metric owner. That person is responsible for explaining movements, flagging anomalies, and driving the conversation about what to do next.
The question that changed how I think about this
A mentor once asked me: "If you had to fly the business using only five instruments — like a cockpit — which five would you pick?" Not twenty. Not fifty. Five.
That constraint is clarifying. It forces you to think about what you actually need to make the next decision, versus what's just interesting to know. Most of what lives in dashboards is interesting. Very little of it is necessary. The discipline of measurement is learning to tell the difference — and having the courage to remove what doesn't pass the test.
Start there. Build your dashboards from that constraint, not toward it.