The Dashboard Nobody Opens
We've built a lot of dashboards, and we've seen the same pattern often enough to name it. A company commissions a dashboard. It launches to genuine enthusiasm. Six weeks later, one person checks it on Monday mornings. Six months later, nobody does.
The cause is almost never the technology. It's that the dashboard answers questions nobody was actually going to act on. Twenty-four charts showing that things are broadly fine is not information — it's decoration. This article is about building the other kind.
Start With the Decision, Not the Data
The single most useful question before building anything: what decision will this number change? If there's no answer, the metric doesn't belong on the screen.
Run it through properly. "We want to track customer churn" is a wish. "When monthly churn in a segment crosses 4%, the account team calls every customer in that segment that week" is a decision — and it tells you exactly what to build: one number, one threshold, one alert, one owner.
This reframing usually shrinks a requested dashboard by two-thirds, and the remaining third gets used. A good rule of thumb: every metric on the screen needs a name attached to it. If nobody owns it, nobody acts on it.
The metrics that reliably earn their place are the ones tied to a lever you can actually pull — conversion rate at a specific step, cost per acquisition by channel, gross margin by product line, time-to-resolution in support. The ones that rarely do are vanity aggregates: total pageviews, cumulative signups, follower counts. They go up. They tell you nothing about what to do tomorrow.
Fix the Boring Layer First
Before any of this works, the unglamorous part has to be right. In our experience this is where most data projects quietly fail:
- One definition per metric. If finance and marketing calculate "active customer" differently, every meeting becomes an argument about whose number is correct. Write the definitions down once.
- Trustworthy pipelines. A dashboard that was silently three days stale once will never be fully trusted again. Show the last-updated timestamp on the screen.
- Access without a gatekeeper. If getting a number requires emailing someone, people stop asking, and decisions go back to being made on instinct.
When Machine Learning Is Actually Worth It
Plenty of problems sold as machine learning are better solved by a well-chosen threshold and a clear report. ML earns its cost when you have genuine volume, a pattern too complex to write rules for, and a prediction that changes behaviour — demand forecasting across thousands of SKUs, fraud scoring in real time, churn prediction with enough history behind it.
It's a poor fit when your dataset is small, when the relationship is simple enough to state in a sentence, or when nobody has decided what they'd do differently if the prediction came back positive. That last one catches more projects than people expect.
The good news for 2026: you rarely need a research team any more. Cloud forecasting services and modern analytics warehouses cover most common cases at modest cost, so the bottleneck is clean data and a clear question — not talent.
Build the One-Screen Version First
If you're starting from scratch, build the version that fits on one screen and answers three questions the leadership team already argues about. Ship it, watch who opens it, and let the second version be shaped by what people actually reached for. Dashboards that grow from real use stay in use; dashboards specified in a single workshop rarely do.
Qodebrik builds financial and operational dashboards that people keep open. Let's discuss the decisions yours needs to support.