Why do most dashboards fail to change decisions?
Most dashboards fail because they show everything that can be measured, so nobody can see which numbers need action, and many charts have no owner or target that would tell anyone whether a change is good or bad.
Consider a fictional 45-person commercial cleaning company, Brightstep Facilities. The operations director, Amara, inherits a dashboard with 34 charts across four tabs. Her notes from the Monday meeting say: 'nobody looks past tab 1. revenue chart is monthly, complaints chart weekly, can't compare. three different definitions of a lost client. no targets anywhere. sales asked why the job count dropped and nobody knew if it was bad.' She wants one screen that the management team uses to decide what to do each week.
How do you choose KPIs for a dashboard step by step?
Choose KPIs by starting from the decisions the team makes, then picking the one or two measures that would change each decision, with an owner, a definition and a target. The steps below keep the list short on purpose.
1. List the recurring decisions
At Brightstep: how many cleaners to roster next month, which clients need a service call, where to focus sales effort and whether quality is slipping.
2. Pick a measure per decision
For each decision, choose the number that would change it. Aim for four to seven KPIs in total. Anything that informs no decision belongs in a report, not on the dashboard.
3. Pair leading with lagging
Lagging indicators, such as monthly revenue or clients lost, confirm what already happened. Leading indicators, such as quotes sent this week or missed visits, move earlier and give time to act. Each lagging KPI should have at least one leading partner.
4. Define each KPI once
Write one definition, the data source, how often it updates and who owns it. 'Client lost' means a contract ended with no renewal within 30 days, for everyone.
5. Set a target or threshold
A number without a target cannot tell anyone whether to act. Add a target and an alert threshold for each KPI.
How should a KPI dashboard be laid out?
Lay a KPI dashboard out with the headline measures at the top, each shown against its target and the previous period, the leading indicators that explain them underneath, and detail only one click away. Use the same time period across a row so numbers can be compared.
Show the date of the last update, the data source and a short note on known limits, for example 'complaints logged by phone only from 3 November'. Public-sector dashboard guidance from the UK Office for Statistics Regulation sets out similar habits, such as clear links to the data source, information on strengths and limitations, and a stated update schedule, and they serve private teams just as well. Check colour is not the only way a status is shown, so the dashboard stays readable for everyone.
What does a weak and an improved KPI set look like?
The weak version keeps all 34 charts and adds a 'Top 10' tab. The Monday meeting still debates what the numbers mean.
The improved version has six KPIs on one screen. Top row, lagging: monthly recurring revenue against target; clients lost this quarter. Second row, leading: quotes sent this week against a target of 25; missed or late visits this week, threshold 3%; open complaints older than five days; cleaner hours rostered against hours booked next month. Each tile shows the owner's initials and a link to the detail. When job count fell, the team could see that quotes had dropped three weeks earlier during a sales vacancy, a cause they could act on.
How do you use AI to draft dashboard commentary and check it?
Use AI to draft the short written commentary under a dashboard by giving it the figures, the definitions and the targets, and asking it to describe changes against target and the previous period without explaining causes it cannot see. Then check every number it quotes against the dashboard.
AI drafts often add plausible reasons, such as 'revenue fell due to seasonal demand', that nobody has verified. Ask the assistant to mark any explanation as a question for the KPI owner. Do not paste client names or personal data into a tool your employer has not approved; totals and definitions are enough for commentary.
How do you check a dashboard is doing its job?
Check a dashboard by asking, after a month, which decisions it changed. If a KPI has not prompted a single question or action, replace it. Confirm every tile has an owner, a definition and a target, that time periods match across rows, and that the AI commentary contained no figure or cause that was not on the screen.
What are the common mistakes with dashboards and KPIs?
The most common mistake is tracking what is easy to count instead of what informs a decision. Others: only lagging measures, so problems appear too late to fix; several definitions of the same KPI; no targets; mixing weekly and monthly figures on one row; and rewarding a KPI so strongly that people improve the number rather than the outcome, such as closing complaints early to hit a target.
Where does this lesson sit in Tosheo.ai?
This lesson sits in the Data Science and Analytics category of the data science course, after the lesson on correlation and causation. In Tosheo.ai the hands-on task asks you to cut a crowded fictional dashboard to a decision-led KPI set and write AI-assisted commentary, and the rubric feedback checks targets, definitions and any unsupported causes. It is a professional foundation for analysts and managers, not a course in a specific BI product.
Sources and scope
Office for Statistics Regulation: regulatory guidance on dashboards, PDF (checked 19 October 2026) ↗