A dashboard is often bought to settle an argument. Two reports disagree, a manager keeps a private spreadsheet, and someone decides that a new visualisation tool will resolve it. Months later the tool is live, the argument remains, and nobody trusts the numbers enough to use them in a meeting.
Most data visualisation failures start before anyone chooses a chart. They start with an undefined metric, an unclear decision, or a vendor demo built on clean sample data. This guide covers the mistakes to avoid, the requirements to write first, and the questions that separate a dashboard people use from one they ignore.
Why projects fail before the first chart
Three things are usually missing at the start. The first is a named decision: who will look at this page, how often, and what they will do differently because of it. The second is a definition: revenue might mean invoiced, collected, net of returns or net of tax, and every department may use a different one. The third is an owner for the data itself, not just for the dashboard.
When those are absent, the project becomes a design exercise. The charts look impressive, the numbers cannot be reconciled, and the business goes back to the spreadsheet that started the argument.
Eight mistakes buyers keep paying for
- Buying the tool before the metric. A licence does not create a definition. Agree the metric catalogue first, then choose software that supports it.
- Mixing grains without saying so. An average of averages, a monthly total shown beside a daily count, or a conversion rate calculated from mismatched denominators will all mislead.
- Hiding freshness and uncertainty. A number that is three days old looks identical to a live one. Show the last refresh, the source and any known gap.
- Building one dashboard for every audience. An operator needs order-level exceptions. A director needs trend and risk. Combining them produces a page that serves neither.
- Using chart types that distort. Truncated axes, dual axes with unrelated scales, 3D effects and pie charts with many slices make comparison harder.
- Ignoring accessibility and mobile use. Colour-only encoding, small labels and hover-only detail exclude users and fail on a phone.
- Leaving definitions in someone's head. A metric without a written formula, source table, filter and owner cannot be audited.
- No plan for permissions, export or retention. A dashboard often exposes more data than the original report. Access must be designed, logged and reviewed.
Requirements to write before you shortlist
Keep the requirement short enough that a vendor can respond to it line by line.
- Decisions and users: the five decisions the first release must support, and who makes them.
- Metric catalogue: name, formula, source, filter, grain, owner and refresh expectation for each metric.
- Data reality: source systems, known quality issues, volume, history and late-arriving records.
- Access model: roles, row-level restrictions, external sharing and audit requirements.
- Output: screen, PDF, scheduled email, embedded page or export to a spreadsheet.
- Performance: acceptable load time at your real data volume, not a sample.
- Support and handover: documentation, training, named support contact and an exit path.
A scorecard that separates demo from delivery
Weight the criteria before you meet anyone, and write down the evidence required for each score. A useful weighting for a mid-sized business is decision fit, data modelling, access control, performance, support and total cost. The exact weights are a business choice, but they should not be invented after the demo.
- Metric governance: can it store a definition, an owner and a change history?
- Modelling: does it handle your grain, joins and late data without duplicating logic in every chart?
- Connector fit: does it connect to your actual systems, including the awkward one?
- Testing: can you compare a dashboard total to the source ledger and explain every difference?
- Security: are permissions tested, logged and enforceable at the row level?
- Handover: will your team be able to build and maintain pages without the vendor?
Run the demo on your own messy extract, not the vendor's sample. Ask them to show a failed refresh, a permission change and a correction to a published number. How a vendor handles those three moments tells you more than any chart gallery.
Pricing questions that make proposals comparable
Do not ask for a single dashboard price. Ask for the cost model and the assumptions behind it.
- Is pricing per creator, per viewer, per data volume, per query or per connector?
- What happens when usage doubles, or when a new source is added?
- Are implementation, data modelling, training and project management priced separately?
- Which environments are included, and is a test environment chargeable?
- What does support cover, and what counts as a change request?
- How does renewal pricing work, and what is the notice period?
- Can you export the underlying data, the definitions and the dashboard logic?
A low licence fee with a large implementation quote is not a cheap option. Compare the first-year total, the run cost and the cost of adding the next department.
Run a pilot before you sign
Choose one decision, one data source and one small group of users. Give the pilot a fixed window and written acceptance criteria. A four-week pilot is usually enough to expose modelling and access problems.
- Reconcile one published number to the source system and record the difference.
- Test the five most common questions the group actually asks.
- Check a refresh failure: who is alerted, and what does the user see?
- Test access with two roles, including someone who should see only part of the data.
- Ask a user to answer a real question without help, and time it.
- Confirm the handover pack: definitions, data model, build notes and support route.
If the pilot needs the vendor present to answer every question, the tool is not ready for your team.
Plan the exit while you have leverage
Dashboards accumulate definitions and history. Before signing, agree how you will leave: raw data export, metric definitions in a portable format, API access, deletion of your data and a reasonable transition period. A semantic layer that lives only inside the vendor's platform is a lock-in risk, even when the contract says you own the data.
Measure the rollout, not the launch
Track weekly active decision-makers, unresolved reconciliation differences, time to answer a standard question, refresh success, definition disputes and support tickets per page. Usage alone is a weak measure: a page can be opened every day and ignored in every decision.
Where this fits in our work
If you need help turning operational data into a small set of trustworthy decisions, start with Business Automation. The pricing page explains how scope is built. For the metric layer underneath the charts, read the KPI dashboard buying guide.
Sources
- UK Government Analysis Function, Data visualisation: charts, accessed 17 November 2026.
- Financial Times, Visual Vocabulary, accessed 17 November 2026.
- W3C, Web Content Accessibility Guidelines (WCAG) 2.2, accessed 17 November 2026.
- Microsoft Learn, Power BI visualisation best practices, accessed 17 November 2026.
Responses (0)
No comments yet. Add the first one below.
Leave a response
Your email address will not be published. Required fields are marked *.