Reporting and dashboards: How to turn business data into decisions

01Start with a decision that needs better evidence

Should you add support capacity, replenish stock or change a sales offer? If a dashboard will not influence a decision like these, clarify the question before commissioning more charts. Assign each main metric a decision, an owner and a review point. Replace “We want to see sales” with “On Monday, the sales manager identifies proposals with no next step and assigns someone to follow them up”.

Reporting is the regular preparation and sharing of results. A dashboard presents selected information for a particular situation. Analysis investigates relationships and possible causes. These activities work together, but a screen full of numbers cannot establish why customers leave. That question may also require transaction details, interviews or a separate test of a hypothesis.

02Separate executive, management and operational views

A director needs performance trends and risks. A manager needs to locate a problem. A person handling everyday work needs the next action. A universal dashboard can quickly become a crowded noticeboard. We recommend starting with one team and a few recurring decisions. Add another view when it serves a different audience or answers a different working question.

03Give every KPI a written definition

A key performance indicator needs an exact definition before people can manage by it. Revenue might refer to orders placed, invoices issued or payments received. Each interpretation has a purpose. Trouble starts when different reports use the same label for different calculations and people compare them as if they were identical.

• Purpose: Which decision does the metric support?
• Calculation: What is counted, and what is the denominator of a rate?
• Scope: Which statuses, countries, channels and customers are included?
• Time: Does the reporting date mean ordering, dispatch, invoicing or payment?
• Exceptions: How do cancellations, returns, test records and missing values work?
• Ownership: Who approves the definition and who corrects the source?
• Freshness: What delay is acceptable, and how is a failed update shown?

For monetary values, specify currency, VAT treatment and any exchange rate with its date. For targets, record the version of the plan. If a target changes halfway through a month, readers must know whether they are comparing performance with the original expectation or a revised decision. Keep definitions together and tell dependent teams when they change.

04Build a data foundation that avoids double counting

List sources such as the CRM, accounting system, warehouse and support platform. Give each an owner, a record identifier and an update method. Decide what one row represents: an order, an order line or a payment. Joining an order to its three lines must not count the full order value three times.

For more substantial reporting, separating events from descriptive information can help. Microsoft's star schema guidance distinguishes fact and dimension tables and stresses a consistent level of detail. You might separate sales records from product, customer and calendar information. A small spreadsheet does not automatically need this architecture, but the principle helps expose faulty joins.

Also decide whether a historical sale belongs to the sales region that existed at the time or to today's region. Changing a customer's address should not silently rewrite last year's report. Repeated imports should recognise and update known records rather than create copies. Check those rules on a small sample before processing the entire history.

05Validate the data before drawing the chart

Start with practical checks: are identifiers duplicated, are dates missing, do totals reconcile with the source, and has all expected data arrived? Select individual records and follow their journey from the original system to the displayed result. A plausible total is not enough: errors can cancel each other out.

A missing value is not automatically zero. Zero might mean no sales; a blank might mean an import failed. Show the last successful refresh and make incomplete data visible. Assign someone to resolve the issue. Without ownership, a warning can become a permanent decoration that everyone learns to ignore.

06Choose a tool around operation and ownership

Test a shortlist with your own data. Assess connectors, permissions, query speed, sharing, export, change management and the person who will maintain the result. The licence price is only one cost. Include data preparation, ongoing operation, support and changes to definitions.

This is not a brand ranking. A prototype might live in a spreadsheet, a shared model in BI and a work queue in an internal application. If moving data between systems is the main obstacle, begin with integrations and APIs. Replacing the charting tool will not repair a missing source.

For screen organisation, see our guide to dashboard design for data applications. This guide focuses on trustworthy numbers and their operation. At handover, require traceable sources, reproducible calculations and an explanation of corrections.

07Take care when averaging rates

If team A completes 9 of 10 tasks and team B completes 50 of 100, averaging 90% and 50% gives 70%. Overall completion is actually 59/110, approximately 53.6%. This is an illustrative calculation. Combine the relevant numerators and denominators rather than automatically giving a small group the same weight as one ten times its size.

Compare equally long, closed observation periods too. Recent proposals may still be open while older ones have final outcomes. Save the filter, cut-off date and closure rule with each comparison. A change in observation time should not be mistaken for a change in the quality of work.

08Design a path from result to explanation

At the top, show the main result, reporting period, comparison and data freshness. Below it, let readers investigate products, teams or enquiry sources. Put individual records in a follow-on view. Microsoft's dashboard guidance starts with the audience and recommends clear, consistent visualisation.

Use bars to compare categories and lines to show change over time. Keep a zero baseline when comparing bar lengths; truncation can exaggerate a difference. Communicate status with words and symbols as well as colour. Include meaningful labels, units and an accessible table of values. Test the dashboard on the screen people actually use, including operation without a mouse.

09Worked example: More qualified enquiries, fewer wins

The following figures are an illustrative model, not client results or an industry benchmark. We compare two closed cohorts of 100 enquiries, observed for equal periods. The first contains 60 qualified enquiries and 18 won contracts; the second has 80 qualified enquiries and 16 wins. Each contract belongs to exactly one cohort.

The qualification rate rises from 60% to 80%, while the win rate among qualified enquiries falls from 30% to 20%. The overall win rate falls by 2 percentage points, a relative decrease of approximately 11.1%. The chart shows counts, not causation. The next decision should be to investigate the qualification definition, enquiry mix and proposal process. These figures do not automatically justify increasing the marketing budget.

10Set access controls and a decision-making routine

Assign access according to working roles. Test direct links, exports and underlying data with an actual user account. In Power BI, row-level security restricts the Viewer role but does not apply to Admin, Member or Contributor roles. Hiding a column in a visual is not an independent security control.

In review meetings, record the finding, hypothesis, action, owner and review date. Send an alert only where someone has agreed how to respond. Alongside alerts about business performance, you need alerts about stale data. Otherwise the team may act on yesterday's failed import as though it were today's operational problem.

11Deliver a first version people can use

We recommend a sequence of work, rather than a universal delivery promise. Select the decisions, write the definitions and validate source data against a historical sample. Build a small view and ask a user to resolve a real situation. Add automation and more teams after checking calculations, access and refresh behaviour.

• Every main metric has an owner and an approved calculation.
• Totals and selected individual records reconcile with the source.
• Missing, late and failed data have distinguishable states.
• Permissions have been checked with an ordinary account.
• A user can find a problem and explain the next action.
• Maintenance, documentation and a review of benefits have owners.

12Common questions about business reporting

Want to connect data with everyday work? Bring one decision-making question and examples of your current information. When designing a custom system, we can identify sources, verifiable measures and a view with a clear user and a clear next action.