Business intelligence: from figures to decisions, with one version of the truth

Ask a wholesaler how many active customers it has and you may get three answers. Sales counts every account that ordered in the past twelve months, finance every account that received an invoice this year, and the webshop every registered buyer. Each number can be defended, yet together they leave the board unable to say whether the customer base is growing.

Closing that gap, from figures scattered across systems to decisions people trust, is what business intelligence (BI) is for. The software matters less than most buyers expect; the agreements behind the figures matter more.

Raw data consists of separate facts: an order line, an hour booked, an invoice. It becomes information when you put the facts side by side: the margin on one customer group has fallen three months running. The decision is what someone does next, such as reopening the price agreements with those customers. A report that nobody acts on is a cost, however good it looks.

So the people who decide set the questions, and IT makes sure the data arrives reliably and on time. Even Microsoft, which sells Power BI, writes that implementing specific tools or features is not the objective of a data culture (source: Microsoft, December 2024).

Data quality: fix errors where they arise

A new dashboard exposes errors that were always there: duplicate customers, a cost centre with two names, hours booked in the wrong week. Do not correct them in the report: that fix must be repeated every period. The UK government's data quality framework, written for the public sector, advises root cause analysis to solve problems at the source rather than temporary fixes, and names six dimensions of quality (source: UK Government Data Quality Hub, December 2020).

Reports on individual employees or customers are likely to fall under privacy law: show totals where totals will do and limit detail by role. Data protection in practice explains the principles.

How organisations grow into BI

1. 1Separate spreadsheetsEach department keeps its own lists, and figures differ per author. Move on when two people build the same figures for one meeting.
2. 2Standard reportsFixed reports with written definitions, often in Excel with Power Query instead of copy and paste.
3. 3A shared modelSeveral sources meet in one model; reports refresh themselves and access is set by role.

Some organisations add a fourth stage, managed self-service: users build their own analyses on the central model without touching the definitions. Skipping a stage rarely works, though. A BI platform on top of loose definitions only speeds up the argument.

In an organisation with one controller, that person may hold three of these roles; write them down anyway. In a group, ownership runs per subject, with the same roles in each.

Many accounting and ERP packages have dashboards of their own, enough for a first step with one source. Excel with Power Query handles a few sources and a small circle of readers. Dedicated platforms such as Power BI, Tableau, Qlik or Looker come into their own when several sources meet in one model and many people read the results.

• Your sources: connectors for your accounting, ERP and CRM systems, or an export you can schedule.
• Definitions in one place: each measure defined once in a model and reused by every report.
• Access by role: a regional manager sees only their own region.
• Licences that fit: the cost of readers and builders over several years, at the vendor's current prices.
• Skills: someone in-house who understands the model, even if a partner builds it.
• Exit: you can take the model, the data and the documentation with you.

Excel or Power BI compares two familiar starting points.

1. 1
   Pick one decisionA decision the management team takes every month and argues about, such as which customers to keep at which price.
2. 2
   Choose three to five KPIsOnly figures that inform that decision, each with a definition sheet (see the toolkit). The KPI dashboard guide covers choosing and presenting them.
3. 3
   Agree one source of truth per KPIName the system and the field that count, and fix the worst errors there.
4. 4
   Build one report that refreshes itselfReconcile it with the ledger for a closed period before anyone relies on it.
5. 5
   Use it for three months, then reviewWhich figures led to a decision, which were ignored and what was missing. Management reporting shows how to explain the variances.