Business Intelligence & Dashboards

There are two revenue figures on the table, and the first half hour of the meeting goes into working out which one is right. What the numbers actually mean barely gets discussed.

We build BI solutions on a clean data model. Every metric is defined once and means the same everywhere, in the dashboard and in an assistant’s answer alike. Business teams run their own analysis without needing IT for every question.

The dashboard is the smallest part

All you see of business intelligence is the surface. Underneath are the sources (ERP, CRM, shop, accounting and the departments’ spreadsheets), then the integration that loads, cleans and combines that data, usually in a data warehouse, and finally the semantic model. It defines how metrics are calculated and how tables relate. Most complaints about dashboards trace back to these layers, not to the surface. That is why we start there.

You notice that reporting has stopped keeping up from small things. The monthly report takes several days every time. Sales and controlling give different revenue figures for the same period. There are plenty of reports, but nobody knows which one is current. And some decisions rest on numbers that are weeks old.

The tool matters less than you would think

The market is large, and almost every tool now ships with its own AI assistant. How well a tool fits the systems you already run matters more than any single feature. If you work with Microsoft 365 anyway, you will usually end up with Power BI, because reports embed into Teams and Excel users find familiar ground. If the data already sits in a data platform that brings dashboards and chat with it, a separate tool is not always needed. And where license costs or data sovereignty come first, self-hosted open source can be the better choice.

Then there are the number and type of users, requirements for data protection and hosting, and whether reports will later be embedded in your own applications. We are not tied to any vendor, and we will tell you if the tool you have is perfectly good.

First settle what “revenue” means

Which bookings count as revenue? When is a customer considered active? Which periods are compared? We settle these questions with the business teams at the start, because a dashboard is only as reliable as the definitions behind it. They go into one central semantic model that all reports use. When a definition changes, it is adjusted in one place and applies everywhere.

The model matters even more now that AI assistants use it too. They calculate with the same definitions as the reports, and if those are poorly described, they calculate wrongly. What such models look like is covered under Data Modeling & KPIs.

Dashboard or chat

In many tools you can now ask a question like “How has revenue in the south developed compared with last year?” and get the answer with a chart, instead of finding a report and filtering it. For ad hoc questions that can already replace a dashboard. Metrics that need to be in view every week still belong in a fixed dashboard, because then everyone sees the same thing. Both run on the same model in our projects.

Answers in chat are only useful once the model is well described, permissions apply there too, and typical questions have been tested before release. That is a topic of its own, covered under Conversational Analytics.

Self-service without sprawl

If business teams are free to analyze on their own, you soon get dozens of versions of the same metric. So we provide vetted, centrally maintained datasets for the teams to build on. With row-level security, each person sees only their own region or cost center, in the dashboard and in chat alike.

Where BI projects get stuck

Rarely for technical reasons. A dashboard with forty metrics answers no question; it is better to start with a few that support a concrete decision. Errors in the source data often surface only when the dashboard is finished, so we check the data automatically in the pipeline from the start. If the old reporting keeps running alongside, everyone compares the two sets of numbers until a fixed cutover date ends it. And without named owners for metrics and reports, dashboards go stale quickly.