Back to articles
    ReportingDataTranslated from Czech

    Custom reporting with a chat over your data: when a custom application makes sense

    I build dashboards in Power BI, Looker Studio, Qlik or Klipfolio. Sometimes, though, a custom application with a chat over the data makes more sense. How I decide and how such a project runs.

    Hand-drawn line chart on paper with a ruler and pens
    Photo: Isaac Smith, Unsplash

    I build most reporting in tools companies already have: Power BI, Looker Studio, Qlik or Klipfolio. Sometimes, though, I conclude that an off-the-shelf tool would get in the way. Then we build a custom application: dashboards, written commentary and a chat where you can ask questions about the company's data.

    When an off-the-shelf tool is enough

    • The company already has a licence and people know how to use the tool.
    • The data sits in a handful of sources with ready-made connectors.
    • The dashboards are used by people who can filter and look for answers themselves.

    In that situation a custom application is an unnecessary cost, and I recommend sticking with what the company has.

    When a custom application makes sense

    • The data is spread across many sources – databases, exports from internal systems, CSV files, web analytics – and needs to be combined continuously.
    • Management wants answers, not filters. Instead of searching through charts, they type a question, for example why orders in one region dropped year on year, and get an answer with numbers.
    • Written commentary is needed – a short comment on the trend, not just a table.
    • The data must stay in Europe and under the company's control, including processing in the chat.
    • People outside the company should see the results, such as partners or branches, each only their own part.

    How such a project runs

    1. Data analysis. Before I design the first chart, I go through the data the company has. The output is a proposal of what to report and what information the data can provide. It often turns out that the most valuable numbers are not tracked by anyone, because they are split between two systems.
    2. A separate data layer. Data from company sources is moved into a separate layer. The client's databases stay unchanged, and new sources can be added step by step.
    3. Dashboards and reports. Tailored to how the company works and who decides what. I don't use templates.
    4. Chat and written commentary. A chat and a regular text summary run on top of the same data.
    5. Further integration. If the company uses Power BI, the data from the application can be shown there too.

    I build these applications with the process described in How I build applications with AI agents. The same dashboards can also show product comparisons with competitors based on public data.

    Rules I stick to

    • Every number is compared with the previous period. The last-year column is visible even while it is still empty.
    • Each metric has one definition, used by both the dashboards and the chat.
    • The data and its processing stay in Europe.

    What to watch out for

    A chat does not replace metric definitions. If the company has not agreed what exactly revenue or an active customer is, the chat will answer wrongly and confidently. That is why definitions are set in the first phase, not while tuning the answers.

    The second issue is source quality. A custom application does not hide errors in the data, it makes them visible. That is useful, but it should be reflected in the budget and the timeline.

    We offer custom reporting with a chat over data at OREGO Productions.

    More articles

    Close-up of a mechanical watch movement
    DataCenyTranslated from Czech

    Comparing products with competitors using public data: what I learned about matching

    Competitor prices and specifications are on websites, in price lists and in data sheets. Collecting them can be automated. The harder part is deciding which competitor product matches yours.

    Two people talking at a table by the window
    PulsavoHRTranslated from Czech

    Why annual performance reviews don't work and how we approach it at Pulsavo

    In many companies, performance is handled once a year with a form. Meanwhile goals go out of date and management learns about dissatisfaction from a resignation. At Pulsavo we build on ongoing work with goals, conversations and feedback.

    Architectural blueprints spread out on a table
    AIVývoj aplikacíTranslated from Czech

    How I build applications with AI agents: Full Agentic Spec-Driven Development

    An AI agent writes code fast. In a larger application, though, without context it repeats mistakes and breaks what already worked. That is why I work with a process in which specifications, decisions, a plan and tests are created alongside the code.