Building Dashboards with AI: From Raw Data to Decision-Ready Visuals

A practical workflow for building dashboards with AI, from framing the question and preparing data to modelling and checking numbers before you share.

Every Monday morning, an Excel file of daily sales from 12 branches lands in the team inbox. Someone spends half a day copying, pasting, rebuilding pivots and redrawing charts. Yet when the boss asks "Which branches need more stock next week?", nobody can answer on the spot. The problem is not the tool. The dashboard was built to "show data", not to "support a decision". This article shows how to use AI at every step without forgetting the parts only people can decide.

What is a decision-ready dashboard?

A good dashboard is a screen that answers a specific business question in a few seconds and makes the next action obvious. AI speeds up the repetitive, technical work: cleaning data, writing formulas, suggesting chart types. But AI does not know whether your business defines "sales" with or without VAT, or before or after returns. You have to define that.

Step 1: Start from the question, not the data

Write a clear decision question

Instead of telling AI "make me a sales dashboard", write "Area managers decide every Monday whether to transfer stock between branches." A question like this lets AI suggest metrics that fit: branch sales versus last week, days of stock cover, and slow-moving items.

Let AI break the question into metrics

Try asking ChatGPT, Claude or Gemini: "Here is the decision the team makes. Suggest 3 primary KPIs and 5 supporting KPIs with formulas." Then choose yourself. Do not accept every suggestion.

Step 2: Prepare the data and the data model

The mechanism many people skip is separating the fact table (each sales transaction) from dimension tables (branches, products, dates) and linking them with keys such as branch code. This structure keeps filters and totals correct whether you use Excel, Google Sheets or Power BI.

• Give AI a sample of 20 to 50 rows and ask which columns belong in a fact table and which in dimensions.
• Ask AI to find anomalies: branch names spelled differently, mixed date formats, negative sales.
• Build a separate date table so week-over-week comparisons are easy.

Step 3: Let AI write formulas and suggest charts

Formulas and measures

Many BI tools now include AI. Microsoft Learn documentation states that Copilot in Power BI can help create report pages and write DAX queries. Google's documentation states that Gemini in Data Studio (the new name for Looker Studio) can create calculated fields from plain language. These features come with licensing conditions and some are still in Preview, so check with your organisation first.

Match the chart to the question

• Comparing groups: a bar chart sorted from high to low
• Trends over time: a line chart
• Headline numbers: a KPI card with a comparison to target or to the previous period

Example: a retailer's weekly sales dashboard

Imagine a cosmetics retailer with branches in malls across Bangkok. The top row has three KPI cards: sales this week versus last week, gross margin, and the number of products that will run out of stock within 7 days. The second row is a bar chart of sales by branch, highlighting branches below target. The last row is a table of products to transfer. An area manager opens it and knows immediately which branches to call today.

A good dashboard does not have the most charts. It tells people what to do next.

Watch out for

• AI can guess column names or formulas wrongly. Always check key numbers against the source by hand, at least 2 to 3 points.
• Averaging averages is a common mistake. Compute from real totals, then divide.
• Personal data. Do not upload names, phone numbers or customer details into general AI tools your organisation has not approved. Remove them or replace them with IDs first, in line with PDPA.
• Write definitions down. Put how each metric is calculated under the chart or on a notes page.

How to get started

1. Pick one decision your team makes every week and write it as a single sentence.
2. Have AI propose KPIs, then cut them down to no more than 3 primary ones.
3. Prepare the data, separate facts and dimensions, and let AI flag anomalies.
4. Build a first version in Excel, Google Sheets or Power BI within one day.
5. Let real users try it for 2 weeks, then remove charts nobody looks at.

If you want to practise the basics first, try the courses Using AI to analyse Excel data and build dashboards or Introduction to data analysis with Power BI on Plants.work Learn. If you are looking for a data role, browse open positions on the jobs page.