23/10/2026

How I use AI to design better Power BI dashboards

Most dashboards fail for a surprisingly simple reason: they answer questions nobody asked.

Most people start a new dashboard by building visuals. They debate line charts versus bar charts, pick colours and decide whether cards need rounded corners, but rarely step back to ask what the dashboard actually needs to answer. The result is a page full of charts that show a lot but explain little. That is a costly mistake: users decide within seconds whether a dashboard is useful to them, and that depends far more on what is on it than on how it looks.

That is why I follow the same three phases for every Power BI dashboard I design: who is it for, what do they need to know, and what should it look like. The design phase combines two inputs: a reusable template and a wireframe for the specific page. Lately, AI has been supporting me in each of these phases and in this blog, I want to show you how.

Dashboard design in three phases: audience and questions, KPIs, and design, using a template and wireframe as inputs.

Who – Know your audience and their questions

A CEO, a department manager and an operational employee rarely need the same information. Executives look for trends, risks and outcomes; managers for explanations and drivers; operational users for actions. A page that tries to serve all of them serves none of them well, so I give each audience its own page.

I start by identifying the primary audience. I ask them what decisions they need to make and which questions they want answered. I complement their input with formal requirements, such as legislations or company targets, and with my own analysis.

Take a company that has committed to 40% women in leadership by 2027 and wants an HR dashboard to track that goal. Depending on the audience, the dashboard needs to answer very different questions:

  • CEO: Are we on track to reach 40% women in leadership by 2027?
  • HR manager: Which departments are furthest behind, and is it a hiring or a promotion problem?
  • Team lead: Are the candidates in my current hiring pipeline helping my team close the gap?

Where AI helps: AI is an excellent sparring partner to check whether you have missed anything. It might raise questions nobody had considered and/or confirms the ones you already have.

PROMPT

I’m building an HR dashboard on our goal of 40% women in leadership by 2027. The audience is our CEO, HR managers and team leads. What questions would each of them want answered?

EXAMPLE OUTPUT (SHORTENED)

CEO HR manager Team lead
Are we on track to reach 40% by 2027? Which departments are furthest behind the target? How many women are in my current hiring pipeline?
How has the share of women in leadership evolved over the last 3 years? Is the gap caused by hiring, promotion or people leaving? Do my shortlists include at least one female candidate?
Where are the biggest risks to reaching the target? Do women and men get promoted at the same rate? Are women in my team ready for a next step (succession)?
How do we compare to our industry? Are we losing female leaders faster than male leaders? How does my team compare to the department average?
Key questions that need to be answered by different stakeholders

Some of these questions the stakeholders had already raised. Others, such as the one on people leaving, had not come up at all. And some you may choose to drop: comparing yourself with the industry only makes sense if reliable benchmark data is available.

The next step is the most important one: together with the stakeholders, I select the questions that really drive their decisions. For the CEO page, we kept three:

  • Q1: Are we on track to reach 40% by 2027, and how have we evolved over the past three years?
  • Q2: Where are the biggest risks to reaching the target?
  • Q3: Where does the gap come from: hiring, promotion or people leaving?

Q1 and Q2 come straight from the CEO column. Q3 is an HR manager question that we kept on purpose: the CEO does not need the detail per team, but does need to know whether to invest in recruitment, promotion or retention. The detailed analysis stays on the HR manager page. From here on, every KPI and every visual on the CEO page has to answer one of these three questions.

What – Turn the questions into measurable KPIs

A KPI (key performance indicator) is a measurable value that shows how well you are performing against a specific objective. The questions describe what people want to know; the KPIs make those questions measurable. Without this step, a dashboard becomes a collection of charts rather than a decision-making tool. The rule is simple: every KPI must answer one of the selected questions, or it does not belong on the page.

Where AI helps: share the selected questions and ask for KPIs that answer them, each with a clear definition and a target or normal range. Without a reference point, a number rarely tells you whether performance is good or bad.

PROMPT

These are the three questions our CEO wants answered about our goal of 40% women in leadership by 2027: [Q1–Q3]. For each question, suggest the KPIs that answer it. Give each KPI a clear definition and a target or typical range.

EXAMPLE OUTPUT (SHORTENED)

Question KPI Definition Target / range
Q1 Women in leadership (%) Number of leadership positions held by women, divided by the total number of leadership positions 40% by year X+1
Q1 Gap to target Difference in percentage points between the current share of women in leadership and the 40% target 0 pp by year X+1
Q2 Gap to target by department Gap to target, calculated per department Flag departments more than 5 pp behind
Q2 Diverse shortlist rate Share of leadership vacancies with at least one woman on the shortlist 75%
Q3 Female hire rate Share of new hires into leadership roles who are women, per period ≥ 40%, otherwise the gap grows
Q3 Promotion ratio Promotion rate of women divided by the promotion rate of men in the same period 1.0 = parity; normal range 0.9–1.1
Q3 Female share of leavers Share of people leaving leadership roles who are women Not higher than the current share of women in leadership
Key questions translated into measurable KPIs/indicators with a definition and thresholds

Within seconds, you have a solid first draft, and because every KPI is linked to a question, gaps in coverage are immediately visible. I combine this AI generated draft with existing sources, such as an internal metrics catalogue, and use it as the basis for a working session with the business. Together, we agree on the key metrics, decide what can be grouped and what can be solved using filters. At this stage, I don’t limit the list to metrics that are already available: if data is missing, I document it so it can be addressed.

Design – Combine the template and the wireframe

Only now does the design work start, based on two inputs. The template defines how every dashboard looks and is reused across projects. The wireframe defines what goes where on this specific page: it builds on the grid of the template and is filled with the questions and KPIs from the previous phases. I prepare the template first, then the wireframe, and bring them together in the final design.

Input 1: the template

Two terms are often confused here. A theme is a JSON file that holds your colours, fonts and default formatting for visuals. A template however (.pbit) is a complete, reusable report file: page size, background, a header with title and filters, a fixed grid, and the theme applied to it. I build the complete template once and reuse it for every report. That saves time and gives all reports a consistent look and feel.

MY STANDARD TEMPLATE

Power BI report template.

Where AI helps: no template yet? AI can help with both parts. Start with the theme, as that is the most time-consuming part to build by hand:

PROMPT

Create a Power BI theme JSON file with colours and font sizes based on the website of my company (intellus.group).

Power BI json theme file

AI generated a theme file with the colours below, together with step-by-step instructions to import it into Power BI. If your organisation has a style guide or an existing template, share that with AI instead of your website for a more accurate result.


I then applied the theme to an empty report, added the header and grid, and saved the result as a template (.pbit). AI can support this step as well: describe the layout you have in mind and ask for exact sizes and positions, so you don’t have to align every element by hand.

Input 2: the wireframe

A wireframe is a rough sketch of the page: no data and no colours, only what goes where and a visual suggestion. This is where the selected questions and KPIs get their place on the page. Moving a sticky note takes seconds, whilst rebuilding a finished page takes hours. I never start from a blank canvas: the grid of my template is the base, and I adapt it to the use case.

A WIREFRAME ON THE GRID OF MY TEMPLATE

Dashboard wireframe with sticky notes


When I fill in the grid, I follow these guidelines:

  • Start top left. Users scan a page in a Z-pattern, so the answer to the most important question belongs in the top-left corner.
  • Move from overview to detail. Headline figures at the top, trends and comparisons in the middle, details at the bottom or behind a drill-through.
  • Limit the number of visuals. If you need more than about eight visuals, consider a second page.
  • Let the question choose the visual. Bar charts to compare categories, line charts for trends over time, scatter plots for correlations. For parts of a whole, try a bar chart before reaching for a pie chart.
  • Every element has to earn its place. If you cannot name the question a visual answers, remove it. The same applies to unnecessary borders, gridlines, legends and decimals: each one costs your audience attention.

Where AI helps: share the grid of your template together with the selected questions and KPIs and ask for a layout in which every box answers one or more of the questions.

PROMPT

Create a wireframe for the CEO page of my HR dashboard on 40% women in leadership by 2027, based on the grid of my Power BI template (screenshot attached). Use these questions and KPIs: [Q1–Q3 + KPI list]. For each box, specify the question it answers, the KPI and a suggested visual type. Place the most important question top left and use no more than eight visuals.

EXAMPLE OUTPUT

AI generated wireframe for dashboarding

Every box traces back to one of the three CEO questions, and the main question, “are we on track?”, gets the most space. Use this as a starting point: refine it yourself or ask AI for alternative layouts.

Bringing it together

With the template and the wireframe in place, the actual design is quick: I add the template into Power BI, place each box of the wireframe on the template and select the right visual.

Where AI helps: provide both inputs and ask AI to create an HTML mock-up of the page to make the dashboard build even easier.

PROMPT

Draw the CEO page of my HR dashboard based on this wireframe: [wireframe]. Use the layout of my Power BI template (screenshot attached) and the colours from my theme. Show the visual type, the question each visual answers and some example data. Deliver it as an HTML page.

EXAMPLE OUTPUT, DRAWN ON MY TEMPLATE

HTML mock-up of a dashboard generated by AI

Once the design is approved, AI can also support the development work. Ask for step-by-step instructions to build the page in Power BI or go a step further and connect an LLM to Power BI Desktop through an MCP server. That setup lets AI work directly on your semantic model, and it can even help with the dashboard itself. My colleague Jurgen explains how to prepare your semantic model for AI, including the MCP server setup, in his blog.

⚠ A WORD OF CAUTION

You could also ask AI to design and build the entire dashboard from scratch. However, you would lose control over the content and the story, so I recommend using AI to propose a starting point, not the result. The best dashboards still need someone who understands the audience, the business context and the story that needs to be told. And whatever you do, don’t give personal or sensitive data to an AI.

Going one step further

AI also helps when standard Power BI visuals cannot tell the story you want. Custom visuals used to require significant development effort, but HTML-based visuals, SVG and Deneb (based on Vega-Lite) now make it much easier to build tailored, interactive visuals directly in Power BI, and AI is very good at writing the code behind them. A full walkthrough is beyond the scope of this blog, but feel free to reach out if you’d like to learn more.

Conclusion

Many organisations invest heavily in collecting, modelling and visualising data, but the real value lies in something much simpler: helping people understand what matters. The best dashboard is not the one with the most data, but the one that makes the next decision obvious.

How do you use AI when designing dashboards?


Written by

Julie Robbrecht

Consultant @ Lytix