Overview
Develop clearer Power BI charts and apply statistical thinking to workplace questions. Attendees prepare data in Power Query, explore distributions, compare periods fairly, create introductory forecasts and investigate local patterns hidden by overall averages. Practical chart-building, critique and storytelling develop confidence in explaining findings, uncertainty and appropriate action.
What you’ll need
- Windows laptop with current Power BI Desktop, or supported access to a Windows environment
- Internet access and permission to add selected certified visuals from the store or an approved organisational collection
- headphones and microphone
Learning outcomes
By the end of the course, attendees will be able to:
- Prepare and group data in Power Query, checking types, missing values and row meaning
- Build and interpret distributions, distinguish continuous data from categories, and assess histogram bins
- Choose and improve charts, accessible design and clear emphasis
- Compare periods fairly, interpret rolling averages alongside raw observations
- Create a built-in forecast and explain its horizon, assumptions, uncertainty and limitations
- Investigate local event clusters against suitable baselines and alternative explanations
- Explain how evidence changes suspicion without treating a pattern as proof or inventing probabilities
- Select additional visuals from the Power BI visuals marketplace using publisher, certification, licensing and organisational checks
- Present an evidence-led story with caveats and a proportionate recommendation
Course topics
Start with the question and data
- Identify the decision, audience and question; import supplied datasets and distinguish categories, continuous measurements and discrete counts.
- Prepare and group data in Power Query; check row meaning, types, missing values and duplicates. Revisit statistical concepts as needed.
- Use paired chart examples to examine heuristics, framing and quick versus deliberate judgement, drawing on Kahneman’s Thinking, Fast and Slow.
Build distributions and interpret variation
- Explore approximately normal, left- and right-skewed, uniform and bimodal datasets; create numeric bins and frequency histograms in Power Query.
- Vary bin widths and boundaries; distinguish histograms from categorical bar charts. Compare groups using consistent bins, scales and appropriate counts or proportions.
- Interpret mean, median, range and quartiles; investigate outliers rather than automatically removing them.
Choose, critique and improve charts
- Compare bar, line and scatter charts with tables; apply graphical integrity, minimal decoration and easy comparisons through consistent scales and small multiples.
- Improve titles, units, labels, colour, contrast, layout and annotations; complete a before-and-after chart makeover and explain the decisions.
- Add and configure a selected shortlist from Microsoft’s Tornado, Sankey, Bullet Chart and Dot Plot. Check publisher, certification, licence and organisational policy: certification is designed to prevent external-service access, not an absolute security or analytical-correctness guarantee.
Compare periods and create introductory forecasts
- Compare totals, rates and like-for-like periods, accounting for days, working days, exposure and seasonality; select a suitable denominator.
- Use a supplied reusable Power Query rolling-average step; plot raw observations alongside smoothing and explain window length, lag and hidden spikes.
- Create a built-in Power BI forecast on suitable time-series data; choose a horizon and interpret uncertainty intervals, trends, seasonality, unusual events and extrapolation limits.
Spot local problems hidden by overall averages
- Investigate fictional or anonymised utility calls inspired by NI Water: ordinary daily totals can hide several calls from one postcode district in a short period.
- Group by district and time window in Power Query; use a matrix heatmap and, where suitable and permitted, a map. Assess clusters against local baselines, exposure, duplicates and alternative explanations.
- Consider window boundaries, refresh frequency, base rates and false alarms; distinguish a pattern’s probability without a fault from a fault’s probability given the pattern. Recommend investigation and further evidence, not an unproven predictive-maintenance model.
Build and present an evidence-led story
- Apply Knaflic’s Storytelling with Data: audience, context, decluttering, selective emphasis and annotations. Critique McCandless’s Information is Beautiful and Tufte’s challenge to presentation formats that obscure evidence.
- Build a small interactive report with purposeful slicers and tooltips; check that filtering and aggregation preserve fair comparisons and precomputed summaries.
- Present “What? So what? What next?” with findings, uncertainty, limitations and proportionate action; answer questions and refine the report after feedback.