DATA4100 · Week 3

Performing Analytics,
UX & Report Writing

You can build charts. Now learn what to do with them — how to frame the analysis, design dashboards people actually use, and write findings that drive decisions.

01The Four Analytics Tiers— what kind of analysis are you doing?
02Dashboard UX Principles— five rules that separate good from unusable
03Report Writing for Data— observation vs insight vs recommendation
04Activities— redesign, write, and peer review
Navigate with ← → arrow keys · all examples use the Superstore dataset
Session Timing — 3 hrs
Four Analytics Tiers20 min
Dashboard UX35 min
Activity: Redesign40 min
Break15 min
Report Writing30 min
Activity: Write summary30 min
Peer Review20 min
Wrap-up + Week 4 preview10 min
The Central Tension

Building a chart is not
the same as doing analysis.

Last week you learned to create visuals in Power BI. That is a mechanical skill — like learning to type. This week is about what you type. Analysis is the reasoning that connects data to a decision. A chart without reasoning is just decoration.

Week 2
Tool Skills
How to build charts
in Power BI
Week 3
Analytical Skills
What to do with them
— and how to say it
Weeks 4+
Advanced Skills
Storytelling, platforms,
and modelling
Analytics Framework
The Four Tiers of Analytics
Every analysis you do sits somewhere on this ladder — knowing where matters for how you frame your findings
1
Descriptive
What happened?
"Give me the facts."
  • Summarises historical data
  • Reports counts, sums, averages
  • Most common in business dashboards
Superstore example:
Total sales by region — West had $725K, East had $678K
2
Diagnostic
Why did it happen?
"Dig deeper into the cause."
  • Compares groups and time periods
  • Uses drill-down and filtering
  • Identifies root causes
Superstore example:
Texas has negative profit — because 68% of orders have discounts ≥50%
3
Predictive
What will happen?
"Show me the likely future."
  • Uses statistical models
  • Forecasting and trend extrapolation
  • Confidence intervals communicate uncertainty
Superstore example:
Tableau forecast: Q4 2023 sales expected to reach $180K ±$22K
4
Prescriptive
What should we do?
"Give me an action."
  • Recommends a specific decision
  • Combines all three lower tiers
  • Requires judgement, not just data
Superstore example:
Cap all discounts at 30% in the Central region to eliminate the $26K annual loss
Key Principle
Most student work (and most Power BI dashboards) stays at Tier 1. The charts you built last week are Tier 1 — accurate but not yet analytical. Moving to Tier 2 and beyond is what separates a data analyst from a data entry clerk.
Analytics Framework — Applied
Same Data, Four Different Depths
How the same Superstore finding escalates through the four tiers
The Question
"What is going on with profit in the Central region?"
TierWhat you askWhat you do in Power BIWhat you say in the report
1 — Descriptive "What is the profit in Central?" Build a bar chart: Sum of profit by region. Central shows $39,706. "The Central region generated $39,706 in profit — the second lowest of the four regions."
2 — Diagnostic "Why is Central's profit low?" Scatter plot: discount vs profit, filter to Central. Add state filter. Texas and Illinois show heavy discounting driving losses. "The low profit is driven by Texas (−$25,729) where 42% of orders carry discounts above 50%, eroding all margin."
3 — Predictive "Will Central improve?" Line chart of Central profit by quarter. Add Tableau forecast. Wide confidence band signals uncertainty. "If current discounting continues, Central is forecast to remain loss-making in Texas through 2024 (95% CI: −$28K to −$22K)."
4 — Prescriptive "What should we do?" Calculate: removing discounts >30% in TX. Model the profit impact using a DAX measure. "Recommendation: Cap Texas discounts at 30%. Modelling shows this recovers ~$18K of the annual loss with minimal volume impact."
Today's Goal
By the end of this session you will be able to classify any analysis by tier, and write findings that advance from Tier 1 observation into Tier 2 or 3 insight — the level expected in your assessment.
Group Activity · 10 minutes

Which tier does each statement belong to?

Read each statement and decide: is it Descriptive, Diagnostic, Predictive, or Prescriptive? Be ready to justify your answer to the class.

A
"Office Supplies accounts for 60% of all orders by volume."
B
"Home Office customers have the highest average profit margin at 14% — likely because they place fewer, more considered purchases with fewer discounts applied."
C
"Based on the 2019–2022 trend, total annual sales are expected to exceed $700K in 2023."
D
"Superstore should discontinue the Tables subcategory in the South region, where it has generated a $14K cumulative loss over four years."
E
"Q4 sales are consistently higher than Q1–Q3 across all four years in the dataset."
Answers: A=1 · B=2 · C=3 · D=4 · E=1
Part 2 — Dashboard UX

A dashboard nobody can read
is worse than no dashboard.

UX (User Experience) in data visualisation means designing for your audience's cognitive limitations — not your own preferences. Five principles determine whether a dashboard enables decisions or creates confusion.

1
Visual Hierarchy
2
Cognitive Load
3
Colour Consistency
4
Interactivity
5
Audience Fit
UX Principle 1
Visual Hierarchy — Lead with What Matters Most
The human eye scans in a Z-pattern: top-left first, then right, then diagonally down-left, then right again
The Rule
1

Place your most important KPI in the top-left. It should be the first thing the eye lands on. Use size, contrast, and white space — not colour alone — to establish dominance.

In Power BI — Use Card Visuals for KPIs
1
Visualizations → Card visual (single large number)
2
Drag profit → Fields → set to Sum. Shows $286,397 in large font.
3
Duplicate for sales (Sum = $2,297,201) and profit_margin (Average = 12%)
4
Arrange the three cards in a horizontal row at the top of your dashboard — this creates an immediate executive summary before the charts.
Common Mistake
Putting a large map or chart in the top-left and burying the KPI numbers at the bottom. The audience sees geography before understanding whether performance is good or bad.
Example: KPI Card Row — Superstore
$2.3M
Total Sales
$286K
Total Profit
12%
Avg Margin
9,994
Orders
Z-Pattern Eye Tracking
KPI Cards Bar Chart Line Chart Map 1st 2nd 3rd
Design Rule
If someone can answer "is the business performing well?" within 5 seconds of opening your dashboard — the hierarchy is working. Test this with a partner.
UX Principle 2
Cognitive Load — Every Chart Added Has a Cost
The more a viewer has to process, the less they understand. Ruthless editing is a design skill.
2

Cognitive load is the mental effort required to interpret a visual. Every additional chart, label, colour, and gridline increases it. The brain has a fixed processing budget per screen.

The Editing Test
The rule
Ask of every visual on your dashboard: "If I removed this, would the audience lose a decision they need to make?" If the answer is no — remove it.
What to cut first
  • Decorative gridlines — reduce opacity to 20% or remove entirely
  • 3D charts — always increase load with zero information gain. Never use them.
  • Legends when labels will do — directly label bars and lines where possible
  • Duplicate information — a data label on every bar AND axis labels is redundant
  • Titles that restate the obvious — "Bar Chart of Sales" adds nothing; "Phones Lead Revenue by 18%" is informative
High Cognitive Load — Avoid
330K 260K 295K 230K 348K 197K 315K Phones Chairs Storage Tables Binders Machines Accessories Bar Chart of Sales Values
7 random colours · dense gridlines · redundant title · cluttered labels
Low Cognitive Load — Better
$348K ← Phones Chairs Storage Tables Binders ★ Machines Accessories Binders Lead in Total Sales
Single colour · highlight the winner · minimal gridlines · insight-led title
UX Principle 3
Colour Consistency — Same Colour Must Mean the Same Thing
Colour is a communication channel. Using it arbitrarily destroys trust in the data.
3

If blue = Consumer in your first chart, blue must mean Consumer in every chart on the dashboard. Breaking this forces the audience to re-read the legend on every visual — destroying comprehension.

Three colour roles
RoleUsageSuperstore example
CategoricalDifferent colours for different categories. Max 6–8 categories.Blue=Consumer, Orange=Corporate, Red=Home Office — consistent across every visual
SequentialOne hue, varying lightness. For quantities on a scale.Light blue = low profit → Dark blue = high profit on the choropleth map
DivergingTwo hues from a neutral midpoint. For loss vs gain.Red = loss-making states, Grey = zero, Blue = profitable states
In Power BI
Format visual → Data colours → manually assign consistent hex codes to each category value. Write these down in your project file so all team members use the same palette.
Consistent Palette — Superstore Dashboard
Recommended Colour Assignments
Consumer
#2471A3
Use in all segment charts
Corporate
#E67E22
Use in all segment charts
Home Office
#C0392B
Use in all segment charts
Alert / Loss
#C0392B — reserved for negatives
Positive / Growth
#1E8449 — reserved for targets met
Warning
Never use red for "Furniture" in one chart and then use red to mean "below target" in another. The audience will be confused about whether Furniture is a problem category. Assign semantic colours intentionally.
UX Principle 4
Interactivity as Navigation — Not Decoration
Slicers and filters should reduce the audience's remaining questions, not multiply them
4

Interactivity is powerful because it allows one dashboard to serve many analytical questions. But every filter you add creates a decision the audience must make before they can read anything. Keep controls minimal and purposeful.

Decision framework for slicers
Slicer fieldGood reason to add?Verdict
YearAudience needs to compare annual performanceAdd it
RegionRegional managers will each use this for their areaAdd it
CategoryProduct managers want category-specific viewsAdd it
Customer793 customers — audience can't meaningfully chooseSkip it
Order IDNo business question requires filtering to one orderSkip it
Ship ModeOnly useful for a specific logistics dashboardMaybe
Cross-filtering in Power BI
When you click any bar or pie slice on a Power BI dashboard, all other charts automatically filter to match — this is cross-filtering and it is interactivity without adding any slicer controls. Teach your audience to click into charts, not just use slicers.
Good Slicer Placement — Dashboard Layout
$2.3M Sales $286K Profit 12% Margin FILTERS Year ● 2019 ○ 2020 Region ● All ○ West ○ East Category ● All ○ Furniture ○ Tech Bar Chart Pie Line Chart (Time Series) Slicers right-column
UX Principle 5
Audience Fit — Design for Who Will Read It
A dashboard for a CEO requires different design decisions than one for a data analyst or a store manager
AudienceWhat they need from the dashboardRight chart typesWhat to avoid
Executive / CEO
5–10 seconds per view
High-level summary. Is the business on track? Where is the biggest risk? KPI cards · single bar chart · one map. Max 4 visuals per page. Scatter plots · matrices · raw tables · anything requiring technical literacy
Regional Manager
10–20 minutes per session
Performance for their area. How is their team tracking against target? Filtered bar charts · time series with target line · table with drill-down National-level aggregates that don't reflect their decisions
Data Analyst
Deep exploration sessions
Full data access. Ability to explore hypotheses. Raw figures visible. Scatter plots · matrices · distribution histograms · all filtering options Over-simplification that hides the data they need to investigate
Marketing Team
Campaign-specific use
Customer segments, product performance, geographic opportunity Donut charts for segment share · map · treemap for product prominence Financial loss/profit metrics that aren't in their remit
Assessment Relevance
Your dashboard assessment asks you to specify your intended audience. This is not decorative — it should genuinely constrain your design choices. Write the audience at the top of your dashboard file before you build a single visual.
Quick Self-Test
Look at your Week 2 dashboard. Could a CEO extract the main finding in under 10 seconds? If no — you have a hierarchy or cognitive load problem. Could a data analyst use it for hypothesis investigation? If no — you have a depth problem. Both are valid dashboards; they just need different designs.
Hands-On Activity · 40 minutes

Redesign your Week 2 Superstore Dashboard

Apply the five UX principles to improve the dashboard you built last week. Work in pairs — one person applies changes, the other reads as the intended audience.

1
Add three KPI cards to the top row: Total Sales · Total Profit · Average Margin. These become your visual hierarchy anchor. (5 min)
2
Remove at least one chart that a viewer could live without. Apply the editing test: "Would removing this hurt a decision?" (5 min)
3
Fix colour consistency. Open Format visual → Data colours for each chart. Assign the same colour to each segment/region across all visuals. (10 min)
4
Add three slicers in a right-hand column: Year · Region · Category. Delete any slicers that don't meet the decision test. (5 min)
5
Rename your chart titles to insight-led titles (e.g., "Phones Lead Revenue" not "Bar Chart of Sales"). Change at least three titles. (5 min)
6
Partner review: Swap screens. Can your partner identify the main finding in under 10 seconds? Give one specific piece of feedback. (10 min)
40 MINUTES TOTAL
Part 3 — Report Writing

A data report is not a description
of your charts. It is an argument
supported by your charts.

The most common mistake in student data reports is restating what the charts show in prose. That adds no value — the audience can see the chart. A report's job is to explain what the data means and what should be done about it.

Common mistake — avoid "The bar chart shows that Phones had the highest sales of $330,000 and Fasteners had the lowest sales of $3,000."
Better — adds meaning "Technology accessories dominate revenue, with Phones and Chairs each exceeding $328K — together accounting for 28% of total sales — suggesting the business is effectively a high-ticket durables retailer despite its broad category mix."
Report Writing Framework
Observation → Insight → Recommendation
Every finding in your report should advance through these three levels — observation alone is not sufficient
Level 1 — Observation (what the chart shows) "Q4 sales are higher than Q1–Q3 in all four years of the dataset. The highest monthly figure occurs in November 2021 at $118,448."
ADD CONTEXT → BECOMES INSIGHT
Level 2 — Insight (what it means for the business) "Q4 sales spike by an average of 35% above the annual monthly mean, creating a predictable seasonal demand pattern. This suggests Superstore can plan inventory procurement and staffing levels in advance of the October–December peak — currently an opportunity that appears underutilised given flat profit margins in the same period."
ADD ACTION → BECOMES RECOMMENDATION
Level 3 — Recommendation (what to do) "Recommendation: Superstore should introduce a Q3 inventory pre-loading strategy, purchasing high-demand subcategories (Phones, Chairs) in August to reduce Q4 stock-out risk and capture margin currently lost to rush procurement premiums. An estimated 3–5% margin improvement is achievable based on comparable retailers."
The Upgrade Formula
Observation → Insight
Add one of: a cause, a comparison, a consequence, or a surprise. Ask "so what does this mean for the business?"
Insight → Recommendation
Add one of: a specific action, a who, a when, or an estimated impact. Ask "what should someone do differently tomorrow?"
The Three-Sentence Test

Any finding can be expressed in three sentences:

Sentence 1: The key finding (observation)
Sentence 2: What it means (insight)
Sentence 3: What to do (recommendation)

If you can't write Sentence 3, your analysis hasn't reached the level required for a business report.

Report Writing
Structuring an Executive Summary
The opening of any data report — 150 to 250 words that a busy executive reads in full
The Five-Part Structure
1
Context sentence — one sentence explaining the purpose of the analysis and the dataset used. Who asked for this, and what data answered it?
2
Headline finding — the single most important insight, stated directly. No hedging, no "it appears that".
3
Supporting evidence — two or three specific data points that back the headline. Exact numbers, not vague language.
4
Key risk or caveat — one limitation of the analysis. Shows intellectual honesty and builds credibility.
5
Primary recommendation — one specific action, stated in one sentence. Actionable, not vague.
Common Mistakes
Starting with "This report will discuss..." — tells the reader nothing. Start with the finding.

Using vague language — "sales were quite high" vs "sales reached $2.3M". Numbers build credibility.

No recommendation — an executive summary without an action item is an observation, not an analysis.
Example — Superstore Profitability Report
Executive Summary

This analysis examines four years of Superstore transactional data (2019–2022, 9,994 orders) to identify drivers of profitability across products, regions, and customer segments.

The Central region's discount strategy is destroying margin: while it generates $723K in gross sales, aggressive discounting — particularly in Texas and Illinois — produces a net profit of only $39,706, the lowest of all four regions.

Specifically, Texas alone accounts for a $25,729 annual loss, driven by 42% of its orders carrying discounts ≥50%. In contrast, the Technology category generates a 20% profit margin even at moderate volumes, confirming that product mix — not sales volume — is the primary profit lever.

Note: this analysis cannot determine whether discounts are being applied correctly per contract or at-will by sales staff — a process audit would clarify this.

Recommendation: Cap all discretionary discounts in the Central region at 30%, and require manager approval for any discount above this threshold. Modelled impact: recovery of approximately $18K annually in Texas alone.

Context Headline finding Evidence Caveat Recommendation
Report Writing
Audience Register — The Same Finding, Three Ways
How you write a finding depends entirely on who will read it — the data doesn't change, but the framing must
The Finding
Texas has a $25,729 annual profit loss driven by discount rates above 50% on 42% of all orders.
For a CEO
Decision-maker. 30 seconds.
"Texas is our only loss-making state at −$25.7K annually. The cause is identifiable and fixable: capping discounts at 30% recovers an estimated $18K. Recommendation: implement a discount approval policy in the Central region before Q4."
Short · decisive · one action · dollar figures
For a Regional Manager
Operational. 5 minutes.
"The Texas market generated $170K in sales in 2022 but posted a −$25,729 profit loss. Analysis of individual orders shows 42% carried discounts of 50% or more — concentrated in Office Supplies subcategories. Orders with discounts under 30% were consistently profitable. A targeted review of discount approval practices in Texas is recommended, beginning with the Binders and Paper subcategories where loss frequency is highest."
Specific · operational · subcategory detail
For a Data Analyst
Technical. Full detail.
"Pearson correlation between discount rate and profit in the Texas subset (n=945 orders) is r = −0.38 (p<0.001), indicating a moderate statistically significant negative relationship. Orders at discount ≥50% show a mean profit of −$68 vs. +$42 for orders at discount <30%. The correlation is strongest in the Office Supplies category (r = −0.44) and weakest in Technology (r = −0.18). Recommend a multivariate regression controlling for order size and ship mode to isolate the discount effect."
Statistical · precise · methodological caveat
The Principle
The underlying data is identical in all three versions. What changes is level of detail, type of language, and what action is implied. Knowing your audience before you write is not a courtesy — it determines whether the report achieves anything.
Report Writing
Annotating Charts — Making Visuals Self-Explanatory
A well-annotated chart does not need surrounding paragraphs to explain it — the insight lives inside the visual
Four Annotation Techniques
1
Insight-led title — rename the chart title from "Sum of Sales by Month" to "Sales peak in Q4 every year — a consistent seasonal pattern." The title is the first text the eye reads.
2
Reference lines — In Power BI: Format visual → Analytics → Constant line / Average line. Add a dotted average line to a time series so the audience can see months above and below average instantly.
3
Data callout — select a specific bar or point → right-click → Show data label and make the font bold. This pulls a key number into focus without the viewer having to read the axis.
4
Text box annotation — Insert → Text box in Power BI. Position it adjacent to the anomaly you want to explain. Write one sentence: "Texas accounts for 65% of all Central region losses."
The Rule
If you have to explain what a chart means in the paragraph next to it, the chart has failed. Fix the chart — don't patch it with surrounding text.
Example — Annotated Time Series
Monthly sales peak in Q4 every year — a predictable seasonal pattern avg $120K $70K $50K $118K peak · Nov 2021 Q4 averages 35% above annual monthly mean Jan '19 Jan '20 Jan '21 Jan '22 Q4 '19 Q4 '20 Q4 '21 avg. line Four annotations: insight title · Q4 shading · peak callout · text box quantifying the finding. None of these require explanatory paragraphs.
Writing Activity · 30 minutes

Write a 200-word executive summary of your Superstore dashboard

Using the five-part structure, write an executive summary for the dashboard you redesigned in the first activity. This will form the basis of your Week 3 submission.

1
Choose your audience first — CEO, Regional Manager, or Marketing Team. Write their name at the top of your document. Every sentence must be appropriate for that audience.
2
Write the five parts: Context (1 sentence) → Headline finding (1–2 sentences) → Supporting evidence with specific numbers (2–3 sentences) → Caveat (1 sentence) → Recommendation (1 sentence).
3
Upgrade test: Underline every sentence that is purely observation (Tier 1). Try to upgrade at least two of them to insight or recommendation level.
4
Word count check: 150–250 words. If you are over 250, you have not edited ruthlessly enough. Cut the least important sentence.
30 MINUTES · TARGET 150–250 WORDS
Peer Review Activity
Giving and Receiving Analytical Feedback
20 minutes — swap executive summaries with a partner and apply the rubric below
Review Rubric — Use These Four Questions
1. Can you find the main finding in under 10 seconds?
Is the headline finding stated clearly in the first two sentences? Or do you have to read the whole summary to find it?
Rate 1–5 · write one specific comment
2. Does the recommendation follow logically from the data?
Could someone challenge the recommendation by questioning the data used? Is the logic chain clear and defensible?
Rate 1–5 · write one specific comment
3. Are there any observation sentences that should be insights?
Find one sentence that describes a chart without explaining what it means. Suggest how to upgrade it.
Find at least one and rewrite it
4. Is the register appropriate for the stated audience?
If written for a CEO, is it concise and action-oriented? If for an analyst, is it technically precise? Or does the register drift?
Rate 1–5 · suggest one change
Feedback Rules
Be specific. "This is good" is not feedback. "The headline finding appears in paragraph three — it should be sentence two" is feedback. Point to specific sentences, not the document as a whole.
What Makes Feedback Useful
Unhelpful feedback "Good work overall but the report could be clearer and more concise. The recommendation is a bit vague. Also consider the audience more carefully throughout."
Useful feedback "The headline finding is buried in sentence 5 — move it to sentence 1. Sentence 3 ('sales were high in Q4') is a pure observation — add the business implication. The recommendation uses 'should consider' which is too weak for a CEO audience — change to 'should implement by Q3'. Word count is 284 — cut the final two sentences about data limitations which duplicate point 4."
Why peer review matters
Reading as an audience — rather than as the author — is the single most important analytical habit to develop. You cannot spot the gaps in your own reasoning because you already know what you meant to say. Your partner doesn't.
After receiving feedback
You are not obliged to accept every suggestion. But for every piece of feedback you reject, write one sentence explaining why. This forces you to interrogate your own choices.
Week 3 Summary
What You Can Now Do
And what's coming in Week 4
Week 3 Learning Outcomes — Achieved
Classify any analysis by tier — Descriptive, Diagnostic, Predictive, or Prescriptive
Apply five UX principles to evaluate and redesign a Power BI dashboard
Write findings that advance from observation → insight → recommendation
Structure an executive summary in five parts and adapt register for different audiences
Annotate charts with insight-led titles, reference lines, and callouts
Assessment Connection
Your dashboard assessment
The redesign activity and executive summary you produced today are directly applicable to your assessment. Specifically: (1) KPI cards in the header, (2) colour consistency across visuals, (3) three slicers, and (4) a written executive summary using the five-part structure.
Week 4 Preview — Using IT Systems for Data Visualisation
Coming Next Week
Power BI Service — publishing dashboards to the web and sharing with stakeholders
Data refresh — connecting Power BI to live data sources (SQL, Excel, SharePoint)
Row-level security — controlling who sees what in a shared dashboard (regional managers see only their region)
Alerts and subscriptions — setting up automated email alerts when KPIs breach thresholds
Preparation for Week 4
Ensure your Power BI Desktop dashboard is saved and complete. You will publish it to Power BI Service in the first activity of Week 4 — bring your student Microsoft account login credentials.
"Statistics are no good unless you have good people to analyse and interpret their meaning."
— Brendan Rodgers · Today was about becoming that person.