Storytelling with Data Before & After: Makeovers for Charts That Drive Decisions

Data does not speak for itself. Storytelling with Data: Before & After shows—through practical makeovers—how the same numbers become noise or a decision tool depending on audience, tension, visual form and focus. For consultants and AI teams, this is the difference between a dashboard graveyard and a board action.

Source note: This article is an original practitioner synthesis of themes from Cole Nussbaumer Knaflic, Mike Cisneros and Alex Velez’s Storytelling with Data: Before & After. It is not a reprint. Figures are educational illustrations; support the original book for the full makeover library and design rationale.

---

0. Why this book still matters​

The first Storytelling with Data book established principles. Before & After is the practicum: dozens of real-world makeovers showing how practitioners moved from confusing to clear. In consulting and AI, teams ship dashboards faster than ever—and decision quality often falls because explanatory craft did not keep pace.

Knaflic’s core insistence: context is not optional. A chart without audience, stakes and a requested action is decoration.

Three commitments organise the makeovers:

1. Explanatory communication is a different job from analysis.
2. Every element on a slide must earn its place.
3. Titles should state the insight, not label the data series.

---

1. Before storytelling — audience and tension​

Figure: ask “Who is my audience?” and build tension around what they stand to gain or lose—not what you find interesting.

1.1 Audience questions​

Question	Why it matters
Who decides?	Titles and metrics must match their job
What do they fear?	Tension and urgency
What action is possible?	Explanatory charts exist to drive action
What is background vs foreground?	Declutter ruthlessly
What do they already believe?	Avoid re-proving the obvious

1.2 Tension — stakes, not drama​

Tension is the gap between current state and desired state with consequences:

Weak tension	Strong tension
“Here is skip rate data”	“Day-1 skips cost £2.1m/quarter in rework”
“GPU costs rose”	“Run-rate exceeds budget by November without action”
“Model accuracy improved”	“Remaining failure mode blocks production sign-off”

Example: Inventory turnover beating a benchmark is not the story for executives—cash freed / holding cost avoided / competitiveness at risk is.

1.3 Big idea and storyline​

Before opening Excel or BI tools, write:

1. Audience
2. Big idea (one sentence)
3. Requested action
4. 3–5 supporting points (pyramid-aligned—see Minto)

---

2. Exploratory vs explanatory​

Mode	Goal	Visual style
Exploratory	Find patterns; many charts OK	Dense, flexible, personal
Explanatory	Make one point clear	Sparse, annotated, audience-ready

Showing exploratory graphs in explanatory settings is a common failure: the audience re-does your analysis live—and often wrongly.

2.1 Handoff discipline​

Role	Delivers
Analyst	Exploratory workbook / notebook
Communicator	Explanatory slide with one point
Reviewer	“10-second comprehension” test

---

3. Makeover principles​

3.1 Declutter​

Remove borders, unnecessary gridlines, redundant legends, 3D, decorative clipart, and colours that encode nothing.

Clutter source	Fix
Heavy gridlines	Light or none
Duplicate labels	Direct labelling
Default legend	Label series on chart
3D bars/pies	Flat 2D
Logo wallpaper	One small logo max

3.2 Focus attention​

Pre-attentive attributes: position, colour, size, bold annotation. Highlight the point; mute the rest (grey baseline series).

Rule: If everything is bold, nothing is bold.

3.3 Choose effective visuals​

Question	Prefer	Avoid (often)
Compare categories	Bars	Pies for many slices
Trend over time	Lines	Dual axes without care
Part-to-whole	Stacked bars, square area	3D pies
Precise lookup	Table + highlight	Chart junk
Distribution	Histogram, boxplot (exploratory)	Pie of quartiles
Relationship	Scatter (exploratory)	Spurious dual axis

3.4 Square area / waffle charts​

Figure: a 10×10 square area graph can communicate part-to-whole more accessibly than pies for some audiences.

When to consider:

• Small number of categories
• Audience struggles with pie angle perception
• Need intuitive “counting squares” metaphor

3.5 Action titles​

Bad: “Skip rate by cohort.”
Better: “Goal-oriented prompts cut Day-1 skips from 32% to 12%.”

Title type	Function
Topic	Names the data
Action	States the insight + implication

Align slide titles with Minto point titles.

---

4. Narrative structure for data stories​

1. Context — situation the audience recognises
2. Tension — what is at stake if we ignore this
3. Insight — what the data shows (one primary point per slide)
4. So what — implication
5. Now what — decision / next experiment / investment

Figure: additional educational before/after makeover page from Storytelling with Data: Before & After.

Figure: before/after craft is not decoration—it is sequencing insight for action.

4.1 Pacing and repetition​

Technique	Use
Horizontal logic	Same chart type across slides for comparison
Build slides	Reveal series stepwise in live presentation
Summary slide	Recap big idea before ask

---

5. Before/after working method​

For each chart:

1. State the intended audience action
2. Screenshot “before”
3. List clutter and confusion sources
4. Redesign with one focal point
5. Write why each change helps cognition (required in the book’s pedagogy)
6. Test with a colleague for 10-second comprehension

5.1 Makeover checklist​

• Audience named
• Single takeaway sentence
• Chart type matches question
• Highlighting guides the eye
• Title states the point
• Next decision is explicit
• Appendix holds detail

5.2 Document the rationale​

Consulting teams should store before/after pairs in a visual standards library—new analysts learn faster from documented “why” than from brand guidelines alone.

---

6. Common makeover patterns (synthesis)​

Before pattern	After pattern
Pie with 8 slices	Bar sorted descending or small-multiples
Dual axis implying causation	Separate charts or indexed line
Rainbow categories	Grey + one accent colour
Chart + paragraph duplicating	Chart + annotation callout only
Dashboard screenshot in slide	One metric per slide for exec readout
Percentages without denominator	Label n and absolute impact

---

7. Consulting and AI applications​

7.1 Artefact map​

Artefact	SWD move
AI eval readout	One metric per slide; annotate failure modes
FinOps LLM cost	Trend + unit cost; highlight actionable driver
Experiment results	Control vs treatment with plain language effect
Risk dashboard	Mute green noise; spotlight exceptions
Pursuit commercial	Tie visuals to value case and explicit needs (SPIN)
Model comparison	Same scale; highlight winner on decision criterion only

7.2 AI-specific pitfalls​

Pitfall	Fix
Leaderboard of 12 metrics	One decision metric + appendix
Confusion matrix raw in exec deck	Highlight error type that blocks launch
Training loss curves	Keep exploratory; show business metric explanatory
Token cost vanity	Unit economics per successful task

7.3 Integration with pyramid structure​

Pyramid level	SWD role
Governing thought	Big idea sentence
Key lines	One slide each with action title
Evidence	Chart supports title only

---

8. Negative cases (expanded)​

Failure	Symptom	Fix
Rainbow colours	Legend required; no focus	Single accent
Dual axis abuse	False causation	Split or index
40-KPI dashboard	No decision	Exception-based exec view
Raw exploratory in exec meeting	Live mis-analysis	Explanatory subset
Story without ask	“Interesting” meeting	Now what slide
Makeover without rationale	Inconsistent team standards	Document why
Chart type mismatch	Bars for long time series	Line chart

---

9. Capstone — redesign one slide today​

1. Pick a noisy slide from your last readout.
2. Write audience, tension, and the decision you want.
3. Produce a before/after pair.
4. Explain five changes in bullets (cognitive rationale).
5. Rehearse a 60-second verbal story ending with the ask.
6. File in team visual library.

---

10. Team operating cadence​

Cadence	Activity
Weekly	One makeover share in team meeting
Per pursuit	Explanatory pack review before steering
Quarterly	Refresh visual standards from new makeovers

---

11. Colour, typography and accessibility​

11.1 Colour with purpose​

Use accent colour for	Avoid
The one series or bar that proves the point	Every category a different hue
Annotations and callouts	Decorative gradients
Exception states in risk charts	Red/green-only KPI walls

Check contrast for accessibility; do not rely on colour alone—add labels.

11.2 Typography hierarchy​

Element	Role
Action title	Insight (largest text on slide)
Subtitle	So-what in one line
Axis labels	Minimal; direct label when possible
Footnote	Source, date, denominator

11.3 Accessibility checklist​

• Chart readable in greyscale print
• Font size ≥ 18pt for exec rooms
• Data labels on key points for colour-blind viewers
• Animation disabled in PDF handouts

---

12. Verbal presentation — the 60-second chart story​

1. Context (10s): “You asked whether the pilot reduced handle time.”
2. Tension (10s): “SLA breaches cost £400k last quarter.”
3. Insight (20s): Walk one visual focal point only.
4. So what (10s): “We can extend to Tier-2 if approved.”
5. Ask (10s): “Approve Phase 2 budget today.”

Practice without reading the slide—if you must read, the slide failed.

---

13. Consulting pursuit pack template​

Slide #	Pyramid role	SWD element
1	SCQA	Tension statement
2	Governing answer	No chart—bold recommendation
3–5	Key lines	One action title chart each
6	Risks	Exception-focused visual
7	Ask	Decision + owners
Appendix	Evidence	Exploratory detail

Pair with Never Split the Difference: present after “That’s right” on the problem frame.

---

14. AI eval readout example (synthetic pattern)​

Slide title (action)	Chart choice
“Citation mode cuts unsupported answers from 18% to 3%”	Before/after bar
“Remaining failures cluster on ambiguous policy edge cases”	Pareto bar
“Human review time drops 22 min → 9 min per case”	Simple line or paired bar
“Recommend: expand to Tier-2 with same eval gates”	Text ask slide

Each title states the insight; appendix holds confusion matrices and full leaderboard.

---

15. Makeover catalogue — common chart types​

Knaflic’s Before & After library repeats patterns. Practitioners should recognise these before states instantly:

Before pattern	Cognitive problem	After move
3D pie, exploded slices	Angle comparison fails	Sorted bar or square area
Dual y-axis	Implied false correlation	Indexed lines or separate panels
Stacked area with 8 series	No focal point	Small multiples or highlight one series
Table with no highlight	Lookup burden	Grey rows + accent on decision row
Map with identical choropleth	Geographic ≠ priority	Bar of top regions only
Sparkline grid without context	No stakes	One trend + annotated event
Waterfall without subtotal labels	Hard to reconcile	Label each bridge step

Document the question type first (compare, trend, part-to-whole, distribution, relationship)—then pick the form.

---

16. Dashboard vs slide — different jobs​

Artefact	Purpose	SWD rule
Operational dashboard	Monitor exceptions	Many metrics OK; alert on breach
Exec steering slide	Decide one thing	One point per slide
Analyst workbook	Explore	Full density
Board pack	Approve / redirect	Explanatory subset only

Negative case: Exporting a live dashboard screenshot into a board deck without redesign—audience drowns, no ask lands.

---

17. FinOps and LLM cost storytelling​

Audience	Tension	Action title example
CFO	Run-rate vs budget	“Token spend tracks 18% above plan—routing Tier-2 to smaller model saves £140k/qtr”
Engineering lead	Cost vs latency trade-off	“P95 latency unchanged when cache hit >60%”
Product owner	Unit economics per task	“Cost per resolved ticket down 34% post-prompt refactor”

Always show denominator (tasks, users, sessions) alongside spend—raw token totals alone mislead.

---

18. Experiment readout template (A/B)​

Slide	Content
1	Hypothesis + decision criterion (pre-registered)
2	Primary metric effect with CI or clear n
3	Guardrail metrics (quality, latency) — exceptions only
4	Segment slice if heterogeneity matters (one slide max)
5	Ship / iterate / kill recommendation

Avoid peeking narrative—state whether the test hit power and duration planned.

---

19. Consulting QA rubric (peer review)​

Score each slide 1–5 before steering:

Criterion	1 (fail)	5 (pass)
Audience fit	Wrong exec level	Named decision-maker
Tension	None	Quantified stakes
Takeaway	Topic title	Action title
Visual form	Mismatch	Matches question
Focus	Rainbow noise	One accent
Ask	Missing	Owner + date

Reject packs averaging <4 on any criterion for C-suite readouts.

---

20. Capstone — team makeover sprint (half day)​

1. Hour 1: Each analyst brings one “before” slide + intended decision.
2. Hour 2: Pair redesign; write five “why” bullets per slide.
3. Hour 3: 10-second comprehension test with a colleague outside the team.
4. Hour 4: Publish to visual standards library; tag by chart type.

---

21. Annotated before/after — annotation checklist​

When annotating charts:

Annotation type	Use for
Direct label	Remove legend clutter
Callout box	Single exception bar/point
Reference line	Target, benchmark, threshold
Shaded band	Acceptable range only
Arrow	One direction of change

Maximum two annotation types per slide—more becomes clutter again.

---

22. Source line and trust​

Every explanatory chart should show:

Element	Example
Data source	“CRM extract, 2026-03-01”
Definition	“Active user = logged in ≥1× in 30d”
Caveat	“Excludes pilot cohort n=12”

Skeptical executives trust provenance as much as design—one footnote line prevents ten minutes of challenge.