From showing data to saying something
Two dashboards can hold identical charts and land completely differently. One is a wall of numbers the reader must decode; the other walks them through context, points their eye at the deviation, and ends with what to do. The difference is not more data - it is narrative structure plus attention control, the craft Cole Nussbaumer Knaflic codified in Storytelling with Data. Tonight we apply it to Daybreak's March dip, then add the analytics-pane features that give a story its spine: reference lines, forecasts, and anomaly detection.
The narrative arc 7 min live
Every effective data story has three acts. Context: what the audience already knows and what they need to decide. Conflict: the deviation from expected - for Daybreak, revenue dipped in March. Resolution: what we recommend doing about it. A dashboard without conflict is wallpaper; a chart without resolution is homework you handed the reader. The cheapest place to install the arc is the title: "Revenue dipped in March" beats "Monthly Revenue" because it does act two for the reader before they even parse the axes.
LiveTakeaway titles - the rewrite exercise4 min▶
A label title names the data; a takeaway title states the conclusion. The reader gets your point in the two seconds they were going to spend anyway. Rewrite every title on your b6 dashboard through this table:
| Label title (names the data) | Takeaway title (states the point) |
|---|---|
| Monthly Revenue | Revenue dipped in March |
| Revenue by City | One city carries an outsized share of revenue |
| AOV by Plan | Pro customers place larger orders than Basic |
| Orders by Status | Cancellations stay a small slice of orders |
| Units by Category | Volume and dollars crown different category winners |
Test: cover the chart and read only the title. If the reader still learns something, it is a takeaway title. If they learn only what the chart is about, keep rewriting.
Self-studyKnow your audience's first question3 min read▶
Knaflic's chapter one is really one exercise: before building anything, write down who the audience is and what you need them to know or do. Different readers bring different first questions to the same dip:
- The exec asks "how bad, and what now?" - they want magnitude, trajectory, and a recommendation in the first breath. Lead with the KPI delta and the action.
- The analyst asks "which segment?" - they want the dip decomposed: which city, which plan, which channel moved. Lead with the breakdown views.
- The ops lead asks "is it still happening?" - they want the freshest point and an alert. Lead with recency.
One dataset, three stories. If you cannot name the audience, you cannot pick the arc - which is why "who is this for?" was the first question in b6's layout session too.
Attention - the gold accent discipline 6 min live
Preattentive attributes - color, size, position - steer the eye in under 200 milliseconds, before conscious reading starts. That is a superpower with one rule: spend it in exactly one place. One accent on THE point, gray everything else. Ten highlighted bars is zero highlighted bars. Our playground has been demonstrating this all course: the line chart marks its minimum point gold automatically, so the March dip lights up before you have read a single axis label.
LiveWhere does your eye land first?3 min▶
Run the trend chart and answer honestly: what did you see first?
That landing was engineered. One saturated point against a calm line is Knaflic ch4 in a single pixel decision: the designer chose your first fixation for you. When you build, make that choice deliberately - accent the point your story is about, and only that point. Everything else earns gray.
LiveAnnotation grammar - point, label, source3 min▶
An accent says look here; an annotation says here is what you are looking at. The grammar is three parts, and shorter is stronger:
- The point: a marker or thin arrow touching the exact data point - not floating near it.
- The label: six words or fewer, stating the event, not the number the axis already shows. "March: revenue dips below trend" works; a paragraph does not.
- The source: a small muted line ("Source: Daybreak warehouse, Jan-Jun 2026") so the chart survives being screenshotted into a deck without you attached.
Analytics extras - the senior toolkit 7 min live
A bare number is an orphan; the extras give it relatives. Reference lines answer "compared to what?", forecasts answer "where is this heading?", anomaly detection answers "which points deserve a second look?". PL-300 tests these under "identify patterns and trends" (Analyze feature, grouping and binning, AI visuals, reference lines, error bars, forecasting, anomaly detection); Tableau ships them as the analytics pane (totals, reference lines and bands, average lines, trend lines, distribution bands, forecasting, predictive model). Same ideas, two toolbars.
LiveReference lines and targets - the cheapest analytics3 min▶
Context is the cheapest analytics you will ever add. A single dashed line transforms "revenue was X" into "revenue was X, below where it should be":
- Target line: the number leadership promised. Every gap to it is a conversation starter.
- Average line: the self-generated benchmark - our March-dip slide uses the average of prior months, so the dip has something to dip below.
- Bands: a normal range (say, min-max of the trailing months). Points outside the band earn attention; points inside earn silence.
Rule of thumb: if a viewer could reasonably ask "is that good?", the chart owes them a reference line.
LiveForecasting and anomalies - use with a spine3 min▶
Forecasts are models wearing a chart costume, and models have assumptions. Two rules keep you honest:
- Show the confidence band, always. A forecast drawn as a single confident line is a lie of omission - the band is the honest part. Tableau's default forecast draws it; keep it.
- Say the assumption out loud. "Assumes seasonality repeats and nothing structural changes" belongs next to the forecast, in words. If March's dip was structural, a naive forecast trained through it is quietly wrong.
Anomaly detection is the mirror image: the tool flags, humans explain. An algorithm can mark March as an outlier in seconds; only b9's investigation can say why. Shipping the flag without the explanation just outsources your homework to every viewer.
Self-studyAI narratives - trust follows governance3 min read▶
The newest extra writes the story for you: Power BI's Copilot can generate narrative visuals and summarize the semantic model, and both are now PL-300 exam material under patterns and trends. Two things to hold at once:
- It is real leverage. A generated first-draft summary of "what changed this month" saves the analyst the boilerplate paragraph and often catches a mover they missed.
- It is only as trustworthy as the model underneath. An AI narrative over an ungoverned semantic model fluently narrates wrong numbers. This is leader session a6's point wearing builder clothes: governance first, then generation. Your b4 measures are exactly what make a Copilot summary safe to ship.
Review AI narratives like you review a junior analyst's draft: check the numbers against the governed measures before it leaves the building.
The March-dip story slide ★ 12 min · everyone builds
We turn the raw trend chart into a slide an exec reads in five seconds. Five moves, in order - each one is a technique from parts 1-3. Sketch it on paper or in your tool of choice; the target is the mock below.
Start from the trend. Rebuild the month x revenue line from part 2's playground - that is the skeleton. The gold minimum marker is already doing the accent work.
Write the takeaway title. Not "Monthly Revenue 2026" - write "Revenue dipped in March". Act two, delivered in the title, before the reader parses a single point.
Place the annotation. A short arrow touching the March point, label above-right of it, six words max: "March: revenue drops below trend". Point, label, done.
Add the reference line. A dashed horizontal at the average of the prior months (Jan-Feb), labeled quietly at the right edge. Now the dip visibly dips below something.
End with the so-what. One sentence under the chart: "We investigate the drivers in the March review and bring a recovery plan." Act three - a resolution, promised. (b9 delivers it.)
Your turn: title it, dress it, restrain it ★ 10 min · build your own
Two fresh panels, no recipe. For each: read the chart, write a takeaway title, then decide which single analytics extra it deserves - and defend the choice out loud. The third card is the hardest skill of the night: choosing to add nothing.
LivePanel 1 · Category revenue - title it, then pick its extra4 min▶
Defend your extra: a category bar chart rarely earns a forecast (categories do not trend through time), but an average line can separate the carrying categories from the passengers. If the gold top bar already tells the story, the honest answer may be a title and nothing else.
LivePanel 2 · AOV by plan - title it, then pick its extra4 min▶
With two bars the comparison IS the story - your title should state which plan wins and roughly by how much. A useful extra here is the overall AOV as a reference line, so each plan reads as above or below the blended number.
Self-studyPanel 3 · What you would NOT add2 min read▶
Restraint is the senior move. For each panel above, name one extra you deliberately left off and why:
- No forecast on categories - forecasting needs a time axis and a believable continuation; a categorical bar has neither.
- No anomaly flags on two bars - with two data points, "anomaly" is just "the smaller one". Detection needs enough points to define normal.
- No second accent - if the gold bar and your annotation both shout, neither is heard. One accent per chart, full stop.
Knaflic's ch8 synthesis is exactly this: everything on the page either advances the story or competes with it. Delete the competitors.
Try it yourself - this week ◐ 25-35 min total
- Rewrite three chart titles at work from labels into takeaways. Cover-the-chart test each one: does the title alone teach something?
- Add one reference line (target or average) to a real chart you own. Watch whether the next meeting's questions change.
- Find one forecast in the wild - an earnings deck, a news chart, a colleague's dashboard - and identify its unstated assumption. Write the sentence its author should have printed under it.
- Bring to b8: the story slide you built tonight. Next session we ship it - refresh schedules, security, and distribution.
Official sources covered
This session carries the storytelling canon plus the analytics-features slice of both major exams. Hands-on vendor click-paths stay with the vendors (mapped in b10). This page covers:
Three questions before you go 🎯 ◐ 90 seconds
1 · What makes "Revenue dipped in March" a better title than "Monthly Revenue"?
A takeaway title does act two of the story in the space the reader was going to skim anyway. Cover the chart: a good title still teaches something.
2 · You want the exec's eye to land on the March point first. The Knaflic ch4 move is...
Preattentive attributes steer the eye in under 200ms, but only if the accent is scarce. Ten highlighted points is zero highlighted points - our playground's single gold dot is the move.
3 · A forecast is going on the exec dashboard. What must ship with the line?
A single confident line hides the uncertainty that is the honest part of any forecast. Show the band, print the assumption ("assumes seasonality repeats"), and let viewers judge.