learn-business-intelligence-with-phoebe / Leader session 4 of 6
Learn Business Intelligence with Phoebe · Leader track · Session 4 of 6

Commissioning dashboards that get used

Most dashboards die quietly: built in two weeks, admired at launch, opened by nobody in month three. The failure is almost never the charting - it is the commissioning. This session gives you the leader's toolkit: a one-page brief that forces the right questions before the build, an acceptance checklist built from Stephen Few's famous design mistakes, and a rollout playbook that plants the dashboard inside a ritual so it actually gets opened.

🟡 Leader track Leaders: execs · sponsors · managers No code · 45 minutes
0-3 · Welcome 3-16 · The brief 16-29 · Acceptance checks 29-45 · Rollout + your turn
Part 0

You are the client, not the builder

Leaders rarely build dashboards, but leaders commission almost all of them - and a vague commission produces a pretty, unused artifact every single time. "Give me a dashboard of everything" is how it starts; a wall of charts nobody opens is how it ends. Today you learn the client-side craft: specify sharply, accept rigorously, and launch into a habit. Last session (a3) made your numbers trustworthy; this one makes them looked at.

Live - presented in session Self-study - read after class ▶ Mini-BI - interactive playground Official sources covered
★ What you walk out with today A five-field, one-page brief template that turns "I want a dashboard" into a buildable request, a five-minute acceptance checklist distilled from Stephen Few's thirteen design mistakes, and a rollout habit loop that gets your dashboard opened twice a week instead of twice ever.
Part 1 · covers Few IDD ch3-4, MS Learn "Scope report design requirements"

The brief before the build 13 min live

Stephen Few's rule and Microsoft's newest Power BI design guidance say the same thing from different decades: know the audience, the question, and the action before anything gets built. Microsoft's redesigned "Design effective reports" learning path literally opens with a module called "Scope report design requirements" - requirements first, visuals later. Your version of that discipline is one page with five fields. If a request cannot fill them, it is not ready to build.

Dashboard brief · one page, five fields 1 Who reads it COO and three VPs, Monday 9am 2 What question it answers Is monthly revenue on plan, and where not? 3 What decision follows Reallocate campaign spend, or hold 4 How often it is read Weekly glance, 30 seconds 5 On what screen Laptop, plus the boardroom TV Notice what is missing: chart types. Fields 1-5 determine the charts - never the other way round.
🔍 Click to zoom - the five-field brief; if a request cannot fill it, it is not ready to build
LiveAudience changes everything6 min

Few's chapter 4 lists the fundamental considerations that should shape any dashboard before design starts: frequency of use, the users themselves, screen size, and the type of data. Change the audience and every one of those answers flips - which is why the same "revenue dashboard" request means two entirely different products:

  • The exec glance: read weekly for 30 seconds, on a laptop or a boardroom TV, standing up. Wants 5 numbers with targets and deltas, zero interactivity, everything on one screen. Success = the anomaly is visible from two meters away.
  • The analyst drill: open half the day, on two monitors, sitting down. Wants filters, drill-downs, exportable detail, and tolerance for density. Success = any follow-up question is answerable without writing SQL.

Build the drill for the exec and they stop opening it. Build the glance for the analyst and they rebuild it in a spreadsheet. Field 1 of the brief exists so the builder knows which product they are making.

One dashboard, one audience When a request names two audiences, that is two dashboards sharing a semantic layer - not one dashboard with twice the charts. Splitting it at brief time costs a page; splitting it after launch costs a rebuild.
Self-studyThe question behind the request4 min read

"I want a dashboard of everything" is the most common commission and the most reliable failure. Nobody reads everything; people read answers. Your job as the commissioning leader is to excavate the Monday-morning question - the specific thing the reader will ask the screen at 9am.

  • Ask "what will you do differently?" If no decision changes based on the number, the number is decoration. Field 3 of the brief is the filter that kills vanity metrics before they are built.
  • Ask "what did you check last Monday?" People's actual recurring question is usually much narrower than their stated wish. "Everything about sales" collapses into "did any region miss plan?" under one minute of questioning.
  • Ask "what would make you stop reading?" Few's chapter 3 frames requirements-gathering as learning what the dashboard must do for the user, not cataloguing available data. Data-driven scoping produces walls; question-driven scoping produces answers.

A useful test: the brief's question should fit in one sentence and contain no "and". "Revenue and churn and pipeline and NPS" is four questions - and probably two dashboards.

Part 2 · covers Few IDD ch2 (thirteen mistakes), ch13 (real critiques)

The thirteen mistakes as an acceptance checklist 13 min live

Chapter 2 of Few's Information Dashboard Design catalogues thirteen common mistakes dashboard designers make; chapter 13 applies them in brutal real-world critiques. You are not the designer - so flip the list. Every designer mistake becomes a commissioner's check: five questions you can ask of any delivered dashboard, in five minutes, with no design training at all.

"Looks right" - easy to fake On-brand colors and a polished logo Gauges, 3D effects, glossy tiles Impressive number of charts and pages Every available metric displayed somewhere Looks great in the launch demo "Reads right" - the real checks Everything on one screen, no scrolling Context beside every number: target, prior Right precision - $1.9M, not $1,943,207.18 No decoration that carries no data Color only where it means something Accept with the right column only. Everything in the left column can be added later - and usually should not be.
🔍 Click to zoom - the acceptance board: judge dashboards by the right column, not the left
LiveFive checks in five minutes6 min

Run these on any delivered dashboard, in order, before you say "approved". Each one is a Few mistake inverted:

  • 1 · One screen? Few's first mistake is exceeding the boundaries of a single screen. If the Monday answer requires scrolling or tab-hopping, part of it will simply never be seen.
  • 2 · Context next to every number? A lone "$1.9M" is trivia. Against target? Versus last month? Numbers without comparison points force the reader to supply memory - most will not.
  • 3 · Right precision? Excessive decimals are false accuracy and real clutter. Exec dashboards round; the detail lives a click deeper.
  • 4 · Any decoration? Gauges, 3D, backgrounds, logos beyond one corner - Few calls this useless decoration, and every pixel of it competes with the data.
  • 5 · Does color mean something? If four bars are four colors for variety, color has been spent on nothing. Color should mark the exception, the target breach, the thing to look at.
Real world

The five-minute sendback. A COO we know runs exactly this list on every new dashboard, out loud, with the builder in the room. First pass usually fails on checks 2 and 3. Second pass usually ships. The builders learned the list within a quarter - and started passing on the first try. Acceptance checks train your whole BI supply chain.

Self-studyContext is the cheapest upgrade3 min read

If you only enforce one check, enforce check 2. The gap between a number and a decision is context, and context is almost free to add:

  • Number alone: "Revenue: $1.9M". The reader must remember the plan, last month, and last year to know if this is good news. Most readers shrug and move on.
  • Number + target + delta: "Revenue: $1.9M · plan $2.0M · -5%". Now the dashboard has done the judging. The reader's next thought is "why?", which is exactly the conversation a dashboard exists to start.

This costs the builder one measure and a few characters of layout. It converts a reporting artifact into a decision instrument - the highest return-on-effort edit in all of BI. When you send a dashboard back, "add the target and the delta" should usually be line one.

Part 3 · covers Few IDD ch14 (design process, rollout, validating effectiveness)

Rollout and the habit loop 6 min live

Few's final chapter is about moving from imagining to unveiling - and validating that the thing actually works after it ships. Here is the leader's translation: a dashboard succeeds when it enters a ritual. Not when it launches, not when people say it is nice - when the Monday standup opens with it on the screen, or the month-end review walks through it page by page. A proper launch has three moves: train the readers (fifteen minutes, live, on real numbers), pin it into the ritual where its question gets asked, and retire the old report - as long as the legacy spreadsheet survives, the dashboard is optional, and optional dashboards die.

LiveMeasure the dashboard itself4 min

Every BI platform counts views - so commission the meta-metric along with the dashboard. Few's chapter 14 insists on validating effectiveness after rollout; this is the lightweight version:

  • Watch usage weekly for the first month. Opens per intended reader, not total opens. Ten views that are all the builder refreshing their own work is a failed launch with good-looking numbers.
  • Run the 2-week silence test. If a dashboard goes unopened for two weeks, it does not get a redesign - it gets a conversation: what question did we get wrong? Sometimes the answer is "the ritual moved"; often it is "the question was never real".
  • Retire without guilt. A dashboard that failed the silence test twice gets archived. Dead dashboards are not harmless: they clutter the portal, dilute trust, and hide the certified content from a3 behind noise.
★ The commissioning loop, closed Brief → build → acceptance checks → ritual → usage review. Notice that four of the five stages belong to you, the commissioner, not the builder. Dashboards get used because leaders run this loop, not because analysts pick prettier charts.
Demo 1 of 2

Write the Daybreak exec brief together ★ 12 min · follow along

Daybreak's founder wants "a monthly dashboard for the leadership meeting". Vague - perfect. We will run it through the five fields live, then check a real chart against the brief we just wrote.

Field 1 - who reads it: the founder plus two leads, in the monthly business review, projected on a meeting-room screen. One audience, one sitting. Already we know: glance product, not drill product.

Field 2 - the question: "everything about the business" collapses under questioning into: is monthly revenue on plan, and if not, where is the miss? One sentence, no "and" joining unrelated topics. That is buildable.

Fields 3 to 5 - decision, cadence, screen: decision = where to focus retention and marketing effort this month; cadence = monthly, five minutes at the top of the review; screen = one projected display, readable from the back of the room. Big text, one screen, no interaction needed mid-meeting.

Now check a candidate chart against the brief. The box below shows monthly revenue as a line - the natural centerpiece for field 2. Does it answer "on plan?" on its own? No: there is no plan line, no target, no delta. Against the brief, this chart is necessary but not sufficient - it locates the dip (look at March) but cannot say whether we are missing plan. That gap is exactly what you would write in the acceptance notes.

The brief is doing the arguing for you Without the brief, "add a target line" sounds like taste. With the brief, it is a requirement: field 2 asks "on plan?", so the chart must show the plan. Commissioning documents turn design debates into requirement checks.
Demo 2 of 2

Your turn: acceptance-check a chart ★ 8 min · you judge

A builder just delivered the chart below for the same Daybreak review: orders by city. Run the five checks from Part 2 on it, out loud, and decide what you would send back. There is no single right answer - but there are checks it visibly fails.

Run checks 1 to 5. One screen: yes, it is a single chart. Context: is there a target or a prior period next to these bars? Precision: are the numbers rounded sanely? Decoration: anything not carrying data? Color: does the highlight mean something, or is it just pretty?

Interrogate it against a brief. Which Monday-morning question does orders-by-city answer for a monthly exec review? If the review's question is "is revenue on plan?", is city even the right dimension - or is this chart an analyst's curiosity promoted to an exec screen?

Write the sendback. One or two lines, requirement-shaped: "add prior month for context", "swap dimension to plan tier - city is not decision-relevant for this audience", or even "cut it - no decision changes on this chart". Cutting is a legitimate acceptance outcome.

Real world

Leaders who check get better dashboards. Builders calibrate to whatever their commissioners actually inspect. If launches only get "looks great!", the left column of the Part 2 board is what you will keep receiving. Five spoken minutes of reads-right checks, twice, permanently raises the quality of everything your team ships you.

Homework

Try it yourself - this week ◐ 20-30 min total

Source material

Official sources covered

This session translates the canonical dashboard-design literature into the commissioning leader's view - the designer's craft stays with Few and the vendor labs. This page covers:

Stephen Few · Information Dashboard Design 2e, ch2-4, ch13-14Thirteen mistakes as acceptance checks; requirements, fundamental considerations, real critiques, rollout and validation
MS Learn · "Scope report design requirements" (Design effective reports path, June 2025)Part 1 · audience, question, action before building - leader view of the module
Google BI cert · course 3, stakeholder presentation angleCommissioning and stakeholder framing here; the builder-side presentation craft stays with Google
Check yourself

Three questions before you go 🎯 ◐ 90 seconds

1 · A VP asks you to sponsor "a dashboard of everything about sales". The first thing to pin down is...

Audience and question are fields 1 and 2 of the brief - they determine everything else, including the charts. Few's requirements chapters and Microsoft's "Scope report design requirements" module both start exactly here: know audience, question, action before building.

2 · A delivered exec dashboard shows "Revenue: $1,943,207.18" as a lone number. Your highest-value sendback is...

Context is the cheapest upgrade: number + target + delta lets the dashboard do the judging, and sane precision removes false accuracy. Gauges are on Few's decoration list, and repetition adds clutter, not meaning.

3 · When is a dashboard actually "done"?

Launch applause and metric-completeness are both "looks right" signals. Few's ch14 ties success to validated use after unveiling: train, pin into the ritual, retire the old report - then watch usage and apply the 2-week silence test.

Leader session 4 cheat sheet · pin this

The five-field briefWho reads it · what question · what decision · how often · what screen. No charts in the brief - fields decide charts.
One dashboard, one audienceExec glance and analyst drill are different products. Two audiences = two dashboards, split at brief time.
The Monday-morning questionExcavate the real recurring question behind "everything". One sentence, no "and", a decision attached.
Few's ch4 fourFrequency of use, users, screen size, data type - settle these before any design work starts.
Five checks, five minutesOne screen? Context beside numbers? Right precision? No decoration? Color = meaning? Then accept.
Cheapest upgradeNumber + target + delta beats a lone number every time. Usually line one of the sendback.
Launch = 3 movesTrain the readers, pin it into a ritual, retire the old report. Optional dashboards die.
2-week silence testUnopened for two weeks = wrong question, not wrong colors. Fix the question or archive without guilt.