Session 8: Storytelling & Final Project
Analysis that nobody acts on has no value, however correct it is. This session covers leading with the answer, writing recommendations a manager can act on, presenting to someone with no time, and the final project: a dashboard over a real messy dataset plus a one-page analysis note.
Learning objectives
By the end of this session you will be able to do each of these without prompting.
- Lead with the conclusion rather than building up to it
- Write recommendations that name an action, an owner and a date
- Present to a manager who has five minutes
- Handle the questions that follow a finding
- Connect the dashboard and the analysis note into one deliverable
- Explain your work to someone who did not build it
The taught content
Lead with the answer
Most analysis is presented in the order it was done: here is the data, here is what I cleaned, here are the pivots, and finally, at the end, the finding. This is exactly backwards for the reader, who has to sit through your process before learning whether any of it matters. State the conclusion first, then give the evidence.
The structure that works is short. The answer in one sentence. Out-of-state revenue grew 14 per cent year-on-year while in-state was flat, and the growth is concentrated in two products. Then the evidence — the two or three figures that support it, each with its count and baseline. Then the recommendation. Then, if anyone wants it, the method. Nobody who only reads the first line is misled, and nobody who wants the detail is denied it.
The reason this feels uncomfortable is that leading with the answer exposes it to challenge immediately, while burying it delays the challenge. That is the point. A finding that cannot survive being stated in the first sentence is not ready to be presented, and it is far better to discover that while drafting than in the meeting.
Recommendations people can act on
A finding is not a recommendation. Ogun buys mostly one category is a finding; it leaves everyone unsure what to do next. A recommendation names an action, an owner and a timeframe: add the two next-best-selling categories to the Ogun price list by the end of the month, owned by the sales lead.
Then it must be proportionate to the evidence. If the finding rests on three months of data from one state, the recommendation should be a trial, not a policy change. Overselling a weak finding is how analysts lose credibility permanently — and credibility, once lost in front of a manager, is very hard to recover, because every future finding arrives pre-discounted.
Finally, say what it would cost and what would count as success. Trial the two categories in Ogun for one quarter; success is 10 per cent growth in Ogun revenue without margin falling. A recommendation with a success test can be evaluated; one without it becomes a permanent maybe that nobody revisits.
Presenting to a manager
Assume you have five minutes and one question. Open with the answer, show one chart that supports it, and make the recommendation. Everything else is preparation for the questions, not content for the presentation. A manager who wants more will ask; one who does not will thank you for being brief, and will remember that you were brief next time they need analysis.
The questions that always come, in roughly this order: so what? — which your recommendation should already have answered; how sure are you? — which your counts, baselines and limitations note answer; what would it cost?; and what are we not seeing? That last one is the interesting one, and having an honest answer is what distinguishes an analyst from someone who makes charts. We cannot see customers who stopped ordering, and that is the biggest gap in this analysis.
Then the discipline of not overselling under pressure. If challenged and you are wrong, say so plainly and revise — that is a fair point, the sample is too small for that claim, and I will come back with a full quarter. Defending a weak finding costs you the room; conceding it precisely costs you nothing and makes the rest of what you said more believable.
The dashboard and the note are one deliverable
The final project is not two pieces of work but one: a dashboard that shows the state of the business, and a one-page note that says what to do about it. The dashboard answers what is happening and is consulted weekly; the note answers so what and is read once, by the person who decides.
They must agree, which is a real requirement rather than an obvious one. Every number in the note must be traceable to the dashboard, and the note's headline must be the thing the dashboard makes visible. If the note claims growth the dashboard does not show, one of them is wrong and the reader will find it.
The note itself is one page: the answer, three supporting figures with counts, the recommendation with owner and timeframe, the success test, and four sentences of limitations. That constraint is the discipline — it forces you to decide what actually matters, and a reader will read one page but will not read six.
Explaining work you did not do today
The last test of any analysis is whether someone else can pick it up in six months and understand what was done and why. Your dashboard will outlive your memory of building it, and the person who inherits it will not have your assumptions.
So document as you go: which file is the source, what cleaning was applied and why, what each calculated field means, and what the known limitations are. The cleaning log from week one is exactly this, and it is why it was worth keeping. A dashboard with no documentation becomes a thing nobody dares change, and then a thing nobody trusts, and then a thing that gets rebuilt from scratch.
The final habit: say the numbers out loud before you send them. Reading revenue grew 14 per cent year-on-year aloud, then checking it against the dashboard, catches the errors that silent reading misses. Almost every analyst has sent a figure they would have caught in ten seconds by saying it out loud, and the cost of being caught is far higher than the ten seconds.
Instructor demonstration
The instructor presents the completed distributor analysis as the model: a dashboard opened live, a one-page note read aloud, the answer stated in the first sentence, three figures with counts, one recommendation with an owner and a success test, four sentences of limitations — then the class challenges it and the instructor concedes one point precisely.
- 01
Open the dashboard before saying anything
Let the class look for ten seconds. Explain that a dashboard should be readable without narration, and this is the test.
- 02
State the answer in one sentence
Out-of-state grew 14 per cent year-on-year while in-state was flat. Explain that the conclusion comes first and the evidence follows.
- 03
Show the one chart that supports it
Not five charts, one. Explain that everything else is preparation for questions rather than content for the presentation.
- 04
Give each supporting figure with its count
Show the case counts beside every percentage. Explain that a percentage without a count cannot be weighed.
- 05
Make the recommendation with an owner and a date
Add two categories to the Ogun price list this month, owned by the sales lead. Explain that a finding without an action leaves everyone unsure what to do.
- 06
State the success test
Ten per cent Ogun growth in a quarter without margin falling. Explain that a recommendation without a test becomes a permanent maybe nobody revisits.
- 07
Keep the recommendation proportionate to the evidence
Call it a trial rather than a policy change. Explain that overselling a weak finding costs credibility permanently.
- 08
Read the four-sentence limitations note
Coverage, exclusions, case counts, falsification. Explain that naming the gaps yourself is what makes the rest believable.
- 09
Answer the so-what question
Point at the recommendation. Explain that if this question is a surprise, the presentation was structured in the wrong order.
- 10
Answer how sure are you
Give the counts and the period. Explain that confidence should be stated as a range of evidence rather than as assurance.
- 11
Answer what are we not seeing
Customers who stopped ordering. Explain that an honest answer here is what distinguishes an analyst from someone who makes charts.
- 12
Take a challenge and concede one point precisely
Agree the sample is too small for one claim and commit to a full quarter. Explain that conceding precisely costs nothing while defending a weak finding costs the room.
- 13
Trace every figure in the note to the dashboard
Show each one live. Explain that if the note claims something the dashboard does not show, one of them is wrong and the reader will find it.
- 14
Hand over the documentation
Show the cleaning log, the source file and the field definitions. Explain that an undocumented dashboard becomes a thing nobody dares change and then nobody trusts.
- 15
Say the headline number out loud and re-check it
Read it, then look at the dashboard. Explain that reading a figure aloud catches what silent reading misses, in about ten seconds.
Guided practice
Final project: a dashboard and a one-page analysis note
You deliver the complete project — a one-screen dashboard over the real messy dataset, and a one-page analysis note that states the answer first, supports it with three figures and their counts, makes one recommendation with an owner, a timeframe and a success test, and states its limitations — then present it in five minutes and defend it.
- 01Confirm the export is cleaned, with the cleaning log complete.
- 02Confirm every pivot, chart and figure reads from the same cleaned table.
- 03Confirm the dashboard fits one screen with no scrolling.
- 04Confirm slicers are connected to every pivot and reset instructions are on the sheet.
- 05Write the answer to your question in one sentence.
- 06Check that sentence survives being read before any evidence.
- 07Select the three figures that support it, each with its count and baseline.
- 08Remove every other figure from the note.
- 09Write one recommendation naming an action, an owner and a timeframe.
- 10Add a success test that would show whether it worked.
- 11Check the recommendation is proportionate to the strength of the evidence.
- 12Write the four-sentence limitations note.
- 13Trace every number in the note back to the dashboard.
- 14Cut the note to one page.
- 15Say every figure out loud and re-check it against the dashboard.
- 16Hand the dashboard to someone who has never seen it and confirm they can answer a question.
- 17Present in five minutes: answer, one chart, three figures, recommendation, limitations.
- 18Take challenges, and concede any weak point precisely rather than defending it.
- 19Revise the note based on the strongest challenge you received.
The standard we hold you to
A one-screen dashboard over the cleaned real dataset with the cleaning log complete, every figure reading from the same table, slicers connected to every pivot and reset instructions on the sheet, verified by someone who has never seen it answering a question unaided; and a one-page analysis note opening with the answer in a single sentence that survives being read before any evidence, supported by exactly three figures each with its count and baseline and with every other figure removed, one recommendation naming an action, an owner and a timeframe with a success test and proportionate to the strength of the evidence, a four-sentence limitations note, every number traceable to the dashboard, and every figure said aloud and re-checked; presented in five minutes as answer, one chart, three figures, recommendation and limitations, with challenges taken and any weak point conceded precisely, and the note revised in response to the strongest challenge received.
Common mistakes and how to fix them
You present in the order you worked
Fix: Lead with the answer. A reader should not have to sit through your cleaning and your pivots to learn whether any of it matters.
You bury the finding to delay challenge
Fix: State it in the first sentence. A finding that cannot survive that is not ready to present, and it is better to discover it while drafting than in the meeting.
You report a finding with no recommendation
Fix: Name an action, an owner and a timeframe. A finding leaves everyone unsure what to do next; a recommendation tells them.
Your recommendation is bigger than your evidence
Fix: Call it a trial. Overselling a weak finding costs credibility permanently, and every future finding then arrives pre-discounted.
You recommend without a success test
Fix: State what would show it worked. A recommendation with no test becomes a permanent maybe that nobody ever revisits.
You present ten charts in five minutes
Fix: Show one and prepare the rest for questions. A manager who wants more will ask, and one who does not will remember that you were brief.
You defend a weak point under challenge
Fix: Concede it precisely and commit to better evidence. Defending costs you the room; conceding costs nothing and makes the rest of your work more believable.
You deliver the dashboard with no documentation
Fix: Include the source file, the cleaning log and the field definitions. An undocumented dashboard becomes a thing nobody dares change and then nobody trusts.
Expert notes
The habits that separate someone who can do this from someone who does it well.
- State the answer first and make it survive the first sentence. Presenting in the order you worked forces the reader through your process before they know whether it matters, and burying the finding only delays the challenge you should want.
- Every recommendation needs an action, an owner, a timeframe and a success test. Without the test it becomes a permanent maybe that nobody revisits, and without the owner it belongs to nobody.
- Keep the recommendation proportionate to the evidence. Three months of one state supports a trial, not a policy change — and overselling once costs you credibility on everything you present afterwards.
- Say every figure out loud before you send it, and hand the dashboard to someone who has never seen it. Ten seconds of reading aloud catches the error silent reading misses, and an outsider finds the assumptions you can no longer see.
Key terms
- Answer-first structure
- Conclusion, evidence, recommendation, then method. The reader is never misled by reading only the first line.
- Recommendation
- An action with an owner and a timeframe. A finding without one leaves everyone unsure what to do.
- Success test
- What would show the recommendation worked. Without it, a recommendation is never revisited.
- Proportionality
- Matching the strength of the recommendation to the strength of the evidence. A weak finding supports a trial, not a policy.
- Analysis note
- One page: answer, three figures with counts, recommendation, success test, limitations. The constraint forces you to decide what matters.
- Traceability
- Every figure in the note being visible on the dashboard. If they disagree, one is wrong and the reader will find it.
- Conceding precisely
- Agreeing a specific weak point and committing to better evidence. Costs nothing; defending costs the room.
- Documentation
- Source file, cleaning log and field definitions. What stops a dashboard becoming something nobody dares change.
Homework before the next session
Rewrite one finding answer-first
Take something you have written that builds up to a conclusion and put the conclusion in the first sentence. Note how much of the rest becomes optional.
Turn one finding into a real recommendation
Action, owner, timeframe, success test — and state whether the evidence supports a trial or a change. If it supports neither, the finding is not ready.
Present your project in five minutes
Answer, one chart, three figures, recommendation, limitations. Time it, then answer the four questions: so what, how sure, what cost, what are we not seeing.
Write the handover documentation
Source file, cleaning applied and why, what each calculated field means, and the known limitations. Enough for someone to pick it up in six months.
Assessment rubric
How this session is marked. The certificate for Data Analytics is awarded on the deliverable, not on attendance.
| Criterion | Passing | Excellent |
|---|---|---|
| Structure | Presents findings clearly. | The answer stated in the first sentence and confirmed to survive being read before any evidence, followed by evidence, recommendation and method in that order. |
| Recommendations | Suggests next steps. | One recommendation naming an action, owner and timeframe, with a success test, proportionate to the evidence and expressed as a trial where the evidence is thin. |
| Presentation | Presents within the time. | Five minutes with one chart and three figures, prepared answers for so what, how sure, what cost and what are we not seeing, and any weak point conceded precisely. |
| The deliverable | Delivers a dashboard and a note. | A one-screen dashboard over the cleaned dataset with connected slicers and reset instructions, plus a one-page note whose every figure traces to the dashboard, verified by an outsider answering a question unaided. |
| Handover | Work is complete. | Source file, cleaning log and field definitions documented, every figure said aloud and re-checked before sending, and the note revised in response to the strongest challenge received. |
Session questions
How long should the analysis note be?+
One page: the answer, three supporting figures with counts, one recommendation with an owner, a timeframe and a success test, and four sentences of limitations. The constraint is the discipline — it forces you to decide what actually matters, and a manager will read one page but not six.
Should I show all my work in the presentation?+
No. Assume five minutes and one question: answer, one chart, three figures, recommendation, limitations. Everything else is preparation for the questions rather than content for the presentation, and a manager who wants more will ask.
What do I say when a manager asks so what?+
Your recommendation, which should already have answered it. If that question is a surprise, the presentation was structured in the wrong order — you built up to the finding instead of leading with it.
I was challenged and I think my finding is weak. What now?+
Concede it precisely and commit to better evidence: that is a fair point, the sample is too small for that claim, and I will come back with a full quarter. Defending a weak finding costs you the room; conceding costs nothing and makes everything else you said more believable.
Do the dashboard and the note have to match exactly?+
Every number in the note must be traceable to the dashboard, and the note's headline must be what the dashboard makes visible. If the note claims growth the dashboard does not show, one of them is wrong — and the reader will find it before you do.
This session is part of
Data Analytics
4 weeks · 8 sessions · ₦50,000 · you leave with a dashboard and an analysis note