The data is there. The story isn't.

·Opening

From a pile to a decision.

Almost every leader we talk to has a version of the same complaint. The data is everywhere, but whenever they need an answer, the answer is never ready. System exports pile up, branch spreadsheets multiply, daily recaps sink in WhatsApp groups. The materials are already here. They're just not in a shape anyone can use.

The road from that pile to a decision breaks into six stages. Data: the raw material. Sorted: grouped by definitions everyone agrees on. Arranged: ordered around the question you want answered. Visual: shown so the eye can compare. Story: given one "so what" sentence. Decision: someone does something differently.

Most data budgets get spent on stage four: dashboards, BI tools, beautiful colors. Yet the stages that decide the outcome are the invisible ones: agreeing on what a word means, choosing a question worth answering, analyzing before you display, and deciding before you look at the data. A pretty chart does not automatically produce insight.

This essay walks through all six stages: what each one means, how it leaks in real businesses, and what you can do about it. Tap each stage below, or let it play on its own.

01Data

The problem isn't too little data.

Count the data sources in your business right now. Usually there's a POS or cashier system, accounting software, one or two operational apps, a spreadsheet per branch, a few Google Forms, and a WhatsApp group where the daily recap arrives as a photo or a long message. In clinics and labs, add a hospital information system or electronic medical records, a lab system, and the national insurance portal.

That isn't a shortage of data. That's a pile.

In Indonesia, the pile has its own character. A survey by OCBC and NielsenIQ found that 77% of business owners who keep books still do it by hand¹⁸. Even at the national level, the Ministry of Health admits that integrating some medical records into its SATUSEHAT platform "is still manual, through periodic matching"¹⁹. If the national health system still reconciles data by hand, it's no surprise that mid-sized businesses do too.

Research from AWS and Strand Partners this year adds a sharper number. Of the Indonesian businesses already using AI, only 21% have data governance, and only 19% have a clear way to define success¹⁷. The tools have been bought. The foundations haven't been laid.

MythEven the statistic about messy data is messy

You may have heard that data scientists spend 80% of their time cleaning data. That number turns out to come from a small 2016 survey in which two categories were added together, and it has been questioned for years². Anaconda's larger survey (3,493 respondents) gives a more grounded figure: about 38% of time goes to preparing and cleaning data¹.

38% is still a lot. But the lesson is more interesting than the number: even a statistic about messy data was born from data that wasn't sorted properly. That theme will keep coming back.

SymptomsSigns your business is still at stage 1

• Reports depend on one person. When the finance admin takes leave, the monthly report takes leave too.
• "I'll pull the data for you" takes three days. A simple question from leadership turns into a small project.
• Important numbers live in chat. The daily branch recap is sent as a photo, then sinks beneath the stickers and good-morning messages.

What the teams that get it right do: they don't start by buying a tool. They make a simple list: every data source, who owns it, how often it's updated, and whether it can be exported. The list usually fits on one page, and it almost always surprises the leaders who asked for it.

You can't ask a pile of data a question. You can only ask data that people have agreed on.

02Sorted

One word, three numbers.

Sorting sounds mechanical: group things by color. In a business, sorting is political. Before data can be grouped, everyone has to agree on what each group means.

Example: a Monday meeting with three revenue numbers. At one corporate health check-up (MCU) provider, the Monday morning meeting opened with a simple question: what was MCU revenue last month? Sales said IDR 4.6 billion. Operations had IDR 4.3 billion. Finance said IDR 4.1 billion. Forty minutes later, the meeting ended without discussing anything except which number was right.

Here's the interesting part: all three were right. Sales counted the value of contracts signed that month. Operations counted participants already examined, times the package price. Finance counted invoices issued after results reached the client's HR team, minus participants who cancelled. Three reasonable definitions, all using the same word.

This isn't a small-business problem. Airbnb, with thousands of employees and a world-class data team, went through exactly the same thing. Its Data Science and Finance teams would sometimes give different answers to a simple question like which city had the most bookings last week, because they used slightly different tables and metric definitions⁵. Their answer was Minerva, a platform that now holds more than 12,000 metric definitions under one principle: define once, use everywhere⁶.

A detail that rarely gets mentioned: Minerva grew out of the A/B testing team. There, a wrong definition leads straight to a wrong decision, and everyone can see it. Only afterwards did the discipline spread to everyday reporting⁶. The lesson: definitions get fixed where a decision is forced. If no decision depends on a number, nobody cares whether it's defined correctly.

TemplateA one-page metric dictionary

You don't need Minerva. The cheap version is one page, signed by finance and operations. Start with the five numbers that get argued about most in meetings.

Before data can be sorted, everyone has to agree on what each group means.

03Arranged

From the system's order to the question's order.

Arranging is the stage teams skip most often, because it feels unnecessary: the data is already tidy. We think the opposite is true. Arranging is the stage where the question enters for the first time.

The difference is this. Sorting is about consistency: everyone counts the same thing the same way. Arranging is about relevance: of everything that's now tidy, what goes up front, and how the pieces relate. Arrange without a question and you get a pivot table. Arrange with one and you get an argument.

Example: 23 sheets arranged by system. A clinic group sent its board a monthly report as a single Excel file with 23 sheets. The order followed the system's modules: registration, cashier, lab, pharmacy, and so on. Everything was correct and complete. But the director only had one question each month: which branch needs help this month, and why? To answer it, he had to open seven sheets and do the math himself.

The same report, rearranged around that question, fit on one page: the three branches with the biggest changes, one line on the cause for each, and links to the detailed sheets for anyone who wanted to dig in. No new data. Just a new order.

None of this is new. Barbara Minto wrote it down at McKinsey in the 1970s as the Pyramid Principle: the answer on top, supporting arguments beneath it, data at the bottom. Amazon took it further. In 2004, Jeff Bezos banned PowerPoint from leadership meetings, arguing that the narrative structure of a memo "forces better thought and better understanding of what's more important than what, and how things are related"⁷. That's an almost perfect definition of the arranging stage.

Bezos was also honest about the cost: great memos are written and rewritten, set aside for a couple of days, then edited again with a fresh mind. "They simply can't be done in a day or two"⁷. Arranging is expensive. That's why it's the stage people skip most.

TemplateThe one-page Monday report

1. Line 1

   This week's question, written as a question. If you can't write it, the report isn't ready.

2. Line 2

   The short answer, in one or two sentences. A reader who stops here still gets the point.

3. Block 3

   The three things that changed most, each with one line on the cause. Not ten, not everything.

4. Block 4

   One chart that supports the answer, with a conclusion as its title.

5. Appendix

   Detail by system for anyone who wants to dig in. This is where those 23 sheets stay useful.

Arrange without a question and you get a pivot table. Arrange with one and you get an argument.

04Visual

Charts that get read, not displayed.

This is the stage most often mistaken for the finish line. Once there's a dashboard, the data project is considered done. Reality says otherwise. BARC's survey of more than 1,000 users found that only 25% of employees actively use BI tools. In large enterprises it drops to 16%⁴. And that one-in-four figure has barely moved across seven years of BARC tracking³.

Example: the 40-widget dashboard. In agency life, the campaign dashboard is part of what you sell to the brand. One of them had 40 widgets across six tabs: impressions, reach, CTR, engagement by platform, all in beautiful colors. We opened the access log: over eight weeks, the dashboard had been opened 11 times. Seven of those were by the agency team itself. The replacement was much simpler: three charts that answered the brand manager's actual question (is this campaign moving sales in the areas we targeted?), sent as an image to a WhatsApp group every Monday at 07:00. No login needed. By the second week, the brand manager started replying with questions.

Why can three charts beat forty? Part of the answer is in perception research that's now 40 years old. William Cleveland and Robert McGill tested how accurately people read different ways of showing numbers. The result: we're most accurate comparing position on a common scale, then length, then angle, then area, and worst at volume and color⁸.

Which means a plain bar chart is almost always more honest than a 3D chart, a layered pie, or a beautiful illustration. And a story should never replace the chart. It should stand on top of it.

But there's another side. Research by Michelle Borkin and colleagues at Harvard, covering more than 2,000 visualizations, found that plain bar and line charts are the easiest to forget. What people remember are visuals with color and objects humans recognize⁹. So there are two different jobs: an accurate chart for making the decision, and one memorable anchor so the decision isn't forgotten after the meeting. Decide which job each chart is doing.

Stephen Few, who has written a lot about dashboard design, has a definition worth pinning to the wall: a dashboard is a visual display of the most important information needed to achieve one or more objectives, fitting on a single screen so it can be monitored at a glance. A dashboard with no objective and no owner isn't a dashboard. It's decoration.

A dashboard with no owner and no question isn't a dashboard. It's decoration.

05Story

One "so what" sentence.

The fastest way to test whether a report has a story: read its chart titles. "Sales by branch, Q3" is a topic. "Branch E is 46% below average, on the same rent as the others" is a story. Cole Nussbaumer Knaflic calls it the Big Idea: one sentence that states your point of view and what's at stake¹¹.

Stories work because people remember stories far better than numbers. That's their strength. It's also their danger.

Daniel Kahneman showed that our confidence in a conclusion depends more on how coherent the story is than on how much evidence supports it. And the most coherent stories are easiest to build when we know less¹². Small samples produce the most convincing stories. That's why data stories need one extra discipline.

DisciplineName the number that would break the story

Every story in a report should come with one sentence: "This story is wrong if ...". For example: "Branch E is behind because of its opening hours. This story is wrong if branches with the same hours are also behind." That one sentence turns a convincing opinion into a hypothesis you can check.

Example: 94% on time, and the biggest client is furious. An MCU provider tracked one number on its main dashboard: results delivered to the client's HR team within the promised time, steady at 94%. Green, every week. Then its largest corporate client sent a formal complaint and started a new tender. When the 6% of late results were broken down, 61% of them belonged to that one client, from the same onsite screening location. The average was right. The story was wrong.

Jeff Bezos has a rule for exactly this: "When the data and the anecdotes disagree, the anecdotes are usually right." Not because data doesn't matter, but because something is wrong with the way you're measuring¹³. He once proved it in the middle of a meeting. The metrics said customers waited less than 60 seconds to reach customer service. He called the number from the conference room. They waited in silence for more than ten minutes.

The complaint from the biggest client is an anecdote. The 94% average is data. The anecdote was right.

A good story makes a number memorable. An honest story names the number that could break it.

06Decision

The stage most often forgotten.

Many data projects end with a beautiful dashboard, and stop there. Yet all data work should end in one thing: someone decides to do something differently. Otherwise those six stages just produce a prettier report.

And this is where the numbers are most worrying. 67% of CEOs in KPMG's survey admitted they had ignored insights from data analysis because the insights contradicted their intuition¹⁰. A study by Accenture and Qlik of 9,000 employees found that 48% often fall back on gut feeling when making decisions, and only 21% are confident in their data literacy¹⁴. An Oracle survey of more than 14,000 people found that 70% had given up on making a decision because there was too much data¹⁵.

That last one matters. The problem isn't always too little data. Sometimes it's too much, with no single question holding it together.

ToolDefault action: decide first, then look at the data

Cassie Kozyrkov, Google's first Chief Decision Scientist, teaches a habit that changed how we read reports. Before looking at any data, set a default action: what you'll do if you get no new information at all¹⁶. The data has only one job: to tell you whether to switch away from that default.

It sounds simple, but the effect is big. The meeting stops asking "what does the data say?", which can be answered a thousand ways. It starts asking "is the data strong enough to move us off our original plan?", which has a yes or no answer.

ToolOne-way doors, two-way doors

Bezos splits decisions into two types. One-way decisions are hard or impossible to reverse, and deserve to be made slowly, with plenty of deliberation. Two-way decisions can be reversed, and should be made quickly by individuals or small teams with good judgment. The trouble is that as organizations grow, they tend to treat every decision like a one-way door²⁰. For two-way decisions, he suggests that around 70% of the information you wish you had is enough²¹.

For data reporting, the implication is clear: not every decision needs all six stages. Changing one branch's opening hours for a month is a two-way door. Closing that branch is a one-way door.

ToolPaired numbers, so the number doesn't get gamed

The moment a number is used for decisions, and especially for bonuses, it starts to change behavior. Marilyn Strathern put Goodhart's law in its famous form: when a measure becomes a target, it ceases to be a good measure²². Andy Grove, the former CEO of Intel, had the antidote: pair every indicator with its counter-indicator. If you measure inventory levels, also measure stockouts. If you measure speed of service, also measure complaints.

Example: a decision in ten minutes. After the clinic's Monday report was rearranged, we added one small block at the bottom: a decision agreed before the data came out. "If new patients at any branch drop more than 15% for two weeks in a row, the branch head and the operations team meet within three days to review doctor schedules and opening hours." When Branch C dropped 18%, there was no debate about whether to act. The meeting took ten minutes.

TemplateDecision brief

A story that doesn't change what anyone does is still just a report.

07Direction

Flip the direction.

Russell Ackoff's DIKW pyramid (data, information, knowledge, wisdom) and plenty of data strategy slides all share one thing: they flow from the bottom up. Collect the data, tidy it, and insight will emerge.

Martin Frické, a professor of information science, wrote a pointed critique of this. In his view, the idea that information "rises" on its own out of data is methodologically wrong. Without a question or a theory at the start, nothing rises²³.

Our experience in the field lines up with that. Teams that build reports successfully work the other way around. They start from the decision that has to be made, move down to the question that answers it, then to the numbers they need, and only at the end touch the data sources. As a result, they only sort and arrange the data that actually gets used.

This also explains why so many data projects feel unfinished. Start from the bottom, and there's always more data that hasn't been cleaned. Start from a decision, and the project is done when that decision can be made better. There's a finish line.

Start from the data, and there's always more data to tidy. Start from the decision, and there's a finish line.

08Limits

Where the pattern bends.

These six stages are a pattern, not a mandatory procedure. Each one has limits.

1. 01While exploring

   Don't rush to a story while you're still searching. Stories force coherence, and exploration needs patience with things that don't make sense yet. Use a story to communicate a decision, not to find one.

2. 02Small decisions

   Two-way decisions don't need six stages. For anything you can undo next month, use enough data, decide, measure, and roll back if you were wrong.

3. 03Data-informed

   Data optimizes the strategy you're already running. It doesn't pick a new one. Andrew Chen, drawing on Adam Mosseri, says metrics are merely a reflection of the product strategy you have in place²⁴. For decisions about direction, data helps, but it doesn't decide.

4. 04Spreadsheets

   Spreadsheets aren't the enemy. Field audits find errors in most real spreadsheets, but the error rate per cell is small, roughly 1 to 5%²⁵. The problem isn't the tool. It's one wrong cell in the place a decision depends on. In 2020, 15,841 Covid cases in England went unreported because an old Excel file format hit its row limit²⁶.

5. 05One place

   Bringing data together also brings risk together. Since 17 October 2024, Indonesia's Personal Data Protection Law No. 27/2022 has been fully in force. Health data counts as specific personal data²⁷. One place for all the data is also one place to protect: role-based access, activity logs, and only the data that's actually needed.

09Self-check

Which stage does your Monday report stop at?

One question per stage. Answer honestly for the report your business uses most.

The first "not yet" from the top is where your report stops. Fix that one first. The stages below it almost always improve along with it.

10Closing

Closing.

A pile of data does need to be sorted, arranged, visualized and told as a story. But that road often looks like a matter of presentation, when the hardest parts are invisible: agreeing on what a word means, choosing a question worth answering, and deciding before you look at the data.

If I had to boil it down to one habit: start from the decision, then build only the data you need to make it better. Everything else can wait.

The data is already there.
What's missing is the decision waiting for it.

This is the work we do at AksesData: taking the data your business already has, from system exports to branch spreadsheets, and turning it into leadership reports that start from a question, not a table. See how AksesData works.