Data-Driven Culture by Gustavo Santos

Data-Driven Culture · by Gustavo Santos

Reality dá signals.

All the time.

  • someone searched and found nothing
  • the project was three days behind
  • a complaint came back
  • a machine stopped for the second time
  • the client quit before paying
  • Tuesday sells different from Thursday
  • one time slot went empty
  • a cost started climbing

Your company is already talking.
The question is: are you listening?

Scroll slowly

One sentence

Data-Driven Culture is learning to look at what is happening, ask better questions, and use the answers to make better decisions.

The formal definition

“The institutional capacity to turn signals from reality into strategic assets that guide decisions, reorganize operations and generate value.”

Capacity: something the organization can do repeatedly. Institutional: not dependent on one person. The rest of this page explains each word with examples.

Truck off-road parked in an open-pit mine at dusk, with an amber light on

Before data, there is reality

A machine stopped for the second time this week.

That is a signal from reality. It happened before any spreadsheet, system, or dashboard. What happens to it now depends on the organization.

Five destinations for a signal

Perceived

Someone notices, writes it down and compares it to last week.

The signal becomes data. And data can become a question.

Ignored

Nobody notices. The machine works again and the day goes on.

For the organization, it didn't happen.

Normalized

Everyone saw it. "This machine is just like that."

It stops generating reaction. Small deviations grow in silence.

Distorted

Someone records it but as "scheduled maintenance."

There is data. Only it tells a different story.

Dispersed

The operator knows. Talked about it at coffee. It's not written anywhere.

Knowledge gets locked in one person and leaves when they do.

The spreadsheet cannot see what was not recorded.

One small case for a big idea

The tapioca that nobody was selling.

An example told by iFood shows the whole chain working from someone's click to an offer that didn't exist.

  1. 01 · Signal

    In a neighborhood, people open the app and search for tapioca.

  2. 02 · Data

    Every search gets logged. Including the ones that don't end in an order.

  3. 03 · Information

    With context, a pattern emerges: there's demand for tapioca there, at certain times and hours and no one selling it.

  4. 04 · Question

    “Where is there demand that nobody is meeting yet?”

  5. 05 · Evidence

    One isolated search isn't enough. The algorithm only acts when demand reaches a volume worth pursuing.

  6. 06 · Interpretation

    It lists nearby restaurants, reads their menus and deduces who already has the ingredients. Then it looks at each one's positioning: tapioca works with a bakery, not an upscale restaurant.

  7. 07 · Communication

    The right places get the message: there's demand for tapioca in your area, you have the ingredients, here's the recipe. Just tap a button.

  8. 08 · Decision and action

    The owner decides to add it to the menu. The bakery could sell tapioca and alone, would never know there was demand.

  9. 09 · Learning

    A question that worked once can be asked again another neighborhood, another product. When this repeats, it becomes capacity.

Hear from the source

Diego Barreto, CEO of iFood, on Flow Podcast.

A 13-minute clip of the interview with examples from this page. Choose where to start:

  1. 1:07174 proprietary models
  2. 3:07Football and couriers
  3. 4:33The tapioca
  4. 8:56Credit for restaurants
  5. 11:42The supermarket test
Watch on YouTube → Full episode on Flow →

Case reported Story by Diego Barreto, CEO of iFood, on Flow Podcast (2025), starting at 4:33 of the clip (42:44 in the full episode). He describes how the algorithm works using a neighborhood as an example. Read the full chronicle →

Data is not the beginning

From signal to strategic asset.

Thirteen steps. Each one only lets you claim what the one before supports. When someone jumps straight from data to decision, you get false certainty that calls itself “data-based”.

  1. 01SignalA happening in the world, before any record. Not data.
  2. 02DataThe partial record of a signal. Not reality.
  3. 03InformationData with context: when, where, in relation to what. Not evidence.
  4. 04EvidenceInformation that has passed through reliability, comparison, method and limits. Not absolute certainty.
  5. 05InterpretationMeaning assigned with responsibility, separating fact from hypothesis. Not loose opinion.
  6. 06CommunicationEvidence translated into shared understanding. Not decoration.
  7. 07DecisionChoosing a path with priority, trade-offs and responsibility. Not a recommendation.
  8. 08ActionActually changing a practice, a process, a resource or a rule. Not intention.
  9. 09ResultReality's answer to the action. Not proof of success.
  10. 10LearningWhat is preserved from comparing what was expected to what happened. Not someone's memory.
  11. 11ValueThe utility created: margin, efficiency, less risk, more trust. Not just money.
  12. 12CapacityThe ability that remains after use and repeats. Not a one-time result.
  13. 13Strategic assetReliable data, rule or learning, governed and reusable. Not data just by existing.

The value is not in possessing data.
It is in what the organization can do with it. Again and again.

Cases the theory uses

Three stories. Three links in the chain.

Cases illuminate a principle, not promise the same result. Each brings the source and type of record.

Delivery bike parked on a wet street at night, with stadium lights glowing in the background

Case reported Link: signal → learning

The game that took couriers off the streets.

iFood has a model that predicts if a courier will be available for each order. One day, the model itself showed a relationship nobody had asked about: when Flamengo or Corinthians played, many couriers left the platform. They stopped their bike to watch the game.

The learning entered the model. Couriers started reporting their team, and the championship schedule became data. On game day, the system calls those who root for another team.

Sometimes the signal arrives before the question.

Source: story by Diego Barreto (iFood), Flow Podcast.

Listen to this clip 3:07

Case reported Link: data → strategic asset

Credit for work, not history.

The small restaurant no bank serves leaves traces every day: if it sells more or less, if it passes on inflation, if clients like it, if it delivers on time. iFood transformed that data into the capacity to grant credit.

“I don't judge him by his credit record. […] I judge the guy's work.”

According to Barreto, the portfolio now exceeds R$ 1 billion, with an average term of 18 months.

The other side. The model assumes the restaurant will stay on iFood. The same asset that gives credit to those banks refuse also increases dependence on those who receive it. Data is power too.

Source: story by Diego Barreto (iFood), Flow Podcast.

Listen to this clip 8:56

Documented case Link: action without deviation management

45min

When automation runs without anyone watching.

On August 1st, 2012, an implementation error made Knight Capital's trading system send over 4 million orders in 45 minutes. Each order was a signal something was wrong. There was no mechanism to notice and stop.

The loss exceeded US$ 460 million.

Automation speeds everything up. Including mistakes.

Source: SEC, Release 2013-222 (10/16/2013).

The right question

Pick a phrase you've heard in a meeting.

“Mondays sell less.”

How do you know?

What would we need to observe to discover:

  • Sales by day of the week, from several months not just one week.
  • Revenue or margin? Showroom or delivery?
  • Holidays and promotions in between.
  • Monday sells less or does the store open later?

“That customer left because they thought it was too expensive.”

How do you know?

What would we need to observe to discover:

  • What they said and what they did before leaving.
  • Similar clients who stayed paying the same price.
  • Complaints about service, deadline or quality.
  • When the price changed and when departures increased.

“The project ran late because of the rain.”

How do you know?

What would we need to observe to discover:

  • How many days it actually rained and on which fronts.
  • Delays that happened on dry days.
  • Material delivery, crew, project changes.
  • What the schedule assumed about weather.

Intuition isn't the enemy. It opens the investigation.
The mistake is letting a hypothesis pretend to be evidence.

And Data-Driven Culture? How do we know?

Do data-driven companies perform better?

It would be inconsistent to ask you for evidence and answer with opinion. So we went to research: one econometric study, three meta-analyses and two studies that measured culture itself.

  1. Study · 179 companies

    5–6%

    higher productivity

    Large companies that took data through to decision made 5 to 6% more than expected, even accounting for their other tech investments.

    Brynjolfsson, Hitt & Kim, 2011 →

  2. Meta-analysis

    125

    studies · 26 countries

    Analytics connects to better operational, financial and market performance. People, management and culture weighed more in value than technology.

    Oesterreich, Anton & Teuteberg, 2022 →

  3. Meta-analysis

    107

    studies

    Technical and social factors matter. Humans and organizational structure come first.

    Oesterreich et al., 2022 →

  4. Study · Turkey

    132

    companies

    Data-driven culture bridges analytical capacity with operational and financial performance.

    Karaboga et al., 2023 →

  5. Study · MSMEs

    149

    small and medium

    Perceiving change and analyzing improved operational performance. Data-driven culture amplified the effect.

    Wong & Ngai, 2023 →

  6. Meta-analysis

    34

    studies

    The effect is positive, and part of it comes through agility: the company needs to be able to react to what analysis reveals.

    Liu, Singh & Al Shibli, 2025 →

The pattern

  1. Technology alone not enough.
  2. Analytics helps.
  3. Analytics + people + management + culture is where value appears.

The number that circulates wrong

“Data-driven companies are 19 times more profitable.”

The source is a 2013 McKinsey survey of 400 executives about one type of analysis: customer analysis. The most intensive users were 19 times more likely to report above-average profitability. It's strong data, but self-reported, comparing extremes and without proving cause.

What the research doesn't say

  • That using data causes profit automatically.
  • That buying technology improves performance.
  • That the effect is the same for every company. None of the studies has a Brazilian sample.
Read the full brief →

All of this can help.

But none of these things is Culture Data-Driven.

  • A dashboard shows a deviation. Doesn't guarantee someone decides.
  • An automation speeds up a flow. Speeds up mistakes too.
  • An AI answers fast. Doesn't guarantee the question made sense.
  • An analysis delivers evidence. Doesn't guarantee it crosses people.

Technology scales the culture that already exists. In mature culture, it amplifies intelligence. In fragile culture, it amplifies noise.

The question stays the same:
what did this information change?

The two engines

Seeing isn't enough. You have to not forget.

Intelligence Engine

Reads reality and transforms it into action.

Perceive, record, build evidence, interpret, communicate, decide, act, learn.

Sustainability Engine

Prevents intelligence from getting lost.

Documentation, memory, traceability, patterns, accountability, continuity.

Intelligence without sustaining becomes an episode. Sustaining without intelligence becomes bureaucracy.

You know that spreadsheet only one person understands?
The “everybody knows that's how it works” that nobody ever wrote down?
The AI prompt only one employee knows how to use?

This has a name: captive intelligence.

The problem isn't the person who knows. It's the organization that doesn't know without them.

Sized to each reality

Same logic. Different sizes.

Data volume changes. Tools change. Context changes. The journey from signal to decision stays the same.

Open notebook on a bakery counter, with pencil tally marks
A bakeryThe bakery that notes and compares already crosses the chain. It doesn't require scale: it starts with recording.
Construction site at dawn, with rebar, wooden forms and a hard hat resting
A construction siteIn physical operations the signal is in the world: deadline, waste, downtime. The challenge is recording what isn't born digital.
Waiting room of a public service, with empty chairs and paper tickets on the floor
A public serviceLines, deadlines, demand, repeat cases. The value changes definition efficiency, transparency, service. The chain works the same.
A digital platform

Data is already born digital and at scale. The challenge becomes governance, asset and power.

A family

Schedule, school, health, money. Life also produces signals and better decisions fit at home.

A mature bakery and a mature bank don't look alike. Maturity isn't the size of the data structure. It's the right dose for the need.

Illustrative scenarios The images on this page were created to illustrate everyday signals. They don't portray the cases mentioned.

Market signal · Civil construction

The construction site started generating data.

Robots from Chinese BrightMaster Robotics arrived in Brazil through Grupo Minas Brisa. They paint, sand, level concrete, carry up to 420 kg, and scan the site in 3D.

Before painting or leveling, the robot needs to see. It scans the site, transforms reality into data, and only then AI decides what to do.

3D
site scan in up to 400 m radius
420 kg
of material per trip, on the logistics robot
3 a 4
equivalent person-days of work per robot, according to Grupo Minas Brisa

Signal → data → decision → action. The robot scans, compares what was built to the project, plans the route and executes. And each task creates new data: square meters, time, material. The site starts being measured while being built.

Read the full story →
Schematic floor plan of a floor being scanned A central scanner sweeps the walls and records points. One wall appears displaced from the project, marking a deviation. project executed deviation
Illustration: the robot scans the floor, compares what was executed to the project and shows where there's difference. In Data-Driven Culture theory, this has a name: deviation management.

Manifesto film

Some ideas need to be seen.

In production

The Culture Manifesto Data-Driven.

The story will be told here.

Culture Data-Driven · Edition No. 1

September 25, 2026

Market signals

The construction site started generating data

Chinese robots that paint, sand, level concrete and scan the site arrived in Brazil. To work, they need one thing above all: data.

4 min read →
  1. 02EvidenceDo data-driven companies perform better?It would make no sense to ask you for evidence and answer with opinion. So we went to the research. What it says is more interesting than “data makes money”.
  2. 03Signals from realityThe tapioca nobody soldThe customer wanted. The bakery could make. The problem was nobody had introduced the two.
  3. 04FundamentalsData-Driven Culture without the fussIt is not Power BI, it is not AI, it is not having lots of data. It is turning reality into consequence.
  4. 05DecisionStop deciding on ‘I think’Experience is too valuable to discard — and too dangerous to become proof without investigation.
  5. All texts →

While you read this page,
your company kept producing signals.

Which of them are you
letting pass?

Data-Driven Culture by Gustavo Santos

Learn to see.
Learn to ask.
Learn to decide.

Start with the tapioca → Read the blog