Data maturity: what level is your company at?

A clinic group with eleven locations bought a BI licence in January. By March the dashboard was live: revenue per location, appointment occupancy, average ticket, all very tidy. In July the director asked why two locations had dropped. The first answer arrived twelve days later. The second, from another team, took fifteen days and did not match the first.

The dashboard was not the problem. The company had tried to skip three rungs at once: it bought an advanced tool and kept beginner processes. The tool simply made the mismatch more visible, and more expensive.

Data maturity is not what you have installed. It is what your company can answer, how fast, and with how much confidence. You can measure that today, with no consultant and no sixty-question assessment.

What data maturity actually means

Data maturity is the degree to which an organisation can turn data into decisions reliably and repeatably. It is usually read in five levels, from reactive to operational. The level is not set by the tools you own, but by the time and confidence with which the company answers its own questions.

Pay attention to the word repeatable. Any company can produce one brilliant analysis when the director asks and three people drop everything. Maturity is doing it again next week, with a different question, without heroics.

The one-question test

There is a diagnostic shortcut that beats any questionnaire. Pick a question your company genuinely needs answered, then time it. A good question is specific, has a number in it, and is not trivial. For example: why did sales fall 12% in July?

What happens after that question reveals your level with uncomfortable precision.

Notice that level 1 is not slow. It never gets as far as slow, because the question is never asked. The drop shows up in the cash position in September and becomes an emergency meeting.

The five levels of data maturity

Level 1. Reactive

The data exists, but nobody asks for it. Decisions rest on experience, on whoever speaks loudest in the meeting, and on what a competitor just did. Reports appear only after something has already gone wrong, and they exist to explain the past to an annoyed shareholder.

Clearest signal: the company knows its revenue, because the accountant reports it, and has no idea where that revenue came from.

Level 2. Manual

Someone consolidates the numbers. Usually an analyst or the manager themselves, at month end, pulling exports from five or six systems into a master spreadsheet. The number exists, but it arrives late, and every team has its own version because each one cut the period differently.

This is the most expensive level and the one least recognised as expensive. A simple simulation: three people spending 6 hours a month on the close add up to 216 hours a year. At $30 an hour, that is $6,480 a year spent purely on producing the number, before any analysis at all. And the work starts from scratch every month.

Level 3. Consolidated

There is a single source. Definitions are agreed in writing: what counts as a lead, when a sale lands in a month, which revenue is gross and which is net. The dashboard refreshes on its own and nobody has to push it.

The bottleneck has moved. Recurring numbers are solved, but every new question joins a queue that runs through one person. The sales team waits two days for a breakdown the dashboard did not anticipate.

Level 4. Analytical

A new question meets the data the same day. Whoever has the doubt can slice it themselves, without filing a ticket, because the data layer is available in language their team understands. This is where a company starts measuring the effect of its own decisions, not just the outcome.

The practical gap between level 3 and level 4 is rarely technological. It is about permission and habit, which makes it a question of data culture far more than of software.

Level 5. Operational

Data leaves the report and enters the process. A conversion deviation fires an alert, a stockout forecast generates a purchase order, a churn score reorders the retention team's queue. The person is warned before they think to ask.

One warning: this level charges maintenance. A model that decides badly has to be corrected, and an alert that fires too often is ignored within three weeks. Level 5 with nobody responsible for calibration turns into automated noise.

A fifteen-minute self-diagnosis

Instead of scoring abstract pillars, look for observable behaviour. You already know the answer to every line below without asking anyone.

Symptom	Likely level
Nobody can state last month's customer acquisition cost	1
Two teams show different numbers for the same metric	2
The report is late whenever the person who builds it takes leave	2
A dashboard exists, but any question outside it becomes an analyst request	3
The media manager opens the data and cuts it by campaign themselves	4
The company records what it decided and returns later to check the result	4
A missed target arrives as a notification before the weekly meeting	5

Mark the highest level for which every earlier symptom is also satisfied. If you show level 4 behaviour but still live with two versions of the same number, you are not at level 4. You are at level 2 with an expensive tool.

Why skipping levels does not work

The natural impulse is to buy the next level's solution. Almost always the problem sits one level below, and the purchase only brings the frustration forward.

Market data shows the size of that gap. The TIC Empresas 2025 survey by Cetic.br, released in June 2026, found that AI use among Brazilian companies rose from 13% in 2024 to 17% in 2025, with small businesses at 15% and large ones at 50%. Adoption is accelerating, but the foundations are not keeping pace: the same survey records that only 31% of companies used a CRM in 2025 and 36% paid for cloud processing capacity.

The study Unlocking AI's Potential in Brazil 2026, run by Strand Partners for AWS with 1,000 business leaders, found the same pattern at a different scale: half of Brazilian companies use AI, but only 15% reached the advanced stage and 47% cannot accurately measure the return on what they invested. A company that cannot measure return is trying to operate at level 5 without having closed level 3.

A beAnalytic study of more than 130 technology leaders, published in September 2025, reads much the same: 22% of Brazilian companies use data strategically and 78% remain in the early stages of maturity.

In practice, the skipped rung fails for one specific reason. An advanced tool assumes the definitions are agreed and the source is trustworthy. When they are not, it multiplies the error with more speed and better charts. It is the same mechanism that keeps a data quality problem invisible until it becomes a wrong decision.

How to climb exactly one level

Pick one move, with a ninety-day deadline. The temptation to fix everything is what stalls most of these projects.

1. From 1 to 2: pick three numbers and produce them every month, on the same date, even by hand. Suggested set: revenue by channel, acquisition cost, conversion rate.
2. From 2 to 3: stop consolidating and start integrating. Choose one source and write the definitions on a single page. The gain comes less from the tool and more from the agreement on what each number means, which usually clears up most of your data silos.
3. From 3 to 4: take the analyst out of the path. Give teams direct access and accept that the first few breakdowns will be wrong. That is learning, not risk.
4. From 4 to 5: pick one repetitive decision and automate only that. A conversion drop warning, a stock alert, an at-risk customer flag. One. Then the second.

One detail changes the outcome: schedule the moment the number gets looked at. A figure nobody reads out loud in the weekly meeting climbs no levels at all, however handsome the dashboard.

Frequently asked questions

How many levels of data maturity are there?

It depends on the model. Most frameworks use four or five levels, and the Brazilian federal government's Data Maturity Model uses five. The count matters less than the sequence: they all describe the same progression, from deciding without data to data triggering the action.

What is the difference between data maturity and data governance?

Governance is one component of maturity, not a synonym for it. Governance is about rules: who can access what, who defines each metric, how data gets corrected. Maturity is the practical result the company achieves, and a company can have a written governance policy and still sit at level 2.

How do I assess my company's data maturity?

Time a real question and watch three things: how long it took, how many different versions appeared, and how many people had to be pulled in. Three days, one version and one person point to level 3 or above. Two weeks and two versions point to level 2.

Do small companies need to worry about data maturity?

They do, and climbing costs them less. A company with ten people and three systems reaches level 3 in a few weeks, because it has less history to undo. The bill gets expensive later, once there are fifteen sources and three parallel spreadsheets defended by three different people.

Do I need to hire a data team to move up a level?

From level 1 to 3, almost never. Those rungs are about agreement and discipline: settle the definitions, pick the source, hit the date. A dedicated team starts to matter from level 4, and even then it usually begins with one person, not a department.

The only rung that matters is the next one

Data maturity is not a badge you earn. It is the answer to a rather tedious question: how long does your company take to know what is happening to it. Companies at level 2 that accept this and climb one rung a quarter will, within two years, overtake the one that bought the most expensive platform on the market and never agreed on what a qualified lead is.

The good news is that the level 3 to level 4 bottleneck has become far cheaper to solve. Sherlok connects Meta Ads, Google Ads, GA4, CRMs, spreadsheets and SQL databases in one place and lets anyone on the team ask a question in plain language and get the finished analysis, instead of joining the analyst's queue. That is precisely the rung where most companies are currently stuck.