Is your data ready for AI? How to assess it in your SMB

Is your data ready for AI? The question cannot be answered in the abstract, it gets answered against a specific use case. If, for that case, the information exists, can be looked up, is complete and in a consistent format, has someone in charge of it, covers the time span the task needs, and can be handled legally, you have solid ground for a test. Gartner projects that through 2026, 60% of AI projects that are not backed by ready data will be abandoned, and checking this beforehand lowers that risk considerably.

Why did this become the bottleneck?

Because tools got cheaper and information did not get organized at the same pace. In a service SMB, the typical situation is that the data that matters is spread across emails, spreadsheets, the inboxes of three different people, and systems that do not talk to each other. AI ends up working from an incomplete picture and proposing reasonable things for a reality that is not yours.

What follows is that review: one question for each condition, about where the information lives, who can look it up, how it is recorded, who is accountable for it, how much history it covers, and whether it can be handled legally. Answer them with a specific case in mind, not with your company's data in general, and you will know before investing whether your data can hold up a first project.

Six checks help determine whether the data is truly ready to support an AI initiative.

What are the six questions?

1. Does the information exist anywhere?

For the case you chose, whether it is proposals, sales follow-up, internal knowledge, or reports, the basic data has to be recorded somewhere you can look it up: a system, a spreadsheet, a shared file. If it lives in scattered emails and in people's memory, for practical purposes it does not exist.

2. Can the people who need it look it up?

Existing is not enough. The test is whether someone on the authorized team can reach that information without asking a specific person for a favor. When every lookup goes through the same person searching and exporting, that is not access: it is a dependency.

3. Is it complete and in a consistent format?

The fields the case needs should be filled in on most records and written the same way. The opposite is easy to spot: frequent blanks, dates written three different ways, and the same company recorded as "S.A. Ltda", "SA Ltda", and "Sociedad Anonima". Perfection is not required; what is required is that the same data point is always written the same way so a pattern can be read.

4. Is anyone accountable for that data?

Every record should make clear who created it, who maintains it, and who to ask when something looks off. Orphan records, the ones nobody knows are still current, are the ones that later produce wrong answers stated with full confidence.

5. Does it cover the time span the task needs?

There is no fixed number of months. The history you need depends on how many records you generate, on whether your operation has marked seasons, and on the weight of the decision you are going to make with them. What you have to check is whether the available history includes the variation that matters. A system change that left six months incomplete can invalidate the comparison even if there are years of records.

6. Can it be legally shared with the tool?

When personal data is involved, the processing has to be reviewed under Ley 8968: on what basis it is processed, for what purpose, with what controls, and in which authorized tool. If that is not clear, or it is not clear what the vendor does with the information afterward, the answer is not yet.

How do you read your answers?

With six yes answers, it makes sense to validate a controlled test with what you already have. With four or five, organize what is missing and check whether any of the gaps blocks everything. With three or fewer, preparing the data is the project, not the step before it.

Which one is missing matters more than how many are missing. A legal or access problem can stop the case even if everything else is flawless, and that is why the count is an orienting reading of this guide, not a traffic light.

The before-and-after view shows the value of cleaning, structuring, and centralizing information before implementing AI.

Do you need to build a data platform?

Almost never. What you have to prepare is what the case requires, with the simplest technical setup that meets access, quality, security, and traceability. Sometimes it is enough to organize a source that already exists. Sometimes you do need to integrate two systems. Starting with the big platform is the most expensive way to find out that the case was the wrong one.

McKinsey reported in 2025 that only 21% of the organizations using generative AI had fundamentally redesigned any of their workflows. Something similar happens with data: the redesign usually starts by clarifying where each thing comes from and who is accountable for it, not by buying infrastructure.

In an SMB, having your data ready almost always looks like a well-structured spreadsheet with someone in charge, and very little like a technology project.

Where does AI help you get organized before you start?

In three preparation tasks that are tedious for a person and cheap for a machine: matching the formats of names, dates, and industries in an existing spreadsheet; sorting loose records, such as prospects or cases, into consistent categories; and spotting duplicates that look different but are the same company. None of the three is the project: they are the step before it when there is organizing left to do.

It is worth clarifying what does not count as ready data. Having a modern customer system does not guarantee it, because the software brand says nothing about the quality of what is inside. Neither does the team knowing where everything is, because that is a dependency on people. And least of all the promise of fixing it along the way: what was not organized beforehand piles up and ends up explaining why the project produced no results. Data quality is not a matter for the technical area either, it belongs to whoever uses the data to decide.

The companies that arrive with their data ready tend to do three things: they go through the six points before buying the tool, they have named the person in charge of that data with time assigned to maintain it, and they review consistency on a defined schedule instead of looking at it at the end. For the full filter that comes before this one, there is "What to assess before opening an AI project in your SMB".

Do you want to know if your case has solid ground?

VegasiO's Discovery web takes 5 minutes and reviews the state of your data against the case you want to tackle. At the end it tells you whether the next step is a Diagnóstico de Adopción IA to get organized first, an Implementación de IA with what you already have, or spending some time consolidating before you invest.