Data governance first, then AI

Broad adoption, little scale

McKinsey’s global survey on the state of AI, conducted in June and July 2025 with 1,993 participants, found that 88% of organizations use AI in at least one function, ten points more than in 2024, and that 79% use generative AI. But only 7% said AI was fully scaled across their organization (McKinsey).

An AI model is only as good as the information that feeds it, and the information of most mid-sized companies is not ready.

The distance between using and scaling has many causes, but in the mid-sized company one dominates: data. An AI model is only as good as the information that feeds it, and the information of most mid-sized companies is not ready.

The data exists, but it isn’t usable

In our work with a range of clients we find a constant pattern. The data exists: in the ERP, in spreadsheets, in advertising platforms, in quality reports, in people’s knowledge. What is missing is for it to be useful for making decisions.

At a food services business group, building baselines for the business cases required reconciling figures that different areas reported differently. Non-conforming product was recorded without a classification that would allow root-cause analysis. At a consumer e-commerce company, there were no automated reports or dashboards to assess performance in real time, and market trends were reviewed without a process for incorporating them into the strategy.

In both cases, the conclusion was the same: before talking about AI, we had to talk about data governance.

What data governance means in a mid-sized company

Data governance sounds like a complex corporate program. In a mid-sized company it can be much simpler and start with three decisions.

The first, defining the critical data: the twenty or thirty indicators and entities (customer, product, supplier, order) the company uses to make decisions. The second, assigning an owner to each one, responsible for its definition and quality. The third, establishing a single source: the system where each piece of data is recorded only once and from which reports are fed.

In the business group’s transformation portfolio, data governance was designed as an enabling initiative with a specific information architecture role. Its value was not measured directly, but by the impact initiatives it made possible.

From data to decisions

With governed data, the next step is not AI, but visibility. An executive dashboard that brings the key information together in one place changes the way a company is managed. Problems are seen earlier and meetings discuss facts instead of perceptions.

In designing a business-strengthening program for an industry association, the use of data made possible something that had not been possible before: characterizing a broad base of entrepreneurs by maturity level and prioritizing support according to their potential. The segmentation, built with diagnostic data, declared revenue, export experience and product alignment, made it possible to design differentiated support paths. Without that data, the program would have treated all entrepreneurs the same.

When AI makes sense

With governed data and visibility, AI becomes a practical tool. The use cases with the best effort-to-value ratio in the mid-sized company tend to be history-based demand forecasting, detection of deviations (in input consumption, in expenses, in quality), automatic classification of unstructured information and assistants for the team’s repetitive tasks.

The recommendation is to start with a use case that has a business owner, available data and a clear indicator. A successful pilot with those conditions gets scaled; a pilot without them stays a demo.

Data protection

Data governance includes data protection. In Colombia, Law 1581 of 2012 regulates the processing of personal data. Any data initiative involving customer, employee or supplier information must have authorization, a defined purpose and security measures.

Data culture

Data governance cannot be sustained by rules alone. It needs a culture in which decisions are made with data and challenged with data. In our experience, that culture is built from the top: when management asks for the figure behind every proposal and reviews the dashboards in meetings, the rest of the organization learns fast. When management decides by intuition, the dashboards are abandoned.

A simple indicator of data culture is how much time management meetings spend debating whether a figure is correct. If it is a lot, the problem lies in data governance. If it is little and the discussion focuses on what to do with the figure, the organization is ready for the next level.

Team capabilities

A mid-sized company doesn’t need a large data team to get started. At a minimum, it needs one person with analytical skills who understands the business, someone responsible for the information architecture who ensures data consistency, and access to specialized capabilities (data science, data engineering) when required, whether through in-house staff or partners. In the project teams we put together, combining a business intelligence profile with a data scientist and a business analyst has been enough for most of the mid-sized company’s use cases.

What doesn’t work is delegating data to the technology area without the business taking part. Data is a business asset; technology provides the infrastructure, but definition and quality are the responsibility of those who use it.