Hei,
Tekoäly on nostanut monet vanhat datanhallinnan kysymykset uudelleen pöydälle.
Yritykset investoivat AI-ratkaisuihin, semanttisiin kerroksiin, ontologioihin ja uusiin data-alustoihin. Samalla joudutaan kuitenkin palaamaan hyvin perustavanlaatuisiin kysymyksiin: kuka datasta vastaa, mitä esimerkiksi asiakas oikeastaan tarkoittaa ja kuinka paljon näistä asioista voidaan ratkaista teknologialla?
Keskustelimme data management- ja data governance -asiantuntija Winfried Etzelin kanssa siitä, miten tekoäly muuttaa datanhallintaa, miksi datan omistajuutta ei voi ulkoistaa ja miksi yritykset rakentavat jälleen enemmän dataosaamista organisaatioidensa sisälle.
Haastattelu on englanniksi.
Data management in the age of AI
Johannes: You didn’t exactly take the traditional route into data. How did you end up here?
Winfried: Not at all. I studied history, political science and law in Germany and Norway. My older brother worked in IT, and I actually found it quite boring.
After university I moved to Norway and started managing a shoe store in central Oslo. And suddenly there was data everywhere: sales data, customer data, inventory, purchasing, stock-outs. You had to use data to make better decisions about what to buy and what to sell.
After that I worked in archive consulting for public organisations in Norway. We were looking at questions such as what information should be preserved and what Norway as a state needs to know about itself.
That was where I started feeling that we were often coming into the process too late. Decisions had already been made, and there were gaps in the information. That gradually pushed me further towards data management.
Later I moved into consulting and oil and gas and went deeper into data management and eventually data governance.
Interestingly, today I feel I am partly returning to my original studies. Organisational design for federated data governance, for example, has a lot in common with political science. And history teaches you source criticism: Who produced this information, why was it produced and in what context? We should ask very similar questions about data.
Johannes: Data management and data governance are often used almost interchangeably. How do you distinguish them?
Winfried: They are closely connected, and you need both.
The danger is that if you only focus on data management, the discussion can become very implementation-driven: How do we do this? How do we put it into the technology stack? How do we monitor it?
Governance asks another question first: Why are we doing this?
Monitoring and enforcement are important, but you need to make sure you are enforcing things that actually matter for your organisation.
A good example is a data catalogue. The tool already contains assumptions about what a data asset is, how things should be classified or what dimensions of data quality you should use. Those might be useful defaults, but they are not necessarily the things that matter most to your organisation.
You have to make those decisions consciously instead of simply doing what the tool tells you to do.
In the DAMA wheel, data governance sits in the centre and connects to all the other knowledge areas. I would actually go even further. Governance operates on a slightly different level. Data governance should ultimately be connected with corporate governance, not treated purely as a technical discipline.
The biggest successes I have seen with data governance have been when the board of directors has become one of the stakeholders.
Johannes: I’ve seen something similar with data catalogue projects. A company starts implementing a large governance tool as an IT initiative and then leaves the business out of the project. A year later they wonder why nobody uses it. What do organisations most often get wrong when they start data governance?
Winfried: One of the biggest mistakes is to start from the technology rather than from the organisation.
The fact that we still talk about “IT”, “data” and “the business” as separate things is part of the problem. If you constantly have to say, “now we need to talk to the business”, you are already positioning yourself outside the business.
A governance tool can support the work, but it cannot decide what matters for your organisation. It cannot decide which data is critical, what your risks are or how responsibilities should be divided.
That is why I see a lot of potential in federated data governance. You bring data governance and data management closer to where the actual work happens and make them part of the business rather than something happening on the side.
Of course, pure decentralisation creates another problem. Different business units have different budgets, capabilities and levels of technical maturity. You can easily end up with fragmentation.
I often compare it with HR. Large organisations have a central HR function, but they also have HR business partners embedded in different parts of the organisation. We probably need something similar with data: central alignment combined with people working close to individual business areas.
Johannes: Data governance can also have a bad reputation. Engineers may see it as something that slows them down, while business people may see bureaucracy and compliance. How would you change that perception?
Winfried: By connecting governance much more clearly to purpose and value.
If governance becomes a list of controls, policies and approvals, people will naturally see it as bureaucracy.
But governance should help the organisation make better decisions. It should clarify what is important, who is responsible, what risks need to be managed and what needs to be consistent.
The problem is not governance itself. The problem is governance that is disconnected from the actual work.
When governance is embedded in the business and helps people solve real problems, the perception changes quite quickly.
Johannes: DAMA gives organisations a fairly comprehensive framework. But companies are never as neat as the diagrams in a book. How do you balance frameworks with pragmatism?
Winfried: There is the DAMA world, the ideal world, and then there is reality.
The real value of the DMBoK is that it gives us a shared language. We can agree on what master data, reference data, data quality or data governance mean and have a foundation for discussing them.
That becomes especially valuable when people from different organisations meet. You can say, “this is how we do it”, ask how another organisation approaches the same problem and still have a common frame of reference.
But you always have to adapt it to the organisation.
I have seen the same problem in enterprise architecture and data management: organisations try to implement something exactly according to the book, and it simply does not work.
You need to understand the reality of the business and adapt the framework to it.
That is also why I am sceptical when someone promises a complete data governance blueprint in three weeks. Data governance is always organisation-specific.
Johannes: And now AI adds another layer on top of all this. How does AI change data management and governance?
Winfried: I see two perspectives.
The first is data governance for AI.
You have to govern the inputs going into the model — the classic garbage in, garbage out problem. You also need to think about what the model is allowed to do, what constraints are required, how the output is governed and what kind of environment the models operate in.
But the other side is AI for data governance and data management.
There are many opportunities to use AI to make data management itself more effective.
The mistake would be to treat AI governance as an entirely separate problem and buy another framework or another tool that supposedly makes you compliant.
We saw the same thing with GDPR. A tool cannot understand your organisation for you. You still need to know your critical data, your risks, your regulatory environment and what the organisation is actually trying to achieve.
Johannes: One thing I increasingly see is companies starting AI projects with external AI consultants without involving their own data teams. That feels strange when the internal data team knows where the data actually is and what is wrong with it.
Winfried: External consultants can be extremely valuable.
Good consultants see patterns across organisations. They can say: “I have seen this problem before, and here is how another organisation solved it.” That outside perspective has real value.
But at the same time you have to create competence and ownership inside your own organisation.
We have seen this movie before with outsourcing. Organisations outsourced large parts of their data work and a few years later realised that nobody internally really understood the data landscape anymore. Then they had to rebuild that competence.
AI can even become a new form of outsourcing. There can be an assumption that instead of having people understand something, AI can simply do it for us.
But if you take that too far, eventually you realise you no longer own or control enough of what is happening.
Johannes: So does that also explain why we are now seeing more companies hiring data people in-house?
Winfried: I am seeing the same trend.
AI has a lot to do with it, particularly because organisations are dealing with systems that can be difficult to predict and control. You need competence inside the organisation.
There is also a lot of discussion now about context and ontologies. You may be able to buy a starting point for an ontology, but you cannot outsource the understanding of what things mean inside your own organisation.
Another interesting development is retraining.
I see large organisations taking people with 10 or 20 years of experience in areas such as finance or core business operations and teaching them the fundamentals of data.
That is a very positive sign. Organisations are realising that they need to combine deep business knowledge with data competence.
Johannes: A lot of data organisations still sit under IT. Even some CDO organisations seem to talk mostly about platforms, Databricks and infrastructure. Should data be positioned differently?
Winfried: The CDO is an interesting role partly because it can be almost impossible.
You may be perceived as an IT person while actually trying to do business work.
Where the CDO sits in the organisation matters. If the role sits under IT, it can naturally become very technology-oriented. I have also seen CDOs under the CFO, which can work very well because they are close to the financial side and the budget. In other organisations, the CDO and CIO operate at the same level.
There is no one organisational model that works everywhere.
Even sitting under IT can work very well if you manage to create strong connections to the business.
What matters is understanding your organisation, where decisions are made and how to build a storyline around what you are trying to achieve.
Executive engagement also makes an enormous difference. I once worked in an organisation where the CEO was personally very interested in data and AI and we had weekly discussions about AI. That creates a completely different environment.
If the board understands data, you are in a much stronger position.
Johannes: Semantic layer is another term that seems to mean almost anything at the moment. Some people mean metrics, others a business glossary, while others talk about ontologies and knowledge graphs. What is your take?
Winfried: Semantic layers have become hot again, although the idea itself is not new.
There is a real need for them, but there is also a reason this has been difficult historically: building shared semantics requires a lot of work from the organisation.
The tooling is better now. AI can help us discover and map relationships, for example. Large vendors are also investing heavily in this area.
But ultimately the purpose should be quite simple: we need to understand what things mean in our business context.
We also need to understand when that meaning changes.
That is important because meaning is contextual. “Customer” may mean something different in finance than it does in marketing. Those differences are not necessarily errors. They are relationships and contexts that need to be understood and mapped.
Johannes: To me there are at least two layers here. First someone in the business needs to decide what a KPI such as churn actually means. Only after that can the data team decide how the metric is technically calculated. Yet much of the semantic-layer discussion seems to start with the second part. Who should actually own definitions such as customer, revenue or churn?
Winfried: The ownership has to sit with the people who understand the business meaning and have responsibility for the context in which that definition is used.
A data team can implement the definition technically, model it and make sure it is consistently available. But it should not independently decide what revenue or churn means for the business.
At the same time, there may not always be one universal definition.
Customer may mean one thing in marketing, something else in finance and something slightly different in another business context. The goal is not necessarily to force all of those into one definition.
The important thing is to make those meanings explicit, understand the context and map the relationships between them.
That is where governance becomes essential. You need a process for deciding who can define something, who can change it, why it changes and where that change has an impact.
Johannes: And there is also an interesting philosophical problem behind this, right?
Winfried: Yes. The symbol grounding problem is very relevant here.
Humans connect words and concepts to experiences and meaning. A machine does not experience the world in the same way.
If I tell you that there is a small stone in my shoe, you immediately understand why that is a problem. You have a physical understanding of what a shoe, a stone and pain are.
AI can receive context explaining that the stone is hard and your foot is soft, but the connection is still artificial.
The more abstract our models become, the harder it can become to connect them back to something real.
That is one of the core challenges in semantic modelling and ontologies. You can model almost anything, but the purpose should still be to connect concepts back to the reality of the organisation.
Johannes: Data teams usually ask a very practical question at this point: where does the semantic layer live?
Winfried: Ideally, it should connect across your tooling rather than live entirely inside one application.
That is one potential limitation of vendor-specific approaches. They can work extremely well inside the vendor’s own ecosystem, but most large organisations have a much more diverse environment.
If the semantic layer maps meaning across the organisation, it needs to connect different systems and different bounded contexts.
So I would not think of it simply as one tool.
The broader problem is that we are still very application-centric. We identify a problem and buy an application for it. Then another problem appears and we buy another application.
Very few people are mapping across those applications and looking at how the pieces connect.
Johannes: Finally, you will be teaching our Data Management Fundamentals and DAMA Certification Preparation course in Helsinki. Why should someone attend if they already work with data every day?
Winfried: There are several reasons.
One is meeting other people who are dealing with the same problems in different organisations. I always encourage people to be active in the course and discuss their own experiences. You can learn a great deal from seeing how somebody else approaches the same challenge.
Certification is another reason.
I have had people in my classes with 20 years of data experience. They already know a lot, but they want a recognised certification that demonstrates the expertise they have built throughout their career.
Then there are people who are newer to data and want to get the terminology and the overall picture right.
A three-day fundamentals course covers a lot: data governance, data quality, metadata, modelling, architecture, data warehousing, integration and other areas. It gives you a broad overview and a strong foundation.
There is value for organisations as well. Certification and a shared framework make it easier to understand what capabilities you have internally and to benchmark competence.
And there is one final reason.
I will make it fun.
Winfried Etzel kouluttaa Ari Hovin Data Management Fundamentals & DAMA -sertifiointivalmennuksessa Helsingissä 9.–11.11.2026.
Kolmen päivän koulutus tarjoaa kokonaiskuvan datanhallinnan keskeisistä osa-alueista ja valmistaa osallistujia DAMA CDMP -sertifiointikokeeseen.
