Agentic AI, data as a product among growing analytics trends

1. The emergence of data marketplaces

Many organizations deploy data catalogs to organize and govern their data.

With the tidal wave of AI development, the topic of governance is more mainstream than ever. Ten years ago, the only ones talking about data governance were experts. … Today, because of AI, data governance seems to be on a lot more people’s lips and in a lot more media.

Data catalogs are software applications that use metadata to inventory an organization’s data, making it discoverable so that engineers and developers can find the data needed to create reports, dashboards and other tools that can be used to inform decisions. In addition, data catalogs let organizations put in place data governance measures, which are policies and controls that simultaneously protect the organization from violating regulations while enabling users to work with data confidently.

Data catalogs have been used most frequently to connect to data stored in data lakes, warehouses and other repositories so that it can be discovered and used.

But according to Christiaens, there’s another way catalogs can be used. Instead of facing down in the data stack with a focus on metadata, they can face up toward data consumers with a focus on data products such as reports, dashboards and now generative AI applications, making them easy to find so that they can serve the needs of the business.

“It’s a way to look at a data catalog as a marketplace,” Christiaens said. “The contents are data products instead of metadata, and the audience is analysts or data citizens. The [catalog] might be the same thing, but the way it’s looked at is different.”

In essence, looking at a data catalog as a marketplace for data is a shift from focusing on data collection to data consumption by analysts and other users, according to Dmytro Lugovyi, Collibra’s senior director of value consulting, who also spoke during the vendor’s webinar.

“[Data marketplaces] will redefine how businesses manage their data,” he said.

Beyond helping analysts and other business users find and operationalize data assets more easily, using catalogs as a marketplace rather than a metadata management tool can better enable organizations to monetize data assets.

Data has value and can be sold. But that’s just one way to monetize data, according to Christiaens. In addition, it can be used to improve core products and services sold to third parties and develop insights that can be marketed to others.

“There is a massive opportunity for many companies,” Lugovyi said.

2. Enterprises embrace data as a product

In conjunction with the emergence of data marketplaces, another rising trend in data management and analytics is treating data as a product rather than merely a collection of information.

While marketplaces are a means of indexing data products and selling them, the products themselves are tools that can inform business decisions that lead to growth or be sold to third parties to create new revenue streams.

Treating data as a product was already becoming a significant trend in recent years due to the rise of data mesh as a data management architecture, increasing attempts by organizations to monetize data and analytics products, and the evolution of the chief data officer role.

“Various things are happening in the market that are making this come alive,” Christiaens said.

But another thing is happening that will accelerate the treatment of data as a product — as an asset rather than just a source of information. It is the rise of generative AI development over the past two years.

The investments enterprises are making in developing generative AI applications have surged since OpenAI launched ChatGPT in November 2022, given generative AI’s potential to make workers smarter and more efficient. Generative AI, however, completely relies on an enterprise’s proprietary data to be of value to that enterprise.

It needs to be trained with proprietary data to understand the business, respond to questions and take on certain repetitive tasks. As a result, generative AI applications, like reports and dashboards, are data products.

“AI, to me, now is generative AI,” Christiaens said. “That means there is a tidal wave of AI projects happening, and AI … is a product of training on your own data.”

With generative AI applications in effect data products, they will need to be governed just as data and data products are governed, he continued.

The data used to train models and applications has to be governed to ensure it is accurate and complete, just as it does when it informs a report or dashboard. And the resulting product needs to be governed so that it is properly used, which means establishing ownership so that someone is responsible for it and implementing processes for its management and use.

“When it comes to ownership, process or even quality, AI is only going to be as qualitative as the data you feed into it,” Christiaens said. “If you feed it low-quality data, don’t expect it to get smarter and clean the data. It doesn’t know how to do that.”

3. Automated decision-making increases

Beyond responding to user questions, generative AI can be trained to take on repetitive tasks that previously had to be performed by human beings.

For example, generative AI applications can be trained to create data pipelines that move data from databases, data warehouses, data lakes and data lakehouses to applications that inform business decisions. Similarly, they can take on data observability, which is the process of monitoring data as it moves throughout pipelines to make sure it remains high-quality.

Both are time-consuming tasks that when automated make developers and other data experts more efficient. As a result, more AI-powered automation of data management and analytics tasks will be a trend throughout 2025, according to Christiaens.

There is, however, a problem organizations will need to overcome to benefit from increased automation through generative AI, according to Lugovyi: Many people fear turning processes over to machines.

“Automation makes [people] scared,” he said. “If someone does something, you can talk to the person to ask why they’re doing something and discuss, but giving autonomy to AI can be problematic.”

That fear is one reason more enterprises haven’t adopted AI, Christiaens added.

“Fear and uncertainty have been why organizations reject automated solutions,” he said. “If someone doesn’t know how something was built and how it works, it’s a black box. Uncertainty is why old AI projects as well as new ones have been such a challenge to get to production.”

The fear can — and will — be overcome, Christiaens predicted.

Change management will play a role, with transparency a key part of that.

If an executive makes 200 decisions per day, some will be good and others bad. But there is time to examine those decisions before taking any action. Automated systems can make thousands of decisions per second and take immediate action.

Governance that not only results in high-quality data, but also shows the lineage of the data so that humans can know about the data used to inform automated processes is key. It’s what provides the transparency that leads to trust. In addition, accountability is important, with someone assigned responsibility for overseeing the performance of AI models and applications to make sure they are accurate and up to date.

“Ultimately, a lot of the data governance practices we learned in the past are immediately applicable and relevant for the AI challenges that are in front of us,” Christiaens said.

Still, he acknowledged that balancing the substantial benefits of AI-powered automation with the risk of outsourcing processes is a delicate task.

4. Agentic AI expands its reach

Just as the use of AI to automate processes will become a more mainstream data management and analytics trend throughout 2025, AI agents will become more ubiquitous this year, according to Christiaens.

Agents are, in a sense, the evolution of chatbots. Chatbots enabled users to ask questions and receive feedback using natural language. Some could also make suggestions. In effect, they were assistants, helping individual workers make more informed decisions.

Agents have the same capabilities as assistants, but go a step further by autonomously performing certain tasks and making certain decisions that save time and add efficiency.

For example, AI agents can be trained to do much of the documentation required of developers as they test applications. In addition, they can be trained to create some of the code. Beyond engineering, they can author report summaries, create presentations and do more to save people time.

As a result, research and advisory firm Gartner predicts that by 2028, a third of all enterprise software applications will include agentic AI capabilities, up from less than 1% in 2024, and about 15% of all work decisions will be made by agents rather than humans.

“Agentic AI, whether it’s Gartner’s prediction or other data, is very real,” Christiaens said.

In 2023, organizations focused on AI infrastructure, including large language models, he continued. Last year, that began to shift toward the applications themselves, including agents.

“It’s about agents that can take actions on your behalf,” Christiaens said. “Those agents are going to be very mainstream.”

But just as automating processes has to overcome a lack of trust, so do agents before their use can become widespread. And just as good governance that includes transparency is key to engendering trust in automation, it is also key to making AI agents trustworthy, according to Christiaens.

“Agents are the future, and when it comes to making sure that they work well, this is where the topic of governance becomes very important,” he said.

Trust, however, is a barrier only because these are the early days of a new technology. Once agents have proved their worth and proved to be trustworthy, they will be like other technologies that once were new and drew skepticism, but eventually became widely accepted.

A century ago, cars drew skepticism, Christiaens noted. Now, trust is the default when it comes to automobiles, and distrust only arises when a problem arises.

“Ten years from now, trust [in AI] is going to be business as usual,” Christiaens said. “Young people now will see it as normal.”

5. Organizations emphasize governance

Underpinning the success of data marketplaces, treating data as a valuable product, and relying on AI to carry out tasks and make autonomous decisions is governance.

High-quality data is at the core of data and AI tools, and governance is how organizations can best put in place the policies and procedures that lead to not only high-quality data, but also the proper use of the resulting applications.

Without governance, the data and AI products in a marketplace can’t be trusted. Neither can the processes and decisions carried out by AI applications.

“Everything starts with the data — it’s absolutely vital,” Lugovyi said. “Those who forget about data quality and transparency will lose their competitiveness.”

As a result, governance, while already important, will become even more so in the coming months and become a major trend in data management and analytics in 2025, according to Christiaens.

Data and AI governance are key to unlocking the value of generative AI for individuals using applications such as assistants and agents. They are also key to scaling generative AI beyond employee use to take on organizational processes and to remaining compliant with regulations placed on the use of data and AI.

“All the responsibility steps we’re used to when building other systems, whether data lakes and warehouses or dashboards, has to be applied to AI,” Christiaens said. “It all boils down to good governance. … You can’t have AI without proper data, so you have to care for your data assets.”

Eric Avidon is a senior news writer for Informa TechTarget and a journalist with more than 25 years of experience. He covers analytics and data management.