Trusted data at the core of successful GenAI adoption

[Pelican Press 0]

This is the hidden content, please

• Sign In
• or
• Sign Up

Trusted data at the core of successful GenAI adoption

As enterprises increasingly invest in developing AI applications, including generative AI, the success of those applications depends on trusted data.

The data itself must be trustworthy, as in high-quality data that is accurate, complete and consistent. And it’s imperative that users have trust in the data that trains and informs AI tools, such as chatbots that enable users to analyze data using natural language, and agents that take on repetitive tasks that otherwise have to be performed by humans.

Without trusted data, applications will deliver inaccurate outputs. Without users trusting data, applications will wind up unused. The result is missed opportunities — missed chances to maintain and improve relationships with customers, missed chances to identify cross-selling and other revenue growth opportunities, and more.

With trusted data, AI outputs are more accurate, leading to more widespread adoption. The result is that employees enterprise-wide are enabled to make smarter, more informed decisions that can lead to growth.

Ataccama, a data management vendor based in Toronto and specializing in data quality, recently partnered with Hanover Research to survey more than 300 senior data leaders for its 2025 data trust report to gauge their success in developing AI applications and discover what barriers prevent success.

Only a third reported having any meaningful success developing and deploying AI applications. As for identifying the primary problem, more than two-thirds said trusted data — or lack thereof — was the culprit.

This is the hidden content, please

• Sign In
• or
• Sign Up

Mike McKee

Data quality and trust were always significant, but with AI enabling more employees within enterprises to make decisions based on analytics and automating tasks that previously were performed by humans, their importance is rising, according to Mike McKee, CEO of Ataccama. As

This is the hidden content, please

• Sign In
• or
• Sign Up

invest in AI development, and those already doing so expand their reach, the importance of data quality and trust will only increase.

McKee recently discussed the critical importance of data trust.

AI is already evolving into the primary interface for data analysis and decision-making, he noted, and it will only become more so as more organizations develop successful AI applications. As a result, data that can be trusted to accurately train applications and lead to trust from those who use the applications to inform decisions and actions has never been more important.

Editor’s note: This Q&A has been edited for clarity and conciseness.

How do you define trusted data?

Mike McKee: At a basic level, it’s no more than being able to trust your data.

That sounds super simple, but one of our fundamental beliefs is that the world has changed from CIO-driven projects, which have been going on for 20 years, to CEO-driven data products. The turning point was ChatGPT 3.5 coming out, and board members and executives suddenly asking whether their organization is using generative AI. They went to … the CIO, and the CIO said, ‘I’ve been working [with data] for 15 years, and you never paid attention. Now, all of a sudden, you care.’

If you think about the data supply chain from the different sources of data to the *** tools, and the use of that data, so many data projects start with governance and catalogs. It’s about transforming the data, making sure it’s cataloged — and ultimately, what matters is whether the business can trust the data.

Similarly, how do you define data quality?

McKee: Once again, the simple answer is high-quality data.

But the trick is quantifying data quality. You can look at the completeness, the uniqueness, the validity — whatever matters most to a particular data set — but the whole idea is to quantify it. We had our product and engineering quarterly business review this morning, and I said that I knew our quality was getting better, but asked, ‘What data shows me that?’ When it comes to our customers, we try to provide a way to quantify data quality along the lines that they believe are most important.

Why is it difficult for organizations to maintain the data quality that leads to trusted data?

McKee: One reason is the

This is the hidden content, please

• Sign In
• or
• Sign Up

of data. Now, there is the digitization of everything. Whether it’s a customer service call, a marketing campaign or website statistics, there are more data sources. The proliferation of data is one thing.

The other thing is that challenge of quantifying data quality, which can be unclear because some data is OK if it’s 80% accurate and some data has to be 99% accurate. Understanding what those thresholds of data accuracy are is important. If it’s the difference between calling someone Mike or Michael, we’re good. If it’s a medical prescription, that can’t be right just 80% of the time. What’s important — and it is a challenge — is knowing what’s an acceptable level of quality and how to quantify it in a world of exploding data sources.

What are the consequences of poor data quality?

McKee: A lot of it starts with customer 360. My family just got a new TV and had to get a new set-top box. I called Verizon, waited a long time, and then they needed to send a four-digit PIN to authorize information on my account. The four-digit PIN was sent to my 23-year-old daughter in Vermont who was on a hike. That [poor data quality] led to customer aggravation. Customer intimacy is missed, the ability to improve customer data and the customer experience, and the ability to cross-sell to customers. There are a lot of everyday examples of bad data and the bad experiences, missed cross-selling opportunities and a lot of other bad things they lead to.

Getting into the relationship between data trust and AI development, are enterprises having success as they increase their investments in building and deploying AI applications?

McKee: In the report, we found that 33% of the 300 companies we spoke to are making meaningful progress with AI initiatives. I was surprised it was that high.

It’s great that this AI catalyst has uncovered the need to have better-quality data. The mistrust of data has been there for a while in a lot of organizations, and AI has become a trigger that shows there’s a problem that needs to be solved. It’s still super early days of AI development, so [three things that need to be done are] making sure the quality of the data going in is high, figuring out where to run the models and getting into the ethics issues related to who can see the outputs from the models.

What is an example of an ethics issue related to who can view AI outputs?

AI is this incredibly powerful engine, and my strong belief is that if organizations don’t leverage AI, they won’t be competitive.
Mike McKeeCEO, Ataccama

McKee: Natural language inquiries are the direction we’re going with AI applications, and anyone could ask about someone else’s salary. You have to put limits on who can see something like that.

AI is this incredibly powerful engine, and my strong belief is that if organizations don’t leverage AI, they won’t be competitive. Organizations that were born in the data world — Meta,

This is the hidden content, please

• Sign In
• or
• Sign Up

, Uber — are dominating. The other 98% of industries out there, from selling baseball hats to hotels, cars, trains and insurance, have been worrying about their own industry, and the companies haven’t thought that they have to be phenomenal data organizations. Now, to be competitive, companies have to leverage their data and have to leverage AI.

What has been the biggest barrier to successful AI deployment so far?

McKee: Historically, all the data sources were cataloged, governed and under control for data projects. Going back to that shift from CIO-driven data projects to CEO-driven data products, it’s important to bite off digestible chunks — something like running a better marketing campaign, creating a better website experience, making sure pricing across different websites is the same.

[Success] is around digestible chunks tied to a business initiative. When data projects are tied to the business, there’s a connection between making a fast decision and data they can trust. Trying to get a manageable amount of data to the quality level that you need to make better decisions faster — finding that balance between the amount of data and the business initiative — is the key to making progress.

What separates those enterprises having early success with their AI initiatives from those that aren’t having success?

McKee: Business involvement, 100%.

At Ataccama, we look at AI and generative AI twofold. One is the need to build AI and generative AI into our products because there’s a fundamental shift happening. Data is exploding faster than people. With that exploding amount of data and a fixed number of people, there has to be automation in the [data management] process. We think AI and generative AI can allow organizations to get more data in good shape, which is super important. The second is allowing business users to be involved in that data management process. The data team doesn’t understand the business’s pain, so from a use case perspective, tying applications to the business user is absolutely essential.

It’s hard. No one is going to look good if there’s a big data project and it doesn’t do anything for the business. To make a project successful, push at the outset to understand what difference the project is going to make for the business. It’s tricky because that’s not how data teams have been wired in the past. But the projects that are successful are the ones with a business initiative tied to it — where the business is involved in getting better-quality data. That’s a truism. There has to be a business rationale behind improving the data.

What benefits are enterprises that have the trusted data needed to successfully develop and implement AI applications seeing that others are not?

McKee: It’s better, or faster, decisions. I’m a huge believer that *** tools — Tableau,

This is the hidden content, please

• Sign In
• or
• Sign Up

Power *** and others — will move more and more toward natural language inquiries. People will ask a question, and *****, the answer will be right there. They won’t have to look for a table. The enterprises that are already using [generative AI tools] are getting better answers and able to access information more quickly.

Do you think it’s possible that within the next one to two years, most organizations will be able to trust their data enough to derive success from their AI initiatives?

McKee: I’d probably go two to three rather than one to two, but they have no choice. There will be a day when, if a company sends a PIN to your daughter in Vermont instead of you, you’ll say, ‘Forget it,’ about that company. If you look at the companies that adopt technology in traditional industries and leverage their data to use AI, they’re completely separating themselves. Uber, for example, is worth more than all the airlines put together. You need to make better decisions faster, and the ones that don’t will lose to the ones that do.

It is possible to quantify data quality, and it is possible to have thresholds for different sets of data that matter. I’m optimistic that having high-quality data and leveraging AI will be there in two to three years. I spent 10 years in cybersecurity, and the Venn diagram between the chief information officer, chief data officer and chief security officer is coming together. There are privacy concerns, security concerns — and managing your way through those will be the [next] challenges.

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.

This is the hidden content, please

• Sign In
• or
• Sign Up

#Trusted #data #core #successful #GenAI #adoption

This is the hidden content, please

• Sign In
• or
• Sign Up

This is the hidden content, please

• Sign In
• or
• Sign Up