CEO Gorkem Sevinc on Data Quality, Ownership, and AI Governance

Data quality is a business problem trapped in technical tools, and Gorkem Sevinc built Qualytics to prove it

Every chief data officer has sat in the room where the CFO or a CRO says the data is wrong, go fix it. Our co-founder and CEO, Gorkem Sevinc, sat in that room for years—first as a CTO, then as a CDO—and kept answering with one fix at a time, until the fixes stopped scaling.

Gorkem sat down with our SVP of Marketing, Nicole Wojno, to kick off our executive interview series. Their conversation covered the founding insight behind Qualytics, the magic that happens when business users get real ownership of data quality, what the market shift from data quality toward data control is telling us, and why augmented still beats autonomous. 

Here’s what we learned from their conversation:

1. Data quality is a business problem trapped in technical tools

Data quality challenges arrived the same way at every company where Gorkem ran the data organization: "I kept on getting yelled at. My CFO, my CRO, people in the business coming and saying this data is wrong, go fix it." 

As the person responsible for the data ecosystem, he found himself constantly writing code to catch each specific issue. At the company he co-founded before Qualytics, that approach had grown to thousands of data quality rules that still could not get ahead of the problem. He calls this cycle expensive whack-a-mole: wait for something to break, fix it, wait for the next break.

"I saw that data quality is a business problem that is trapped in technical tools, and I wanted to change the script."

He and our co-founder and CTO, Eric Simmerman, arrived at that conclusion from different angles, each building similar solutions at different companies. The great escape from whack-a-mole started with using AI to generate and maintain data quality rules, but that was just the beginning. 

Getting proactive with AI-generated rules solved detection, but detection raised its own question: "Now that I found something, what do I do with it? Just letting somebody know is not enough. I need to be able to drive downstream workflows." And when he went looking for a product that could do all of this, he became a customer of what is now his competition. "I just saw that they're missing the boat, they're targeting the wrong persona." That gap is what Qualytics was built into.

2. Data quality works best when business users co-own it

The next level of maturity arrives when the business co-owns its own data quality. Gorkem’s clearest proof point comes from our customer MAPFRE USA, a global insurance group whose centralized data governance team used to sit in the middle of a vicious cycle. Underwriters wrote documents describing their business logic, the governance team had to understand, implement, test, and validate that logic, and the two sides went back and forth until they solved it. Since adopting Qualytics, MAPFRE has elevated access to business SMEs so underwriters can define their own rules directly.

"Having underwriters who are not technical people at all coming in and saying this is an entity resolution problem, or this is a reconciliation that we need to do. That's how data governance becomes data quality, and they have actual real business impact."

Natural language opens the door to business SMEs even wider. "Business users may not know how to code, but they know how to use an LLM," Gorkem points out, and they already know how to define business logic in plain conversation. Taking a 173-page regulatory document, mapping it to your data, and applying it can happen with much less friction when the business user closest to the logic can do it directly instead of routing through a governance team.

"Data quality is a team sport,” Gorkem says. The business impact shows up when the people who know the logic can act on it directly.

3. Your AI is only as good as the data it's trained on

Enterprises are putting far more budget behind AI and data governance than they were three years ago, and Gorkem connects the two directly: AI governance and data governance are very similar things, because "your AI is only as good as the data that it is trained on." He describes AI as a "glorified calculator" that performs exactly as well as what it's fed.

"If I train AI on pictures of dogs with three legs, AI is going to tell me dogs should have three legs."

In an enterprise, AI has to be given context, and trusting that context is critical to the governance of AI in production. The stakes become clear once people start asking AI systems questions nobody scoped in advance. A board or executive-level dashboard used to have a team behind it, manually checking every number before it went out. 

“In today's world, you may have a CFO that goes around the whole thing and starts having a conversation with your data and asking for previously unknown insights. That’s the power of AI, right? Well, if AI is coming up with previously unknown insights on data that has never been looked at before, they’re going to make some bad decisions.”

"That governance cannot happen without humans. That's why we [at Qualytics] are augmented, not autonomous."

The division of labor behind that line is specific: AI can generate and maintain around 95% of the rules a company may ever need, so humans can focus on the 5% that is high-impact business logic and can never be automated.

4. Regulated industries need rules, not just model verdicts

The market has taken different angles at solving the data quality problem, from hand-written technical rules 20 years ago to black-box model-based approaches today. A model can tell you the data looks wrong, but it can't say why, and Gorkem's objection to this approach comes down to what happens when someone has to defend the decision built on top of it.

"If I'm a regulated bank, I have to be able to defend that to the regulator. I have to say exactly what failed, when, and how. If you're not running rules and you're running a model, that's not going to work. It's not really auditable or explainable."

Regulators want to see the specific rule that was violated, at a specific time, by a specific data point, including whether the rule was originally AI-generated or person-generated and how it has been edited since. The rules doing that work are often, in Gorkem's words, "very common sense logic": a patient should not be older than 130 years old or born in the year 1700. Entity resolution follows the same pattern, where Jonathan Smith, John Smith, and Jon Smith need to be resolved to the actual person behind them.

The impact goes well beyond compliance. In verticals like financial services, insurance, supply chain, and manufacturing, Gorkem sees mismatches between systems carrying hundreds of millions of dollars in impact that aren’t caught by a model but by the reconciliation between a source and a target system.

5. Validate-at-use, because the cost of bad data compounds

Every enterprise is deploying an LLM copilot or already has, and some are running agents in production. What tells those systems which data is good? How can users trust data that may or may not have been validated before? Gorkem's answer starts with context: every system asking about a metric needs to see the same quality signals, down to "we had five anomalies on this customer acquisition cost metric in the last month, here is what they were, and here's the resolution." 

From there, the question becomes when the check happens.

"I need to be able to run data quality checks, or data controls, at the point of use. That's validate-at-use."

To explain why timing matters so much, Gorkem points to the 1:10:100 rule of data quality, a concept from 1992 (attributed to George Labovitz and Yu Sang Chang) that is still relevant today. The rule explains the compounding cost of bad data: Someone enters a bad value into a system of record. Caught at the point of entry, the fix costs the company $1. Caught after the number has spread into other systems, $10. If the bad value is left in place long enough to drive a bad decision, it costs the company $100 to fix.

The 1:10:100 rule was written decades before AI entered the picture, and Gorkem extends it to where enterprises are today: "Now I'm training an AI model that includes that bad data point. It's probably $10,000 at that scale. If I'm going into a copilot that is exposing signals trained off bad data, that's probably a million. And now I have autonomous agents interacting with my data, presenting it, making decisions… that may be millions in impact.”

”Start with trusted context, and expose the controls to every single derivative use of data that actually comes in and uses that data.”

6. Big investments in data demand investment in data quality

Interest in data quality and data governance has grown by "an order of magnitude," in Gorkem's estimate, because buyers are now rethinking their spend. An enterprise putting millions into Databricks or Snowflake, and hiring people to work with that data, is realizing the investment only pays off if those people can trust what they're working with. 

Gorkem is hearing more buyers asking: "Am I arming my team with the right tooling?" A big investment in data demands a matching investment in the quality of that data. That buying pressure is why Gorkem predicts the category won't keep its name.

"I think we're going to stop calling it data quality. It's my data management suite, which may have a little bit of MDM, a little bit of data quality, a little bit of data cataloging, a little bit of context. The best of breed is going to become best integrated."

Data quality is everybody's job

The one thread that ran through the entire conversation: data quality stops being a technical problem the moment ownership stops sitting with a single team. Business users should be able to define their own rules instead of routing them through governance, AI should carry the repeatable 95% so people can focus on the judgment calls, and validation must happen at the moment a system or human uses the data.

For Gorkem, that ownership is, ultimately, a matter of culture:

"Your data governance is not somebody's job, it's everybody's job. It's the corporate culture, the corporate mindset, that I own my data and I own the quality of my data."

This conversation kicks off our executive interview series. Subscribe to our newsletter (form in footer) to catch the next one!