Ville Satopää, Associate Professor of Technology and Operations Management at INSEAD, recently hosted a webinar on one of the most quietly damaging problems in AI adoption: data quality. He was joined by Tom Kunz, former Data Manager at Shell; Kinda El Maarry, Head of Data at GotPhoto.com; and Tom Redman, President of Data Quality Solutions.

At the INSEAD Alumni Forum 2026 in Oslo, Satopää polled senior executives on a simple question: Would you bet your entire bonus on the accuracy and completeness of the data feeding your organisation's most important decisions? The majority said they would refuse. A substantial number said they didn't know how good their data was. Only four respondents said they'd bet confidently. Their responses point to a data reliability problem that most organisations would rather not face.

What the panel made clear is that data quality problems rarely become urgent until something breaks badly enough that denial is no longer an option. As Redman put it, organisations aren't ignoring poor data quality so much as absorbing the associated cost. "A typical salesperson, let's say they work 10 hours, three hours of it is [spent] preparing the data that they should already have," he said.

The panel agreed that responsibility for resolving the issue is frequently misplaced as a problem for IT or a central data team to fix. “People hear the word data, and they think it's a technical problem," said El Maarry. "But if you look at it at a deeper level, it's a business process problem that is showing in the data." The more effective approach, which El Maarry has seen work from the ground up, is to connect the people producing data with those using it, making explicit what's at stake when something arrives late, incomplete or formatted incorrectly. At meal kit company HelloFresh, for instance, something as simple as a shared document between teams produced an immediate improvement in quality, not because processes changed overnight but because the people creating the data understood the consequences of getting it wrong. "As soon as the producers understood what was at stake, that was it," she said.

Kunz's experience at Shell carried the same logic. "The business has to own it, because they will be the ones who benefit from it,” he said. As for Redman's advice to anyone wondering where to start? “Find a problem and a person with an open mind willing to solve it," he said. At the end of the day, AI can’t rescue companies from bad data, and human-led processes must be in place to build trust and reliability in the input.