Your Best AI Use Case Is Not in the Boardroom. It Is at the Desk Nobody Visits.
Last week KB Financial Group, one of South Korea’s largest banking groups, ran something it called the 2026 AI Agent Competition. Three hundred and sixteen employees, across a hundred and sixteen teams from seven affiliates including KB Kookmin Bank and KB Insurance, built AI agents meant to solve real problems in their own jobs. Twelve teams reached the final. The grand prize went to a customer-care AI control center, built by people who answer customer calls for a living, not by the group’s AI strategy team.
That detail is the whole story.
Executives think adoption is further along than it is
A useful, uncomfortable data point came out earlier this year from a workplace AI proficiency study covering thousands of employees. Executives at the surveyed companies overwhelmingly reported a clear AI strategy, wide adoption, and a culture that encourages employees to build their own solutions. Employees told a different story. A quarter said they had no work-related AI use case at all, and of the use cases people did report, only a small fraction looked likely to produce real return. The researchers called it a use case desert: people know how to open a chatbot, they just cannot picture what to point it at.
This is not a training problem you can fix with a slide deck, and it is not really a technology gap either. It is a distance problem. The people who could name the highest-value use case in your company, the one where AI removes a genuine daily headache, are usually the furthest from whoever is deciding the AI roadmap.
Why the floor beats the strategy deck
An executive sees the business as a chart of departments and outcomes. The person answering support tickets sees the same three questions arrive forty times a day in slightly different words. The person keying in supplier invoices sees exactly which fields the ERP makes them re-type from a PDF that already has the number. That specificity is the raw material of a good embedded-AI use case, and it does not survive being summarized upward through two layers of management before it reaches whoever holds the AI budget.
KB Financial’s answer was to run a competition and let hundreds of people compete to name and build the fix themselves. A small or mid-sized company obviously cannot run a 316-person hackathon, and does not need to. The mechanism that matters is not the scale, it is the direction the question travels.
A version of this that fits a ten-person team
You do not need a competition. You need thirty minutes and an honest question, asked to the people doing the work rather than about them. Ask your support person, your bookkeeper, your ops lead: what is the one thing you retype, re-check, or re-explain every single day that a competent assistant could do for you? Write down every answer without judging it yet.
Then apply the filters we have written about before for picking the workflow that is actually worth embedding AI into: does it repeat, is a wrong answer cheap to catch, and can you name the number it should improve. Most of what your team names will fail that test. One or two will not, and those are worth more than anything a consultant would have guessed from outside.
The training gap we wrote about last month is real, but it assumes people already know what to point AI at and just need the skill. KB Financial’s competition suggests the opposite problem is often bigger: the skill exists, the target does not, because nobody with the budget asked the person with the desk.
If you want help running that thirty-minute conversation properly and turning the answer into a workflow worth building, that is exactly what our free assessment is for.