AI Use Case Contest · Internal Best Practice Awards
Internal AI Use Case Contest and Awards
What Is an AI Use Case Contest?
- One form that collects scattered usage in one place
- Security check as the first gate; scoring weighted toward reproducibility
- Don't add up applicants' estimates as company-wide savings
What an AI use case contest changes is that the training team learns who is using AI well, and where. After company-wide training, real use happens in scattered pockets, and admin logs show only counts, not what people did. A contest collects that scattered usage through a single form, and the form becomes the draft of the casebook. Held at year-end, it gives you both a results report for this year's training and the basis for next year's training plan, which is why contests cluster in October through December.
The first things to decide in the design are the entry unit and the template. Individual entries bring in many submissions, but most are small tips. Team entries bring fewer, but they surface cases that changed a work process. Limiting the template to 'which task, with which tool and prompt, and what changed before and after' in one A4 page and one before-and-after figure protects both the entry rate and the quality of judging. Beyond quantitative impact such as time saved, the scoring should weight replicability, meaning whether another department can follow the template and do the same. As for rewards, a chance to present to the CEO, inclusion in the casebook, and a note in the HR record last longer than cash prizes.
A contest also works as a filter for risk. Submissions will include cases built by entering customer data into external AI services or using material with unclear copyright, and giving an award to one would mean the company had officially endorsed a violation. You need a structure where the security and privacy check is the first gate, so only cases that pass reach the final round. Conversely, cases caught at this gate become material for improving the guidelines. State in the announcement that applicants are told why they were rejected but face no penalty; that is what keeps honest entries coming the following year.
The most common mistake in practice is taking the figures in submissions at face value. 'Saves 10 hours a week' is usually the applicant's own estimate. If you add these up and report '30,000 hours saved company-wide per year,' next year's executive expectations will drift away from reality. Another is the same department and the same people winning every year. When marketing and planning teams sweep the awards, production and logistics staff see the event as 'nothing to do with us,' and the gap in AI use actually widens. Split the awards by job function or set up a separate category for frontline work. And without a named owner and a schedule for actually spreading the selected cases after the awards, the cases remain just a list of winners.
John KimInstructor John Kim, author of “B2B Marketing Design” and CEO of Nanum Consulting
Miru LeeExpert in future careers, self-branding, and AI tools trainingThis carries over the manufacturing tradition of suggestion systems and quality-circle competitions, where shop-floor improvements were collected, rewarded, and shared, into the use of generative AI. With no single founder or theory behind it, it became established in practice after 2023, as companies that rolled out generative AI company-wide adopted it as a way to show results.
Consider a hypothetical case. Company A, a logistics firm with 2,300 employees, rolled out generative AI company-wide in the first half of the year, but had nothing to show for what had actually changed. So it decided to hold its first AI use case contest in October, ahead of the year-end management report. The submission period was four weeks.
- Manager Jung of the training team limited the submission form to a single A4 page with four fields (task, tool, prompt, and before/after numbers) and announced separate categories for office and frontline staff.
- 87 submissions came in over four weeks. At the first gate, the information security team screened out 6 cases in which customer addresses had been uploaded to external services. Manager Jung told those applicants why and stated explicitly: "There will be no penalty."
- In the final round, entries were scored 30 points for impact figures, 40 for replicability, and 30 for scalability. One of the judges, the head of the sales division, proposed: "The time savings are just the entrant's word, so let's follow the template ourselves first and then score."
- Following that proposal, employees from other departments tried to reproduce the top 10 entries using only the template. Of those, 3 could not be reproduced, so the rankings changed, and first place in the field category went to a dispatcher who drafted replies to customer complaints.
- The 12 cases presented to the CEO at the December awards ceremony were compiled into a casebook and distributed. When replication was tracked the following February, 6 of them had spread to at least 3 other sites, and those 6 became hands-on exercises in next year's training plan.