AI 창업 아이템이 심사에서 떨어지는 이유, 심사위원석에서 본 것
Across the applications I saw while running the 2025 Gyeonggi Generative-AI Public-Data Startup Competition, points were lost in almost exactly the same places.
I ran the 2025 Generative-AI Public-Data Startup Competition for Gyeonggi Province and GBSA, and sat on evaluation panels for MSS, KISED and KOCCA around the same period. Applications leading with AI have multiplied — and they fail in strikingly similar places.
가장 흔한 감점, 「AI를 쓴다」와 「AI가 필요하다」의 혼동
Half the pitch is model names and feature lists. That is not what the panel wants. It is what breaks if you solve this without AI. Put generative AI on a problem that rules already solve and technical difficulty rises while the commercial score falls.
“We use GPT” is not a differentiator. Everyone does now.
What breaks most often in public-data programmes
- Never checked the update cycle, 연 1회 갱신되는 데이터로 실시간 서비스를 설계한 지원서가 매년 나옵니다
- Never actually opened the dataset, 목록에는 있는데 결측이 절반인 데이터셋이 흔합니다. 열어본 팀은 발표에서 표본을 보여줍니다
- An idea that works without the data, 공공데이터 과제인데 데이터가 장식인 경우, 과제 적합성에서 먼저 걸립니다
The moment the score goes up
It is the sentence “done by a person today this takes X minutes per item, Y items a day, with a Z% error rate.” Once the numbers are there, the improvement claimed on the next slide becomes verifiable — and verifiable claims score.
A slide that ends with “efficiency improves significantly,” whatever technology sits behind it, leaves the evaluator with no box to fill.
In short
- Write, in one sentence, why it has to be AI
- Open the dataset you plan to use and bring a sample
- Measure the current state in numbers — that is the basis for any improvement claim
Si necesita una conferencia o un taller sobre este tema
Adapto el formato a los objetivos y al público de su institución, desde una charla de 90 minutos hasta un curso intensivo de 2 días. Normalmente respondo en menos de un día.
