Last week I got 47 client emails, each needing a 'reply today or not?' call. I worked until 1 AM.

Back then I only knew how to ask AI to write little essays

Before, I'd prompt ChatGPT with 'tell me if this client is worth pursuing,' and it'd come back with a polite paragraph: 'This appears to be a promising client, I suggest...' — I nearly wasted time chasing the wrong people. Then I learned AI has a new move: hand you a score, a yes/no, a label. The English-speaking world calls these Decision Models. Basically, AI stops writing essays and starts scoring.

Who's already using this? Real names

Lao Chen runs an independent consulting practice in Hangzhou. Every morning he uses OpenAI's new Decisions feature to bucket client messages into three tiers — urgent / normal / junk — done in ten minutes. He says it's '10x faster than reading them myself.' Michelle's team at Cloudflare released clef last month — it scores audio and images to detect spam or violations. Perplexity is also building pplx-decider, claiming 94.5% accuracy across 1,000+ tests. These aren't PowerPoint concepts — they're running in production today.

What's the cost to try today?

Money: OpenAI Decisions runs roughly $0.10 per 1M input tokens. My 47 emails cost pennies in RMB.
Time: Half a day to get it running, assuming you'll actually read the API docs.
Technical barrier: Not high, but you need to know what an 'API key' is (basically your AI account password). If you've never heard the term, start with ChatGPT's web 'Custom GPT' — tell it 'only reply with a score from 1 to 10.' Same vibe.
First step: open platform.openai.com → log in → Playground → pick 'Decisions' → paste one client email → see what it gives you.

One thing per stage

Just starting (under 5 clients/month): Skip it for now. Sorting 47 emails by hand takes maybe 30 minutes — getting your business running matters more.
1-2 steady clients: I'd pick one painful use case and try — say, prioritizing inbound leads or scoring contract terms. One at a time. Get one working before the next.
Scaling (10+ emails/day, multiple clients, maybe a team): I'd push hard on this. I save 1.5 hours a day now — that's one extra thing every day. The longer you wait, the more you lose.