Last Wednesday I stared at Excel for half an hour

Last month's 17 closed deals were scattered across three spreadsheets, and I zoned out staring at the "what to do next" column. I wanted to know if I could sign 5 more next month, but just getting the data into a format AI could chew on was a real headache. So I tried the dumb approach: dumped the whole mess into ChatGPT and asked—hey, can you actually predict this?

The result wasn't mysticism, but it wasn't a magic bullet either. I used to think AI prediction was just machine fortune-telling. I got stuck three times on this step before it clicked: you've got to format the data into structured tables for AI to learn from. Here's my takeaway: it can help you figure out the "rough direction"—don't expect it to make decisions for you.

So how does this actually work? Zhang Lei's approach opened my eyes

My buddy Zhang Lei runs a one-person Xiaohongshu (a major Chinese social media platform) agency in Hangzhou. At the end of each month, he copies his data—lead volume, closed-customer tags, average order value—straight into ChatGPT and fires off this prompt: "Based on the data above, predict which type of customer is most likely to close next month, and give me three specific action recommendations."

He says it's not 100% accurate, but last month he followed the advice and pushed two extra posts targeting moms-and-babies—and conversions really did jump 30%. He admits "it might be a coincidence," but he's willing to keep testing.

The underlying logic: ChatGPT can find patterns in your historical data. Customer age distribution, inquiry timing, average order value—AI can learn from all of it. But there's a big prerequisite: your data has to be structured (Excel sheets, CSVs), not screenshots of chats.

What does it cost to replicate today?

Money: ChatGPT Plus is $20/month (about 145 RMB). The free tier works too, just slower and more prone to cutting out.

Time: First-time setup is roughly 1 hour (including exporting chat logs into a spreadsheet). After that, each prediction run takes about 10 minutes.

Tech barrier: You need to copy-paste from a spreadsheet and save one prompt template. That's it.

First step: Right now, open chat.openai.com, paste in any one sheet of last month's sales data, and type: "Please analyze this data, find the three most obvious patterns, and predict next month's trend." See what it says before deciding whether to continue.

Which stage are you at?

Just starting out (0 customers): Don't bother yet. You don't even have baseline data—forcing AI to predict is asking it to make stuff up. Run for three months first.

Have 1-2 stable customers: Worth trying once. Compile your inquiry records and closed deals into one spreadsheet, run a prediction, and see if AI's direction matches your gut. Use it as a "second opinion."

Scaling (5-20 customers): Worth taking seriously now. Spend 30 minutes a week letting AI look at your data—it doesn't sleep, doesn't charge extra, and can pull you a bit out of "going with my gut on every deal." AI prediction is just a reference—final calls are still yours.