Monday morning, 7 client messages, and I almost started over

Last Monday morning, I was revising a quote in a Shenzhen café when a client asked about delivery time, payment methods, and our return policy. I copied an old email and scrambled to answer each question one by one. I spent 40 minutes replying and still missed one message afterward. That was when I started experimenting with AI to organize my reply drafts. It wasn’t sending anything for me.

I treat it as a second draft, not an automated customer service bot

The setup is simple: I put the client’s questions, our service boundaries, and past conversations into one document. Then I ask AI to draft replies in this order: confirm the question, provide the information, suggest a next step. I check the pricing, commitments, and tone myself before copying the draft into the chat window.

I know Chen Yan, who runs customized training. She organizes students’ questions every evening in Hangzhou. She has AI group repeated questions first, then prepares one reply template for each category. In 20 minutes at night, she completes work that used to take nearly two hours. She also made a mistake: in her first version, she added “refunds are returned via the original payment method” to every course. She had to remove it manually. It showed why the boundaries of the source material have to be explicit.

What it costs to catch up today: ¥30, 40 minutes, no coding

I started with the chat tool I already used, without rushing to buy a complex system. I spent about ¥30, mainly on one month of an AI subscription. Preparing the material took 20 minutes, and building the first template took another 20. The technical bar is low: knowing how to copy and paste, and recognizing what must never be sent automatically. The first step is to open the AI chat tool I already use, create a document called “Frequently Asked Client Questions,” and write three real questions plus one rule about what must never be promised.

When I was just starting out, I handled only the three questions clients asked most often. With one or two clients, I organized the tone and commitments separately for each client. While scaling, I turned recurring questions into team templates and kept a human review step. Not everyone needs this method. When there is little source material or communication is highly personalized, writing manually may be faster. There’s no rush to try it either: save one real question and test it when I have time.

Match the workflow to my stage, don’t automate everything at once

Just starting out: I use one real question to run three drafts and get familiar with how AI rewrites it.

Working with one or two clients: I build a reply library with clear boundaries and review missed edits once a week.

Scaling up: I let teammates maintain the templates together, but a person still confirms pricing, contracts, and delivery dates.