6 min read
It is 9:40 on a Tuesday night in early December, and Rosa is at her kitchen table with a cold cup of tea, typing the same answer for the eleventh time that day. “Yes, the cedar and fig candle ships in a gift box. No, I can’t guarantee it arrives by the 18th unless you choose express.” Behind her, 60 candles are curing on racks, and she still has labels to print.
Rosa is a composite, not a real person. This is an illustrative story built from common patterns in small product businesses, and the numbers are realistic examples, not measured results. The workflow, however, is one you can copy.
Rosa makes hand-poured soy candles and sells them through her own online shop and at weekend markets. She works alone. Most of her year is pleasantly steady, but November and December bring in about 40 percent of her annual sales, and with them a flood of messages. Last year she spent around 11 hours a week on customer email during the peak, and she remembers little of that season except being tired.
This year she decided to change one thing before the rush. She did not want a chatbot on her site or a new subscription she would have to learn. She wanted her evenings back.
The discovery: 12 questions were eating her December
Rosa began by exporting her email from the previous November and December. She pasted about 200 customer messages, with names and addresses removed, into her AI assistant and asked one question: “Group these by what the customer is really asking, and count each group.”
The answer stung a little. Roughly 70 percent of her messages were versions of just 12 questions. Will it arrive in time? Can you ship to my mother’s address? Is this safe around pets? Can you add a gift note? How long does a candle burn? Do you offer wholesale?
She had been treating each message as a fresh conversation. In truth, she had been answering the same FAQ by hand for years.
The setup: one afternoon, one AI project
Rosa did not need new software. She already paid for an AI assistant, so she created a single workspace inside it, using Claude Projects, though a custom GPT would have worked as well. Our comparison of the three main options can help you choose.
Into that workspace, she placed three documents:
- The answers file. One approved answer for each of the 12 questions, written in her own words.
- The facts file. Candle sizes, burn times, ingredients, shipping cutoffs, her returns policy and her market schedule.
- A voice sample. Twenty of her warmest real replies, so that drafts would sound like her and not like a corporate help desk.
Then she wrote short instructions: “Draft a reply to the customer message I paste. Use only the facts in these files. If a question isn’t covered, say so and leave a note for me. Never promise a delivery date. Keep it under 90 words.”
That took about three hours on a slow Sunday afternoon in October.
The rule that mattered most: never promise a date
The instruction about delivery dates was not there for style. Rosa knew her shipping cutoffs, but she also knew that carriers slip in December. An AI that cheerfully guessed “it will be there by the 20th!” could create an angry customer, or a real problem. In the US, the FTC’s mail and internet order rule expects sellers to have a reasonable basis for any shipping time they state.
So the draft always quotes her published cutoffs, and only she sends the message. If you have read our guide to catching made-up facts, you will recognize the reasoning. Give AI an approved source, then check its work before a customer sees it.
The routine: three batches a day
The bigger change turned out to be a habit, not a tool. Rosa stopped answering email as it arrived. Instead, she set three short sessions: 9:00, 1:00 and 7:00.
- She opens her inbox and pastes each message into the project.
- She reads the draft, fixes anything that feels off, and adds a personal line where one belongs.
- She sends it, then moves to the next one.
Most replies took under a minute. Anything unusual, such as a wedding order of 150 candles, went into a separate “me” folder for proper attention. These were the conversations she actually enjoyed, and now she had time for them.
She also fixed the source of many questions. She took the top five from her answers file and placed them on her product pages, directly above the buy button, along with a clear shipping-cutoff banner. Fewer people needed to ask.
The result, and what went wrong along the way
In this illustrative example, Rosa’s holiday email time falls from about 11 hours a week to about 4. Response time improves too, from “sometime tomorrow” to the same day, and a few customers mention how quick she has become.
It was not perfect. The first week brought three real mistakes:
- A draft suggested a scent she had discontinued, because her facts file was out of date. She now updates that file every Sunday.
- A customer asked about a candle near a cat, and the draft sounded too certain. She rewrote that answer to be more cautious, and it now tells people to follow the label and ask their vet.
- Her first drafts all opened the same way, which started to read like a template. She added a line telling the AI to vary its openings, and she keeps writing her own closing line.
She also made one decision that saved her from trouble: she kept customer addresses and order numbers out of the workspace and looked those up herself in her shop system. It added a few seconds per reply, and she felt better about what she was sharing with a third party.
What you can borrow this week
You do not need to make candles for this to work. It suits anyone whose busy season is really a repeated-questions season: photographers, tutors, bakers, gift sellers and repair shops. With the holidays close, try this:
- Pull last season’s messages and remove personal details.
- Ask AI to group them and find your top 10 to 12 questions.
- Write one approved answer each, in your own voice.
- Build one project with your answers, facts and tone samples.
- Add your hard rules, such as never promising dates, refunds or prices that are not in the file.
- Batch your replies into two or three set times a day.
- Move the top five answers onto your product pages.
Our nine holiday quick wins have more ideas if you are short on time. What question do you answer more than any other in your busy season? Tell us in the comments, and we may build a template around it.
