Replying to flower order enquiries with AI

Most enquiries are the same few questions in different clothes. Can you do this on that day? What does it cost? Do you deliver there? An AI draft helps most when you've already answered those questions yourself and only need the words.

Do the checking first

The tool can't see your cooler, your diary or your driver's route. Before you ask for a draft, settle four things on your own:

• Do you have the flowers, or can your wholesaler supply them in time?
• Is there room in the work schedule for that day?
• Can you reach the delivery address, or will it be pickup?
• What's the price, including delivery?

Then give the tool those answers. A message to it might read: "Reply to this customer. Say yes to a hand-tied bouquet in pinks and whites for Friday afternoon, pickup only, at the price I give. Keep it friendly and short." Those facts are yours. The tool's job is the tone.

One maker's guide says to tell the model to stick to the material you've supplied and leave out what it already "knows"¹. In plain terms, say: "Only use the facts I've given you. If something's missing, ask me." The same guide says to let the model admit it's unsure¹. That helps stop it inventing a delivery time.

A reusable reply shape

Most good replies have the same parts, in this order.

1. The answer, yes or no, in the first line.
2. What you can do, with the date and time stated in full.
3. The price and what it includes.
4. What you need from the customer, such as the delivery address, a phone number and the card message.
5. A plain sign-off with your shop name.

Save that shape as a note you paste at the top of each request. Your drafts will come out in a steady order and you'll spot gaps quickly.

Keeping it from sounding like a robot

People notice stock phrases. A few habits keep a draft sounding like your shop.

• Paste in two or three of your own past replies and say "match this style." Remove customer names and addresses first.
• Ask for short sentences and no exclamation marks, unless that's how you write.
• Cut openers like "Thank you for reaching out." Start with the answer.
• Read it aloud. If you'd never say a line to a customer at the counter, change it.

A tool can also sound very sure while being wrong. One maker's help page notes that a model "might sound confident" even when its answer is incorrect². Sounding sure proves nothing about the details.

What to keep out of the tool

A customer's name, phone number and address are personal information. The Australian privacy regulator advises against pasting such details into public AI chat tools³. That's Australian guidance. If you're in another country, look up your own regulator.

Try this: write "the customer" in the request, then add the real name to your own reply afterwards.

Cases to answer yourself

• Anything involving an allergy or a health concern. Answer from what your supplier has told you, or say you don't know.
• Large or mixed orders, such as events. Reply by phone or with your own wording.
• Complaints. A person should write those.
• Orders where the date is close and your stock is tight. Say what you can do, then confirm.

Read before sending

OpenAI's guidance calls the output a draft to be checked, not something to rely on as it stands². Anthropic's guide also says to validate critical details¹. For a flower order, those are the date, the time, the address and the price. A short routine for this sits in the pre-send checklist.