What you will end up with
You will have a simple system where AI sorts your incoming emails, drafts replies to routine questions, and flags the ones that need you. Nothing important goes out without you seeing it first.
The aim is not to hand over your inbox. It is to stop spending your mornings answering "what are your hours?" so you can spend that time on the emails that actually need your judgement.
Before you start
You need three things in place:
- An AI email assistant set up on your inbox. Rapid Ready Ai offers an AI Inbox & Email role for this. Because it connects to your mailbox, it is quoted before you order.
- Basic information about your business given to the AI. That means what you do, your pricing, your delivery times and your policies. It works only from what you tell it, so gaps here become mistakes later.
- Access to your email inbox. You will want about 20 recent emails open as real examples.
Set aside an hour. Most of the work is deciding things you already know in your head but have never written down.
Step 1: Tell the AI what counts as urgent
Write two short lists. The first is what needs your attention quickly. The second is what is routine.
Urgent usually includes:
- Payment problems, such as a failed payment or a disputed charge
- Complaints or unhappy customers
- Time-sensitive requests, such as "I need this by Friday"
Routine usually includes:
- Questions about hours, pricing or how to order
- Booking confirmations
- General enquiries you answer the same way every time
Be specific. "Anything from an unhappy customer" is clearer than "important emails". Add a few words that signal trouble, like "refund", "damaged", "still waiting" or "disappointed".
What you should see: a short written definition of urgent and routine that you could hand to a new staff member on their first day.
Tip: use real emails
Pick two urgent and two routine emails from your inbox and note why each belongs where it does. The reasons help the AI more than the labels do.
Step 2: Set up sorting rules
Next, give the AI the categories you want and what to do with each one. Four is plenty to start:
- Customer enquiry: a question about your products or services. Draft a reply.
- Invoice: a bill from a supplier or a payment question. Do not reply. Mark it for you.
- Complaint: an unhappy customer. Do not reply. Flag it for you.
- Booking request: someone wanting a time or date. Draft a reply, but flag it if the time is not clearly available.
For each category, include one real example. If you run a plumbing business in Hamilton, a booking request might be "Can someone come look at a leaking tap on Thursday?"
What you should see: new emails landing in the right category. Check the first ten by hand and note any it got wrong.
Step 3: Draft replies for routine questions
Pick your four or five most common questions. For most small businesses these are opening hours, pricing, how to order and delivery times.
For each one, write the answer you would want a customer to get. Then paste in one or two past replies you were happy with, so the AI picks up your tone. If you write short and friendly, show it that. If you are more formal, show it that instead.
Also tell it what not to do. For example: "Never promise a delivery date. Never offer a discount. Never guess."
What you should see: draft replies that read like you wrote them on a good day. They will not be perfect. That is what Steps 5 and 6 are for.
Step 4: Flag what needs you
Now list the emails that must reach you before any reply goes out. Start with:
- Complaints
- Custom quotes or anything priced outside your standard list
- Anything the AI cannot answer from the information you gave it
- Anything involving a refund, a change to an order or a special arrangement
The last point matters most. Tell the AI that if it is unsure, it should flag the email rather than guess.
What you should see: a short list of flagged emails each day, each with a note on why it was flagged. If everything is flagged, or nothing is, see the section below.
Step 5: Run the pre-send checklist
Before any AI-drafted reply leaves your inbox, check it against this list:
- Is the price correct? Compare it with your current price list.
- Is the scope right? Does the reply promise only what you actually offer?
- Are the dates accurate? Check delivery times, opening days and any public holidays.
- Does the tone match your business? Would a regular customer think it came from you?
- Is there anything the AI got wrong? Names, product details and anything it seems to have made up.
AI can sound confident and still be wrong. Prices and dates are where mistakes cost you, so never skip those two.
What you should see: most drafts pass in under a minute. The ones that do not tell you what to fix in your instructions.
Step 6: Approve and send
If the draft passes, approve it. If it is close, edit it and send it yourself. If it is off, rewrite it and make a note of what the AI misunderstood.
Keep that note. After a week you will see patterns, such as the AI always getting one product name wrong, and you can fix each one at the source.
What you should see: your inbox sorted, routine replies waiting for a quick check, and flagged emails clearly separated.
If it does not work
The AI flags too much or too little
Your definition of urgent is probably too vague. Add more real examples to both lists. If it flags every question about price, say which price questions it can answer and which need you. If it misses complaints, add the exact phrases your customers use when they are unhappy.
The replies sound wrong
Give it better samples of your tone. Choose replies you are proud of, not just the first ones you find. Add plain rules like "keep it under five sentences" or "say kia ora, not dear customer". Then test it on the same emails again.
The AI is missing context
This usually means the information you gave it has gaps. Check that it knows your products, your standard answers and your policies on returns, delivery and bookings. If a reply is wrong, ask yourself whether you ever told it the right answer. If not, add it.
Next steps
Once this runs smoothly for a couple of weeks, you can widen it. Good candidates are:
- Invoices: sorting supplier bills and pulling them into one place for you to review
- Booking confirmations: sending a standard confirmation once you have approved the time
- Follow-up emails: a gentle nudge to someone who asked for a quote and has gone quiet
Add one at a time, and keep the same checklist in place. A person should still see anything involving money or a promise.
If you have questions about any of this, you are welcome to get in touch through our contact page.
