Guide · Email automation
What email automation actually does, where plain rules are enough, how drafts in your own voice work, what it costs in New Zealand, and which parts a person should keep.
"A staff member missed an email and we missed the job." That is how one owner put it to me, and some version of it comes up in a lot of first conversations.
It is rarely anyone's fault. A busy shared inbox can take a couple of hundred messages a day. Most are noise: receipts, newsletters, reply-alls, a supplier's automated statement. The one that matters looks exactly like the others until someone opens it, and by then it has been pushed off the first screen. Hiring another person to watch the inbox does not fix it. It moves the same problem to a new desk.
It does five jobs, in order, on every message as it arrives. None of them is sending an email on its own.
| Job | What happens | Example |
|---|---|---|
| Sort | Works out what kind of message it is and how urgent. | New enquiry, complaint, supplier invoice, job update, or noise. |
| Extract | Pulls out the details someone would otherwise retype. | The address, the order number, the date they want, the amount. |
| Route | Puts it with the right person and files it against the right record. | Tagged to the job in your job-management system, copied to the project lead. |
| Draft | Writes a reply for a person to check, using your own information. | A quote follow-up that already has the customer's last order and your current lead time in it. |
| Chase | Notices what has not been answered and says so. | "Three enquiries from Tuesday still have no reply." |
Chasing is the one people forget to ask for, and it is the one that fixes the missed-job problem. Sorting makes the inbox tidier. Chasing makes sure nothing sits there unanswered.
A lot of email handling needs no AI at all. If you can write the rule on a whiteboard, use a rule. Rules are free, instant and never guess.
I build it in that order: rules for the plumbing, AI for the judgement steps. It keeps the running cost down and it means fewer things can go wrong. There is more on the difference in automation vs AI workflow vs AI agent.
The most useful output is the plainest: a short list for each person at the start of the day. It says what needs a reply from you, what you are waiting on from someone else, and what has gone quiet for three days.
Where the list shows up matters more than how clever it is. One owner told me that if staff have to go to a separate page and open a tool, they won't use it. He was right. So the list should arrive where people already work, as a Teams message or an email first thing, with the drafts sitting in the Drafts folder of the mailbox they already use. Nobody logs into anything new.
AI can draft replies that sound like you, but only if it has your real emails and your real facts to work from. The common complaint about AI drafts is "I spend more time editing the draft than anything else." That happens when the draft is written from nothing. Give it a few dozen of your own sent replies, your price list, your policies and the customer's history, and the editing drops to a glance and a tweak.
A person should still press send. In 2024 a Canadian tribunal ordered Air Canada to honour a bereavement refund its website chatbot had wrongly promised. The airline argued it was not responsible for what the bot said, and lost. The lesson for an owner is simple: whatever goes out under your name, you own. A thirty-second read before sending is cheap insurance.
The exception is the low-risk acknowledgement. "Thanks, we have your enquiry and will reply by tomorrow" can go automatically, because nothing in it can be wrong.
Real inboxes are messier than any demo, so the design has to start with the awkward cases. These are the ones that turn up every time:
The rule I build to is that anything the system is unsure about goes to a person with a note saying why, and it never guesses. Before go-live it runs in test mode against your real mail for a couple of weeks: it does everything except act, and you read what it would have done. Every decision is logged, so when something is misfiled you can see why and fix the instruction.
The system reads the mailboxes you nominate and nothing else. For most businesses that means the shared ones, such as info@, accounts@ or orders@, and not everyone's personal mailbox.
A custom email automation build typically costs NZ$1,000 to $8,000 fixed price and takes two to five weeks. That is the document and email intelligence range on my pricing page. One shared mailbox with sorting, a morning list and drafts sits at the lower end. Several mailboxes that also file into Xero and a job system sit at the top.
Once it is live there are two running costs: hosting, monitoring and support on a monthly plan from $99, and the AI usage itself, which for a small business is usually a few dollars to a few tens of dollars a month. I have broken those down in what AI costs to run each month. Prices exclude GST.
Partly, and you should try the free parts first.
It can, but for most businesses it should not, at least at first. The safer set-up is that AI drafts the reply and a person reads it and presses send. Fully automatic replies are reasonable for low-risk messages such as an acknowledgement that an enquiry has arrived.
Yes. Microsoft 365 and Google Workspace both let an approved system read a mailbox as mail arrives and save drafts back into it, so nobody has to change how they send email. Shared mailboxes such as info@ or accounts@ work the same way.
Yes, if it is given real examples. A voice profile built from a few dozen of your own sent emails, plus your actual prices, policies and customer history to draw on, gets drafts close enough that editing takes seconds. Without those, drafts sound generic and take longer to fix than writing from scratch.
It can be, if it is set up deliberately. Use business accounts or a system in your own cloud account, limit it to the mailboxes that need it, keep the data in Australia or New Zealand where you can, make sure nothing is used to train a public model, and tell staff what is being read and why. The Privacy Act 2020 applies to this like any other use of personal information.
A custom email automation build typically costs NZ$1,000 to $8,000 as a fixed price and takes two to five weeks. Running costs are a monthly hosting and support plan from $99, plus AI usage that is usually a few dollars to a few tens of dollars a month for a small business. Prices exclude GST.
Bring a typical day of your shared inbox to a free 30-minute call. I will tell you which parts are a rule, which need AI, and what it would cost to stop things slipping through. Discovery is always free.
Book a free call