What an AI CRM does, and what it should never do without you

An AI CRM is a customer database with a reader and a drafter built in. The database part keeps your contacts, enquiries, deals and the history of each in one place. The AI part reads the messages that come in, tells you what each one is, drafts a reply and suggests what to do next. What it should never do without you is anything a customer can see or rely on: send a message, state a price, or close, delete or change a record. A good one prepares; a person decides.

What is a CRM, before any AI is added?

CRM stands for customer relationship management, and for a small business it means one list instead of five. The enquiry that arrived by email, the phone call you took in the car, the quote you sent last Tuesday and the invoice that followed all hang off the same customer, so anyone who opens the record can see where things stand.

Without one, that history lives in an inbox, a notebook, a spreadsheet and your head. It works until two enquiries arrive on the same afternoon, or until someone else has to answer for you.

What does the AI part actually do?

Mostly the reading and the first draft. In practice the jobs are these:

• Reading. It reads an incoming message and says what kind it is: a quote request, a complaint, a question, a newsletter. That lets the urgent ones rise to the top.
• Summarising. One line instead of six paragraphs, and on a long deal, where things stand and what is still unknown.
• Drafting. A reply written for you to read, correct and send.
• Suggesting. The next step: open a lead, add a task, ring them back, move the deal on.
• Filling in. A phone number or job title from an email signature, offered for a field that is still blank.

None of that is judgement. It is preparation, and preparation is where a small business loses its evenings.

What should an AI CRM never do without you?

Anything that leaves the building, and anything that cannot be undone. The line is easiest to see job by job.

The job	Fine for the AI to prepare	Needs a person to press the button
A customer's email arrives	Read it, label it, summarise it	Decide whether it becomes a lead or is ignored
A reply is needed	Write a draft	Read it, change it, send it
A price is asked for	Find the price you already wrote down	Quote any figure that is not already yours
A deal has gone quiet	Point it out and suggest a next step	Move it, close it or mark it lost
A record is incomplete	Offer the missing detail	Accept it, and never overwrite what a person typed
A quote or invoice is due	Gather the lines	Generate it, number it and send it

The test for the right-hand column is simple. If the customer could hold you to it, or if you could not take it back, a person should have pressed the button, and the record should show who did.

Why does "a person presses Send" matter so much?

Because the AI will sometimes be wrong, and the cost of being wrong depends on who saw it. The Office of the Australian Information Commissioner says so directly in its guidance on commercially available AI products: "AI systems are known to produce inaccurate or false results", and the risk is especially high with generative AI, which the guidance describes as probabilistic in nature.

A wrong sentence in a draft is a small thing: you read it, fix it and move on. The same sentence sent to a customer is a statement your business made. That is why the same guidance tells an organisation choosing an AI product to consider "how human oversight can be embedded into processes". In a CRM, the Send button is that oversight.

What happens to your customers' details?

This is the question to ask before any other. The OAIC's guidance says privacy obligations apply to any personal information put into an AI system, and to what the AI produces where that contains personal information. It also recommends, as best practice, that organisations do not enter personal information into publicly available generative AI tools.

Pasting a customer's email into a free chatbot is exactly that. An AI built into the system that already holds your customer records is a different arrangement, but only if the vendor can tell you what the arrangement is. The guidance also says businesses should update their privacy policies with clear information about their use of AI. It is written for organisations with obligations under the Privacy Act; whether or not yours is one of them, its questions are the right ones to ask.

What should you ask before you choose one?

Six questions, and a vendor that cannot answer them in a sentence each has answered them.

1. What exactly is sent to the AI model, and is anything hidden first?
2. Can the AI send, change or delete anything by itself, and where is that written down?
3. Is the AI's reading switched on by default, and who can switch it off?
4. Does the record show which steps the AI prepared and which person took them?
5. Where does a price in a draft come from?
6. Is the AI a separate purchase, and what are its limits?

The same questions apply to quotes: an AI that prepares the lines is useful, and what the finished document must show is covered in quotes and tax invoices with GST.

How does HarvestFlow Grow draw the line?

HarvestFlow Grow is an AI CRM for Australian small businesses, and it answers the six questions this way.

What is sent. The AI reads the email you forward to your workspace's own address, the replies to email you sent from the workspace, and a chat or call notes you paste in. The subject and text go to an AI model with the names and contact details your workspace knows hidden first. It does not read chat apps or social accounts, and it does not connect to your mailbox: you forward to it.

What it can do by itself. Read, rank and suggest. It answers what kind of message it is, a one-line summary, the products named and how urgent it reads. The AI never sends, files or changes anything; every suggestion is a button that waits for you.

Who switches it on. Reading incoming email is off in every workspace until its owner turns it on, and the owner can turn it off at any time.

What the record shows. Press Draft a reply, change what you like, then press Send. Your timeline shows the email as drafted by AI and sent by you, and each step the AI suggested and you took is marked on the record.

Where a price comes from. The draft never writes a price of its own. A figure reaches a draft only inside an entry you wrote yourself, quoted word for word.

What it costs. The AI is on every plan, from the free start to Pro, with a daily limit and a monthly allowance set by the plan. The limits are on the pricing page, and the AI page lists everything it does and everything it never does.

Following up is where this division of labour shows most plainly. The AI can tell you a deal has gone quiet and draft the message; whether to send it, and when, is yours, which is the subject of how to follow up a quote without becoming a nuisance.

General information, not legal advice.