ChatGPT is genuinely useful in a business for drafting, summarising and reading documents you paste or upload into it. What it can't do on its own is see inside your job system, stock system, CRM or accounts, keep those systems up to date, or give you the same answer every time with a record of who approved what. If the work you want help with sits inside one document or one email, a business plan of ChatGPT (or a similar assistant) may be all you need. If the work moves between your systems, the missing piece is usually an integration or a workflow, with AI built into one step and a person approving what matters.
This article is for owners and finance or operations managers who've been asked "what are we doing about ChatGPT?" and want a straight answer before spending money.
What ChatGPT is genuinely good at
Used well, a chat assistant saves real time on work that starts and ends with words. Good uses in an established business include:
- Drafting: first drafts of quotes, tender responses, policies, procedures, letters to suppliers and routine customer emails. A person still edits and sends them.
- Summarising: long contracts, meeting transcripts, board papers, tender documents and industry guidance, boiled down to the parts you need.
- Reading documents you give it: pulling the key dates and terms out of a contract, listing the line items from a PDF quote, or comparing two versions of a document.
- Explaining and rewording: turning technical notes into plain English, or a rough procedure into clear steps.
- Thinking out loud: testing an argument, listing questions to ask a vendor, or planning a meeting.
All of these share one thing: the information goes in through the chat window and the answer comes out of the chat window. A person carries it from there.
What it can't do on its own
It can't see your systems
ChatGPT knows what you type, paste or upload, plus what it was trained on. It doesn't know which jobs are over budget, what stock is on the shelf, which customers owe you money or what was ordered yesterday, because that information lives in your job, stock, CRM and accounts systems.
Some business plans now offer connectors to common tools such as file storage and email, and the list changes often, so check what's current for your plan. But the systems that run an operation (an industry job system, an older stock database, a payroll or rostering system, a spreadsheet that's become the real system) are rarely on that list. And reading a file is a long way from updating a job record with the right checks.
So the usual pattern is a person exporting a report, pasting it in, asking a question, and copying the answer back somewhere. That's the same information typed in again, which is what we call double entry (the same information keyed into two places). It's also how mistakes creep in.
Privacy and data settings matter
What happens to the information you put in depends on the plan and its settings. OpenAI states on its business data page that, by default, it doesn't train its models on data from its business plans and API. Free and personal plans can work differently, and terms change, so check the current terms for the exact plan your staff use, and decide what can and can't go in. Our AI policy for staff post has a template for that.
If your business is covered by the Privacy Act, personal information needs extra care. The OAIC has guidance on commercially available AI products; as best practice, it suggests not entering personal information into publicly available AI tools. We're not lawyers, so get advice if you're unsure.
It isn't consistent
Ask the same question twice and you can get two differently worded answers, sometimes with different numbers. That's fine for a first draft. It's not fine for a stock count, a margin figure or a payroll amount. Anything that needs the same answer every time belongs in a system with rules, not a chat.
There's no audit trail in your systems
When someone asks ChatGPT to work something out and then keys the result into your accounts, your accounts show the person keyed it. There's no record of what was asked, what source data was used or who checked it. For anything that affects money, stock, pay or customers, finance and auditors will want to know where a number came from. A workflow can record that; a chat window can't.
Microsoft Copilot and the other assistants have the same basic limits. Whichever one your staff use, it only knows what it's been given access to, and it doesn't replace the connections between your systems.
The real question: is the work in the chat, or between your systems?
Most of the "AI" requests we hear about turn out to be about something else. When an owner says "I want AI to tell me our margin by job every Monday", the hard part isn't the AI. It's that the job costs sit in one system, the invoices in another, and the timesheets in a spreadsheet. That's a data silo problem (information stuck in separate systems that don't share it), and our why your systems disagree post explains it in plain words.
Once those systems are connected and the data is clean, ordinary reporting answers the Monday question, and AI becomes a useful extra rather than the foundation. If you're unsure which of these you need, our companion article AI, automation or integration? Which one your business actually needs walks through it, and the AI fit check gives you a quick read in a few minutes.
When you need an integration or a workflow instead of a chat window
Here are some common requests, and what usually sits underneath them.
"I want ChatGPT to enter supplier invoices." For example, if your team reads PDF invoices and keys them into the accounts, then checks them against purchase orders in another system, the useful fix is a workflow: AI reads the invoice, the workflow matches it to the order and the delivery, and a person approves anything that doesn't line up. Our AI workflow automation page explains how we build that.
"I want AI to answer questions about our jobs." That needs the job data to be reachable and trustworthy first. Often the first step is joining up the job system and the accounts so there's one set of numbers, then reporting on it. See systems integration for growing businesses.
"I want AI to tidy our customer list." AI can suggest likely duplicates, but someone has to decide the rules, and the clean list has to flow back into every system that uses it. That's a data cleansing job with AI helping, not a chat.
"I want AI to write our reports." If the numbers going in are late or disagree, a well-written report is still wrong. Fix where the numbers come from first.
A simple decision guide
Ask these questions about the task you have in mind.
- Does the work start and finish in one document or one conversation? A chat assistant on a business plan is probably enough. Write a short staff policy and start.
- Does someone copy information out of one system, ask AI, then copy the answer into another? You need an integration, so the information moves without being retyped.
- Does the task need the same answer every time, or touch money, stock, pay or customers? You need a workflow with rules and a person approving, with AI doing only the reading or drafting step.
- Do people argue about which number is right? You need clean, connected data before AI will help. Start there.
- Not sure? Pick one repetitive task and run a small, time-limited trial. Our guide to running an AI pilot safely sets out the steps.
What it looks like when it's working
Picture a Monday morning. Supplier invoices that arrived over the weekend have already been read and matched to orders and deliveries. Your accounts officer approves the ones that match in a few minutes and looks properly at the three that don't. The margin report for last week's jobs is ready, from one set of numbers. Staff still use a chat assistant every day for drafting and summarising, under a simple policy. Nobody is pasting exports into a chat window and copying answers back.
That's AI built into the work, with a person in charge, rather than AI sitting off to one side.
Where we fit
We don't sell ChatGPT plans or set up office software. We help with the parts a chat window can't do:
- Connecting your systems so job, stock, CRM, payroll and accounts data moves without being retyped.
- Building AI into workflows for the reading and drafting steps, such as invoices, dockets, orders and emails, with a person approving anything that matters.
- Cleaning the data so reports and AI are working from one set of numbers.
If you want an independent view of where AI fits in your business before committing to anything, our AI consultant for small business page explains how we work. The AI readiness assessment looks at your processes, systems and data and tells you which tasks are worth trying first, and the back-office AI pilot tests one of them on real work with a fixed end date.
Frequently asked questions
Is ChatGPT safe to use for business?
It can be, with the right plan and rules. Use a business plan whose terms say your data isn't used for training by default, check the current terms yourself, keep personal information out where you can, and write a short staff policy on what can go in.
Can ChatGPT connect to our accounting or job system?
Some business plans connect to common tools such as file storage and email, and that list changes. Most industry job systems, stock systems and older databases aren't on it. Connecting those, so data moves and records update with the right checks, is integration work.
Should we buy ChatGPT business plans for all our staff?
That's a decision about drafting and summarising, and it can be a reasonable one. Just don't expect it to fix work that moves between systems. Check current plans and terms with the provider; we don't resell or recommend particular plans.
What about Microsoft Copilot?
It's another assistant with the same basic limits: it only knows what it's given access to, and it won't connect your operational systems to each other. We don't set up Microsoft 365 or Copilot; we work on the systems and workflows around them.
Can ChatGPT do our month-end or our reporting?
It can help explain figures and draft commentary. The numbers themselves should come from your systems through a repeatable process, so they're the same every time and you can trace where they came from.
If you'd like to talk through where ChatGPT helps and where your business needs connected systems instead, book a free initial chat.