ChatGPT or a custom build, what is the actual difference?
The ChatGPT vs custom AI question comes down to who is doing the work. A subscription puts a capable assistant next to a person, and that person still has to be at the desk, still has to ask, and still has to paste the answer somewhere. A built machine sits inside your process and does the step itself, whether anyone is at the desk or not.
Both are legitimate. They are just answers to different questions, and a lot of money gets spent because someone asked for the second and bought the first.
When is a subscription enough?
When the work is varied, occasional, and already being done by a person who is happy to do it.
Writing a tender response. Rewriting a job ad. Making sense of a long document. Getting unstuck on a spreadsheet formula. Drafting the awkward email. These are one off, high judgement tasks where the value is in making a capable person faster, and a general tool is genuinely excellent at them.
If that describes your problem, buy the subscription, train the team on it, write the one page policy about what data may go in, and spend nothing else. That is not a lesser answer. It is the right answer, and we will tell you so.
When is a subscription not enough?
When the task is repetitive, happens whether or not anyone is available, and has to touch your own systems to be useful.
The tell is usually one of these. The work arrives at times nobody is working, like the enquiry at nine at night. It needs your actual data, your price book, your job history, your stock, your client files, not general knowledge. It has to write back into a system, not just produce text. It has to happen consistently, the same way, every time, without depending on which staff member is on that day.
A subscription cannot do those, not because the model is not smart enough, but because nobody is there to operate it. That is the whole gap. Our post on what to automate with AI first works through how to pick the lane where that gap costs you the most.
Is a custom build just a wrapper around the same model?
Partly, and that is not the interesting part. The model is a component you can swap. The build is everything around it.
What you are paying for in a build is the connection to your systems, the rules that decide what happens in which case, the handling for the awkward twenty percent, the approval gates that keep a person accountable for what goes out, the logging that lets you prove what it did, and the maintenance that keeps it working when a supplier changes a format. The model is the cheapest part of the machine and the easiest to replace.
Anyone selling you a build who cannot describe those parts is selling you a wrapper, and you can spot it by asking what happens when the input is unusual.
What does each one cost you?
A subscription costs a small monthly fee per person and your team's time to operate it. A build costs a project fee and a monthly fee, and buys back the time entirely.
We publish our prices, so you can see the shape of the second answer on the solutions page rather than having to sit through a discovery call to find out. What we would say about comparing the two is that the honest denominator is not dollars, it is hours. A subscription that saves each of five people two hours a week is doing more for you than a build that automates a task nobody was spending time on.
Can you start with one and move to the other?
Yes, and starting with the subscription is usually the sensible order for a business that is not yet certain.
Run the subscription first. Watch where your team actually uses it, and more importantly, watch where they keep using it for the same task in the same way. Repetition is the signal. A prompt that someone has pasted forty times is a process that wants to be a machine, and now you have forty examples of the correct output, which makes the build faster, cheaper and far more likely to be right.
That evidence is worth more than any workshop. It is also why we would rather start with a working demonstration on your own data than a slide deck, which is set out in our method.
How do you decide this week?
Answer three questions honestly. Does this work happen when nobody is available? Does it need our own data to be any good? Does it need to be done the same way every time?
Three noes means buy a subscription and get on with your day. Three yeses means you have a build. A mixture means start with the subscription, keep a note of what it cannot reach, and revisit in a quarter with real examples in hand.
If you want a second opinion on which side of that line your problem sits, get in touch and describe the process. We will tell you if the answer is a subscription, and it costs you nothing to hear it.
For general guidance on digital tools and obligations for Australian businesses, business.gov.au is the plain English starting point, and the AI Ethics Principles are worth reading before you commit either way.