ChatGPT is free. Why would I pay you?

Short answer

A general assistant is enough when a person is in the loop every time, the volume is low, and a wrong answer is merely annoying. You need something built when the work is repeated at volume, has to draw on your own data, must connect to systems you already run, or when a confident wrong answer costs real money. If you are in the first situation, keep your money.

We lose work to this question, and we would rather lose it honestly than win it by pretending the free tool is worse than it is. It is very good. For a lot of businesses it is genuinely the right answer.

Where the free tool is enough

Drafting emails, descriptions and posts, which you then read and edit.
Translating and improving text, with someone who speaks the language checking it.
Explaining something you do not understand, as a starting point rather than an authority.
One-off analysis where you paste in the data yourself and sanity check the output.
Anything you would have done yourself in ten minutes, done in two.

Notice what those have in common. A person is present for every single one, checking the output before it matters. That is not a small detail, it is the entire safety mechanism.

Where it quietly stops working

The failures are rarely dramatic. Nothing crashes. You just end up with something that looks right and is not, or a process that never quite becomes a process.

The four walls you hit

It does not know your business. Your prices, your stock, your history, your clients. It will invent all of them convincingly.
It does not connect to anything. Every input is pasted by a person and every output is copied out by a person.
It does not repeat reliably. The same question can produce materially different answers, which is fine for a draft and fatal for a process.
It cannot be measured. There is no accuracy figure, no test set, and no way to know if it got worse this month.

What we add

Retrieval over your documents, with the source attached to the answer
Integration, so output lands in the system where the work happens
Constrained outputs that come out the same shape every time
A test set and a measured number, so quality is observed rather than assumed
A designed refusal path, so it says when it does not know
Something that keeps running when the person who set it up is on holiday

You are not paying for the model. The model is the cheap part. You are paying for everything that makes its output safe to act on without reading it first.

The honest arithmetic

The question is not whether the free tool can do the task. It usually can, once, with someone watching. The question is what happens at volume.

Take a task that takes a person four minutes with the free tool, including pasting the input, checking the output and copying it out. At ten a day that is forty minutes, and you should not hire anyone. At two hundred a day it is a full-time job that exists only because nothing is connected.

Somewhere between those two numbers, paying once for something built stops being an expense and starts being cheaper than not doing it. Where exactly depends on your volume and what an error costs you, and it is a calculation you can do yourself before you ever speak to us.

If you have run the numbers and you are near the line, tell us the task and the volume. If the free tool is still the right answer, we will say so in the first reply.

Tell us what you are trying to fix

Three situations where it is not a close call

The answer has to come from your own material. Your contracts, your archive, your catalogue. A general assistant does not have it and will confidently produce something in its place.
A confident wrong answer is expensive. Legal, financial, medical, regulatory, anything you sign. You need citation, verification and refusal, none of which comes as standard. That is the subject of stopping a model from inventing a citation.
It has to run without you. Overnight, at weekends, when the person who knows the trick has left. A workflow that depends on someone remembering the prompt is not a system.

What we will tell you for free

If you describe the task, we will tell you honestly whether the free tool covers it. We turn work down on this basis regularly, and there is a whole article about it: when you do not need AI. Turning down a project costs us one invoice. Selling one that gets abandoned costs the relationship and every referral behind it.

And if the answer is that the free tool is enough for now, that is a perfectly good outcome. Come back when the volume changes.

Frequently asked questions

Is ChatGPT enough for my business?

Often yes. If a person checks every output, the volume is low, and a wrong answer is only annoying, a general assistant is genuinely the right tool and you should not pay anyone to replace it. It stops being enough when the work is repeated at volume, has to draw on your own data, must connect to your systems, or when a wrong answer is expensive.

What does a custom AI system give me that a chatbot does not?

Access to your own data with the source attached to every answer, integration so output lands where the work happens, outputs that come out the same shape every time, a measured accuracy figure against a test set, a designed path for the system to refuse rather than guess, and something that keeps working without the person who set it up.

At what point does it become worth paying for?

When the time spent pasting inputs, checking outputs and copying results back adds up to more than the build costs, or when the cost of one confident wrong answer exceeds it. Both are calculations you can do yourself before contacting anyone.

Will you tell me if I do not need you?

Yes, and we do it regularly. Describe the task and the volume and we will say whether the free tool covers it. Delivering a project that gets abandoned costs us far more than declining one.

Is the free version safe for confidential business information?

Treat consumer chat tools as unsuitable for client data, contracts and anything privileged unless you have checked the specific terms that apply to your account. Where confidentiality matters, the data handling has to be established in writing, which is part of what a built system gives you.