ChatGPT for business: plans, pricing and when each one pays off

In many SMEs, ChatGPT adoption has already happened without anyone deciding on it. Someone in admin opened a personal account, pasted in a contract to get a summary, and it worked. The following month three more people were doing it, each with their own account and their own card, and the company has no idea which documents have gone through it.

From there the conversation turns into a purchasing decision, and it almost always comes framed as “what is the best AI for business?”. Reading comparisons is a poor way to answer that question, because for office work the major assistants do roughly the same thing and you notice the difference by trying them for a week with your own documents. Other things are decided on paper: which plan you sign up for, what commitment you get on your data and whether you want your team to have a tool or a process to run without anyone having to remember it.

Free, Plus, Business and Enterprise: what really sets the four plans apart

It is worth getting one expectation out of the way. “ChatGPT for business” means the same chat interface you have already used, with an admin layer on top and a different contract behind it. OpenAI sells it in four tiers: a free plan, a paid personal plan, a team plan billed per user and a custom contract for large organizations. The product names and the split of features between tiers change often, so check the plans page on the day you sign up, not in an article you read months ago.

What you get when you move from the free plan to a paid one

For a one-off query on a quiet day, the free version answers well and you won’t notice anything odd. The difference shows up when usage stops being curiosity and becomes routine, when someone opens it again and again throughout the morning and needs it to be available every time.

What you buy when you move to the paid plan is, above all, predictability: access to the most capable models without being rationed halfway through a task, higher usage limits, long files and workspaces where the assistant keeps a project’s context between sessions. For someone who uses it half an hour a week none of that matters, and for someone who has it open all day it is the difference between a tool and a toy that switches off when you need it.

There is a second difference that matters more in a company, and it isn’t about features. A free account is in one person’s name, with their email and password. The day they leave, the history leaves with them, and while they are there, nobody can audit what they have uploaded. The personal plan doesn’t help here either. What you need is a team plan, and that is a different decision.

When Business makes sense and when you need to negotiate Enterprise

The question people type into Google is how many people can use ChatGPT for business, and the useful answer goes the other way: from how many does it pay off. The threshold comes early. As soon as there are two or three people using it for real work, and above all as soon as someone has to add new starters and cut off access for people who leave, scattered individual accounts become unmanageable. That is what the team plan is for: one invoice, a user dashboard and a data commitment that applies to the whole organization instead of depending on each person ticking a box. It usually requires a minimum number of seats, which you need to check when signing up.

Enterprise comes in when requirements that have to be negotiated one by one appear. Someone has to sit down and agree how it integrates with the corporate identity system, how long data is retained and what support is put in writing. It has no published price because it is agreed case by case. A reasonable indicator: if nobody in your company has the role of negotiating a software contract with a sales rep, you are not ready for that tier yet.

How much ChatGPT costs per person per month, and the bill nobody budgets for

Personal and team plans are billed per user per month, in the region of €20 to €25 a month per person. That is an order of magnitude, not a verified rate for your case, and it varies with the specific plan, the country and its VAT, and whether you pay monthly or commit to the year upfront. Check the exact amount on the pricing page on the day you sign up.

Enterprise doesn’t publish a figure. It is quoted by number of seats and length of commitment, and in that conversation they will ask you how many people will use it and what security requirements you have to meet.

And then there is the line item almost nobody budgets for, because it doesn’t look like a subscription. If you connect the model to a process, what you pay for is the API, and there are no seats there: you pay for the text you process. A workflow that sorts emails all day can cost less than a subscription or much more, depending on how much text you feed it and how many passes it makes. You can’t estimate that number off the top of your head; you measure it with a small pilot.

The most expensive part of an automation project doesn’t usually show up on the OpenAI bill. What’s expensive is the hours: building the workflow, testing it with real cases until it stops breaking and going back to it when the provider changes something. In the quotes we prepare, that line weighs more than the model.

Buying licenses is not the same as buying automation

This is where an SME spends its money in the wrong place. It signs up for a team plan for the whole staff expecting the company to automate itself, and what it has bought is chat accounts, with admin controls and a decent contract, but chat accounts all the same. A few months later half the licenses still haven’t been opened and the conclusion drawn is that AI didn’t work, when what didn’t fit was the product they bought.

There is work someone sits down to do and work that happens on its own

The subscription takes care of work that starts when a person sits down and decides to do it. Someone drafts a proposal, summarizes the minutes of a meeting or rewrites a product page, and each time they open the chat, paste in the context and judge whether the result is good enough. That work improves a lot with a good subscription and needs no code.

The other kind of work is the kind that has to happen even if nobody remembers. Every order that comes in, every form that gets filled in, every invoice that lands in the inbox. There, the model is one piece of a workflow built in an orchestration tool such as n8n, Make or Zapier, with its trigger, its error handling and its log. A boring example: an email comes in, one step classifies it and extracts the data, another creates the record in the CRM and another sends an alert if something doesn’t add up. That is paid for by API usage and it works at three in the morning.

A team plan doesn’t trigger a single process

A team plan gives you user management, a single invoice, a shared workspace and better terms for your data. None of those four things triggers a process. Nothing kicks off when an email arrives, nothing writes to your CRM, nothing retries if a call fails. You still need a person to open the chat.

What automates is everything around it: a trigger that notices something has happened, credentials to read from and write to your systems, a plan for when the model returns something strange and a place where it all gets recorded. The model is the easy part of the setup. The hard part, and where almost every project that reaches us gets stuck, is having your data accessible and organized.

It is also worth knowing when it doesn’t pay off, which is half the answer. A process that happens a handful of times a month, or whose rules change every two weeks, doesn’t pay back the setup or the maintenance, and we usually advise against it even if it is technically possible. There, a subscription with a person in front of it is cheaper and gets going the same day.

Your data: what is used for training and what changes when you sign a business plan

The question always comes up on the same day, when someone wants to paste the customer list into the chat and someone else tells them not to even think about it. It is worth having the answer before that moment.

On consumer accounts, conversations may be used to improve the models unless this is turned off in the settings, and turning it off depends on each person remembering to do so. On business plans the starting point is the opposite: workspace content is not used for training by default, and whoever administers it controls access and how long the history is kept. For many companies that is the compelling reason to move away from scattered accounts, more than any added feature. The specific terms are in the plan’s contract and should be read before signing.

There is something that contract doesn’t change. Anyone with access to the shared workspace sees what is inside, so uploading a folder of sensitive data there is the same as handing it out to the whole team. And if you process personal data about customers or employees, the business plan doesn’t spare you the data protection work: you still need to review the legal basis and the provider’s role as data processor. Spend half an hour with whoever handles your GDPR before uploading anything.

With API automations the framework is different from the chat, and the weak spot is usually in your own workflow rather than with the provider: where you store the copy of the processed document, who reads the execution logs, what happens to the data if the orchestration runs on a third-party service. With n8n on your own server you control that part, with all the good and bad that comes with controlling it yourself.

At La Teva IA that is exactly where we start, looking at which processes you have, which ones can be automated with enough quality and which ones aren’t worth it, before recommending any tool or plan. If you want to have that conversation, tell us which process is eating up your hours.