Three options are on the table. Add Copilot to the Microsoft licences you already pay for. Buy ChatGPT for the team. Or have something built that fits how your business actually works.
Everyone selling one of them has a confident view. The decision is genuinely simpler than the sales conversations suggest, and it turns on one question that nobody tends to ask first.
The question that decides it
Does the useful answer depend on your own information?
If the work is generic, drafting, summarising, rewriting, explaining, then a general tool is fine and the cheapest adequate one wins.
If the useful answer requires knowing what is in your files, your emails, your customer records or your policies, then you need something connected to those things, and that narrows the field immediately.
Almost every disappointing AI rollout comes from getting this backwards: buying a general tool and expecting it to know things it has never been shown.
Copilot
The argument for Copilot is proximity. It sits inside the software your team is already in, and it can reach the documents, mail and messages they already have, with existing permissions respected.
That last point is the real value and it is consistently undersold. If someone cannot open a file today, Copilot will not surface its contents to them tomorrow. Your existing access controls carry over, which removes an entire category of worry.
It suits you if the business runs on Microsoft 365, the work is document-heavy and meeting-heavy, and the value is in summarising and finding things across material you already hold.
The honest caution is that Copilot inherits the state of your files. If your SharePoint is a decade of unstructured folders with permissions nobody has audited, it will confidently surface things to people who technically had access and were never meant to look. That is not a flaw in the tool, it is an existing problem becoming visible. Tidy first.
ChatGPT and the general assistants
Better at open-ended thinking, drafting and reasoning through a problem. Cheaper per seat. Faster to adopt, because there is nothing to configure and people already know how to use it.
It suits you if the work is writing, planning, analysis and general problem-solving, and staff mostly bring the context with them in the prompt.
The honest caution is the account tier. A business tier is not a nicety, it is the difference between a defensible arrangement and an undocumented disclosure of data to a third party, which matters considerably once customer information is involved.
Something custom
Custom does not mean training your own model. Essentially nobody your size should be doing that, and anyone proposing it should be asked why very directly.
It means using an existing model inside something shaped around one specific job: connected to your data, constrained to a particular task, with the output landing where the work actually happens rather than in a chat window somebody has to copy from.
It suits you if there is one repeated, well-defined, expensive task, and the general tools keep almost working but need too much hand-holding to be worth it.
The honest caution is that this is software. It has a build cost, a running cost and a maintenance cost, and it needs an owner. The right time is when you can name the task, count how often it happens, and estimate what it currently costs. If you cannot do all three, it is too early.
A decision you can make this afternoon
Work through these in order and stop at the first yes.
- Is the value mostly in your own documents and email? Copilot, assuming you are already on Microsoft 365 and your files are in reasonable shape.
- Is it mostly writing, thinking and drafting? A general assistant on a business tier.
- Is it one specific repeated task the general tools keep half-solving? Now a custom build is worth pricing.
- Can you not name the task? Buy nothing yet. This is the most common position and the correct answer is to wait.
Landing on four is not a failure. It is the honest result for a large share of businesses, and acting on it saves more money than any of the other three answers.
On paying for both
Plenty of businesses end up with both, and it is not automatically waste. They are genuinely different tools.
It becomes waste when nobody decided. Licences bought in a burst of enthusiasm, assigned to everyone, used properly by three people. That is the most common way AI budgets get spent badly, and it is worth checking against what this is genuinely costing you before renewing anything.
Buy for a named group with a named job. Expand when they ask for more seats, which is the only reliable signal that something is working.
The honest summary
Most businesses need one general assistant on a business tier, and nothing else, for at least a year.
Copilot earns its price where the value is locked inside your own documents. Custom earns its price where one specific task is expensive and repeated. Everything else is enthusiasm.
If you are weighing a proposal and cannot tell which of the three it actually is, describe what has been quoted and you will get a straight read, including if the answer is that you do not need it yet.
Common questions
Do we need both Copilot and ChatGPT?
Rarely at the start. They solve different problems, and paying for both before anyone has a named task is the most common way this budget gets wasted. Buy one for a named group and expand when they ask for more seats.
Is Copilot worth it if our files are a mess?
Not yet. Copilot inherits the state of your files and their permissions, so a decade of unstructured folders produces confident answers drawn from documents people were never meant to open. Tidy first.
When does a custom build actually make sense?
When you can name the task, count how often it happens, and estimate what it currently costs. If you cannot do all three, it is too early, and a subscription will tell you more for less.