Take a 20-person consultancy on Microsoft 365. Their IT provider recommended Copilot for the whole team. Six months in, three people used it daily: the ones who lived in Outlook, Teams and Word. Everyone else had switched to ChatGPT because it worked with the rest of their stack. The firm was paying for licences across the board and getting value from a fraction of them.
This is the most common version of a buying mistake I see with AI tools. The decision gets made on what sounds credible rather than what fits how the business actually works. This guide is designed to close that gap.
What is the actual difference between ChatGPT and Copilot?
ChatGPT, made by OpenAI, is a standalone AI assistant you access through its own app, mobile app, or API. It works with whatever you bring to it and runs independently of your existing software stack. Microsoft 365 Copilot is embedded directly into Word, Outlook, Teams, Excel and PowerPoint, and it draws on your existing emails, documents and calendars to help you work faster inside those tools, not alongside them.
The price difference reflects the different value propositions. ChatGPT Plus costs around £16 per month for an individual. Microsoft 365 Copilot runs at around £25 per user per month, charged on top of whatever your firm already pays for Microsoft 365 Business Standard or Premium. That premium reflects integration depth that only pays off when the integration is actually used.
The distinction comes down to context versus flexibility. Copilot’s value is that it lives inside tools your team already uses. It can summarise a Teams meeting while you’re still in it, draft a Word proposal drawing on related SharePoint files, or compose an Outlook reply using context from a calendar invite. ChatGPT works outside those tools, and what you get from it depends entirely on what you put in.
When does ChatGPT make more sense for your business?
ChatGPT suits firms whose work is varied, creative, or distributed across tools that are not part of the Microsoft ecosystem. It runs alongside Google Workspace, Mac, Linux and bespoke sector software without requiring a Microsoft 365 licence at all. Its Custom GPTs feature lets you build task-specific assistants, a proposal-drafting bot or a client-FAQ responder, in hours rather than weeks, with no engineering work required.
For owner-managed businesses that are not fully committed to Microsoft 365, or where only a handful of staff need AI assistance regularly, ChatGPT Team at per-seat pricing is also likely to be cheaper than rolling Copilot out to all users at £25 per month. UK Microsoft partner FormusPro notes that Custom GPTs are particularly well suited to fast, flexible internal tools where deep integration with Microsoft Graph is not needed.
The governance risk with ChatGPT is real and worth taking seriously from day one. When staff use it without clear policies, client data can end up in prompts without anyone realising what that means. In May 2023, Samsung restricted employee use of generative AI tools after discovering that staff had uploaded sensitive internal source code and meeting notes to external AI services. The root cause was a policy gap, not a flaw in the tool. The NCSC makes the same point: treat prompts sent to public AI services as potentially visible to the provider, and design your policies accordingly before staff start using them with real data.
When does Copilot earn its extra cost?
Copilot makes sense when a meaningful share of your team’s day is spent inside Microsoft 365, particularly on structured document, email and meeting work. It reads context from your SharePoint files, Exchange inbox and Teams conversations, then drafts, summarises or generates within those tools while respecting existing permissions, so staff only see what they would already be allowed to access.
The governance story is also cleaner for firms with regulated or enterprise clients. Microsoft’s existing contractual framework already covers Copilot, including its Data Protection Addendum and EU Data Boundary commitments. Microsoft states that customer content processed by Copilot is not used to train foundation models for other customers. If your contracts already reference Microsoft 365 as a sub-processor, Copilot sits inside that existing structure without requiring you to introduce and explain a new supplier relationship.
Microsoft’s 2023 early-access study, covering 600 users, reported that 70% said Copilot made them more productive and 68% said it improved the quality of their work. Treat vendor-run studies with appropriate scepticism. The use cases they describe, summarising meeting notes, drafting proposals from related documents, generating a first-pass Excel analysis, are genuine and repeatable. They just require that Microsoft 365 is already the main working environment, not a tool people use occasionally alongside other platforms.
What does getting this call wrong actually cost?
The cost of the wrong decision usually shows up in one of two ways. Over-buy Copilot licences for a team where staff mainly need occasional AI assistance, and you are paying around £6,000 to £9,000 a year for 20 to 30 users without the productivity gains to justify it. Under-govern ChatGPT, and you are one careless paste away from a client data incident that triggers a regulatory review.
The ICO expects organisations using generative AI with personal data to identify a lawful basis and, where processing is high-risk, to carry out a Data Protection Impact Assessment. Failure to do so can attract enforcement action, including fines of up to £17.5 million or 4% of global annual turnover for the most serious breaches. Choosing a tool without clarity on data flows does not reduce that exposure. It means you cannot demonstrate compliance if a client or regulator asks.
Lock-in risk deserves attention too. Building workflows around Copilot and Microsoft-specific extensions increases dependence on that ecosystem at a time when the Competition and Markets Authority is actively reviewing the Microsoft-OpenAI partnership and its implications for market competition. On the other side, building everything around ChatGPT’s API creates a parallel dependency if pricing, model access or usage terms change. Neither is a reason to avoid these tools. Both are reasons to design processes so you can adapt rather than rebuild if the landscape shifts.
What should you ask before you commit to either?
Before signing up for either licence, start with where your team actually works. If they are in Outlook, Teams and Word for the majority of the working day, Copilot’s in-context assistance is worth the premium. If your stack is mixed, or creative and research work dominates, ChatGPT covers more ground for less money per seat. Custom GPTs give you a practical path to building something specific to your operation without any engineering work.
The second question is about data. What will staff actually put in front of the AI? If the answer includes personal information, client files or regulated data, you need clarity on where that data goes, who can access it, and what your contractual obligations are to clients. Both Microsoft and OpenAI publish documentation on data locations and transfer mechanisms. Reading it is not optional if you are in a regulated sector or handling client data under a confidentiality agreement.
Third: how will you govern use before you scale? An AI acceptable-use policy that references ICO and NCSC guidance takes a day to draft and prevents the kind of incidents that create real problems. If you decide to standardise on one tool, you should also consider whether to block consumer alternatives at the network level, otherwise staff will use both, which is a harder governance position than choosing one deliberately.
The practical answer for many owner-managed businesses is to pilot with three to five staff on a specific use case for two to three months, tracking time saved and errors avoided, before committing to a broader rollout. That way the licence cost is a structured test rather than a sunk cost.
If you want a clear view of which option fits your team before you commit, Book a conversation.
