How do I decide whether to use Microsoft Copilot, ChatGPT Team, or a custom AI assistant?
15 May 2026
How do I decide whether to use Microsoft Copilot, ChatGPT Team, or a custom AI assistant?
For most UK SMEs, the sensible order is: start with Microsoft Copilot if you are a Microsoft 365 business, add ChatGPT Business for teams doing writing, research, analysis, sales, marketing, coding, or operations work, and only commission a custom assistant once you can name the exact workflow, data sources, owner, compliance risk, and expected return. Copilot and ChatGPT are seat costs. A custom assistant is a project, normally from £3,000 to £40,000+ depending on integration, security, and workflow complexity.
The honest recommendation for most UK businesses
Here is the practical answer. If your team uses Outlook, Teams, Word, Excel, PowerPoint, SharePoint, and OneDrive every day, Microsoft Copilot should usually be the first platform you test. It sits inside the tools your people already use, which reduces training friction and helps with governance.
If your team needs better reasoning, writing, summarising, research, spreadsheet analysis, coding help, creative work, or flexible problem solving across departments, ChatGPT Business is usually the stronger second step. OpenAI's UK business pricing page currently lists ChatGPT Business at £15 per user per month when billed annually, with a 2 user minimum. It also states no training on your business data and support for GDPR, SOC 2 Type 2, SAML SSO, and MFA. Source: OpenAI ChatGPT pricing.
A custom AI assistant is not the place to start unless you have a specific job for it. Good custom assistants answer a narrow business question, follow your process, use your documents or systems, and create measurable value. Bad custom assistants are expensive demos that look clever for two weeks and then get ignored.
My bias is clear: we help businesses implement practical AI, including custom assistants where they make sense. That does not mean custom is always best. In many cases, Copilot or ChatGPT Business is the more sensible, cheaper, faster answer.
How the three options compare
The mistake is treating these as three versions of the same thing. They are not. Copilot is best understood as AI inside Microsoft 365. ChatGPT Business is a general AI workspace for teams. A custom assistant is a designed system for one or more business workflows.
| Option | Best for | Typical UK cost | Main risk | When to choose it |
|---|---|---|---|---|
| Microsoft Copilot | Microsoft 365 productivity, meetings, documents, email, Teams, Excel, PowerPoint | Microsoft UK pricing changes by plan, but paid Copilot seats are commonly around the mid £20s per user per month before VAT, depending on licensing and commitment | Paying for seats before staff know how to use it properly | You already live in Microsoft 365 and want controlled, organisation-wide productivity gains |
| ChatGPT Team, now ChatGPT Business | Writing, analysis, strategy, research, coding, sales support, marketing, operations, data interpretation | OpenAI UK page lists £15 per user per month billed annually for Business, with 2+ users | Shadow AI, poor prompting, unclear data rules, duplicated use alongside Copilot | You need a more flexible AI workspace and your work is not limited to Microsoft documents |
| Custom AI assistant | Specific internal workflows, knowledge retrieval, customer support, proposal drafting, compliance checklists, CRM tasks, operational automation | Usually £3,000 to £10,000 for a simple assistant, £10,000 to £40,000 for a serious operational assistant, and £40,000+ for complex integrations | Building before the process, data, owner, and ROI are clear | The workflow is repeated often, uses business-specific knowledge, and has measurable value |
There is also a fourth option: do nothing formal and let staff use free consumer AI tools. That is common, but it is not a strategy. It creates data leakage risk, inconsistent quality, and no shared learning. The UK government's AI Adoption Research found that only 1 in 6 UK businesses were using at least one AI technology in 2025, and that limited AI skills and ethical concerns were common barriers. Source: GOV.UK AI Adoption Research.
When Microsoft Copilot is the right answer
Choose Microsoft Copilot when the problem is mainly Microsoft 365 productivity. That means summarising Teams meetings, drafting emails in Outlook, turning Word documents into PowerPoint decks, analysing Excel sheets, finding information in SharePoint, and helping staff move faster inside the Microsoft environment.
Copilot is strongest when your files are already well organised and permissions are clean. That last point matters. Copilot can surface information from across Microsoft 365 based on the user's existing permissions. If your SharePoint structure is messy and too many people can access sensitive folders, Copilot does not magically fix that. It can make the mess more visible.
For a 30 person UK business, a broad Copilot rollout at roughly £25 per user per month is around £750 per month before VAT, or about £9,000 per year before you add training, adoption support, and internal admin time. That can be good value if staff save even 30 minutes a week. It is poor value if people keep using it as a slightly better spellchecker.
Copilot is usually the right first choice if you meet four conditions: you already pay for Microsoft 365, staff spend a lot of time in Teams and Outlook, your document permissions are reasonably tidy, and you want a managed enterprise tool rather than a loose collection of AI experiments.
When ChatGPT Business is the right answer
Choose ChatGPT Business when the work is broader than Microsoft 365. It is often better for open-ended thinking, long-form writing, planning, analysis, brainstorming, coding, structured research, role-play, document review, and turning messy inputs into useful outputs.
The price is also easier to justify for small teams. At £15 per user per month billed annually, a 10 person team is £150 per month, or £1,800 per year before VAT if billed in pounds. For teams that write proposals, produce reports, analyse data, handle marketing, or support clients, that is a small investment if adoption is managed properly.
ChatGPT Business is not automatically safer just because it is paid. You still need rules. Staff need to know what they can paste into it, what they cannot paste, how to check outputs, and when human review is mandatory. Under UK GDPR, if personal data is involved, you need a lawful basis, appropriate safeguards, and sensible controls. The ICO has been clear that AI does not remove normal data protection obligations. If your prompts include customer data, employee data, legal material, financial records, or health information, treat the rollout as a governance project, not a software subscription.
ChatGPT Business is usually the right choice for leadership teams, marketing teams, sales teams, analysts, consultants, developers, operations managers, and anyone who regularly turns information into decisions or documents.
When a custom AI assistant is worth the money
A custom AI assistant is worth considering when the task is repeated often, the answer depends on your business knowledge, and the value of a better process is obvious. Examples include a proposal assistant that uses your case studies and pricing rules, a customer support assistant trained on your policies, a compliance assistant that checks documents against an internal standard, or an operations assistant that pulls data from your CRM and produces next-step recommendations.
The honest cost range is wider because custom work is not a seat licence. A simple knowledge assistant using curated documents might cost £3,000 to £10,000. A proper operational assistant with user roles, testing, analytics, guardrails, workflow design, and integrations might cost £10,000 to £40,000. If it touches multiple systems, handles sensitive data, requires audit logs, or becomes business-critical, the budget can go beyond £40,000 quickly.
You should not build custom just because Copilot or ChatGPT feels generic. Build custom when the business case is boringly clear. For example, if 8 staff each spend 3 hours a week preparing similar client documents, and an assistant can cut that by half, you are saving 12 hours a week. At a blended cost of £35 per hour, that is £420 a week or about £21,840 a year. A £10,000 assistant starts to make sense. If the saving is theoretical, wait.
Usage data also matters. IAB UK reported that 24 million people used AI tools in January 2026 and that 76% of total time spent on AI tools went to ChatGPT. It also reported that 44% of time spent on AI tools occurred during work hours in August 2025. Source: IAB UK AI usage statistics. In plain English, your staff may already be using AI at work. A custom assistant can be a way to channel that behaviour into a safer and more useful process, but only if the workflow deserves it.
A simple decision framework
Use this order before spending money.
- Map the work. List the 10 tasks where staff lose the most time. Be specific. Do not write 'admin'. Write 'turning meeting notes into CRM updates' or 'drafting first versions of client proposals'.
- Separate broad productivity from workflow automation. If the task is general writing, summarising, or document work, start with Copilot or ChatGPT Business. If it follows a repeatable business process, consider custom.
- Check data sensitivity. Public information and internal templates are low risk. Customer records, employee data, contracts, legal advice, finance, and health data are higher risk.
- Estimate the value. If the saving or revenue upside is under £5,000 a year, do not build custom. Use an off-the-shelf tool and training instead.
- Pilot with 5 to 10 users. A smaller well-supported pilot beats a big announcement followed by low adoption.
- Measure adoption, not enthusiasm. Ask what changed after 30 days. Hours saved, documents produced, response time reduced, errors prevented, or revenue influenced.
For many UK SMEs, the best first 90 days looks like this: clean up Microsoft permissions, run a Copilot or ChatGPT Business pilot, create an AI usage policy, train staff on 5 to 8 real workflows, then decide whether one custom assistant is justified.
When this is NOT right for you
Do not buy Copilot for everyone if your Microsoft 365 environment is chaotic, your SharePoint permissions are poor, or nobody has time to train staff. You will spend money and get scattered usage.
Do not buy ChatGPT Business if your main goal is deep integration with Outlook, Teams, SharePoint, and Microsoft admin controls. It can still be useful, but it is not a direct replacement for Copilot inside Microsoft 365.
Do not build a custom assistant if the process is not documented, the data is messy, the owner is unclear, or the value is vague. Custom AI does not rescue a broken process. It usually exposes it.
Also do not rush into any of these if you handle sensitive personal data and have no data protection process. Start with governance, permissions, and a simple AI policy. It is less exciting than buying tools, but it prevents expensive mistakes.
The bottom line
If you are choosing today, here is the direct recommendation.
Pick Microsoft Copilot when your business runs on Microsoft 365 and you want broad productivity gains with familiar tools. Pick ChatGPT Business when you need flexible thinking, writing, research, analysis, and team collaboration outside the Microsoft bubble. Pick a custom AI assistant only when one high-value workflow is clear enough to design, test, measure, and improve.
If you want to explore which route makes sense for your business, book a free call. No pitch, no pressure, just an honest conversation about your current tools, risks, and likely return.
Is This Right For You?
This comparison is right for you if you run a UK business and are choosing where to spend AI budget in the next 30 to 90 days. It is especially useful if staff are already experimenting with AI, but you do not yet have a clear policy, platform choice, or implementation plan.
It is not right for you if you need a deep technical architecture review, regulated clinical advice, legal advice, or a formal procurement document. In those cases, use this as a first filter, then involve your IT, legal, data protection, and security leads before you buy or build anything.
Frequently Asked Questions
Is Microsoft Copilot safer than ChatGPT Business?
Not automatically. Copilot may fit more naturally into Microsoft 365 governance, but safety depends on your permissions, policies, staff training, and data handling. ChatGPT Business also includes business privacy and admin controls. The safer option is the one you configure and govern properly.
Can ChatGPT Business replace Microsoft Copilot?
Sometimes, but not if your main use case is working inside Outlook, Teams, Word, Excel, PowerPoint, and SharePoint. ChatGPT Business is stronger as a flexible AI workspace. Copilot is stronger as Microsoft 365 assistance.
Should every employee get an AI licence?
No. Start with the people who have clear use cases and enough work volume to justify the cost. For most SMEs, that means a 5 to 10 person pilot before a company-wide rollout.
When should we build a custom AI assistant instead of using Copilot or ChatGPT?
Build custom when the workflow is repeated often, uses your internal knowledge, needs system integration, and has measurable value. If you cannot explain the workflow and the return in plain English, do not build yet.
How much should a UK SME budget for a custom AI assistant?
A realistic starting range is £3,000 to £10,000 for a simple document or knowledge assistant, £10,000 to £40,000 for an operational assistant with integrations and governance, and £40,000+ for complex or regulated use cases.
Do we need an AI policy before using these tools?
Yes. It does not need to be a 40 page document. You need clear rules on what data staff can enter, which tools are approved, who reviews outputs, how errors are handled, and what happens with customer or employee data.
What is the best first step if staff are already using free AI tools?
Do not panic and do not ignore it. Run a short discovery exercise, identify the most common use cases, set minimum data rules, then move active users into an approved business tool with training and monitoring.