Microsoft Copilot vs Custom AI Automation for UK SMBs

Microsoft has spent Q1 2026 pushing Copilot harder than any product launch in the M365 era. Most UK SMBs running Microsoft 365 have either deployed it or sat through a Microsoft partner pitch about deploying it. The renewal conversation is now happening in finance directors’ inboxes across the country.

The question being asked is rarely the right one. “Is Copilot worth £24.70 per user per month?” pulls the conversation toward a yes or no on a single product. The better question is which jobs Copilot is shaped to do and which it is not, and where the rest of your work needs to be served by something else.

This post sets out what Copilot does well, where it stops, and how to think about the trade-off before your next M365 renewal.

The real question is where your work actually happens

Microsoft Copilot’s value is locked inside the Microsoft 365 boundary, and the right way to evaluate it is to ask how much of your work actually sits inside that boundary. Inside it, Copilot is genuinely strong. Outside it, anywhere your work touches non-Microsoft systems, customer-facing channels, or autonomous workflows, Copilot does not reach.

For a UK SMB whose people spend most of their working day in Outlook, Teams, Word, Excel, and SharePoint, with little need to move data between business systems, Copilot is a productive investment. For a UK SMB whose work routinely crosses into Salesforce, HubSpot, Xero, Zendesk, customer-facing email, or autonomous data flows, Copilot reaches a useful but limited part of the working day, and the bigger productivity gains sit elsewhere.

Most UK SMBs we audit fall in the second category. The Copilot question for them is not “do we want a smarter Outlook” but “what is the right shape of AI investment given where our work actually happens.”

What Copilot does well inside the Microsoft 365 boundary

Three points up front:

• Copilot reads and acts on the content already inside your Microsoft 365 tenant: Outlook mail, Teams chats and meetings, SharePoint documents, OneDrive files, Excel data, Word drafts.
• It produces real productivity gains for individual workers in roles where the work is mostly composing, summarising, analysing, or extracting information from those sources.
• The integration is tight, the security model is unified with the rest of Microsoft 365, and deployment is largely a licensing decision rather than an engineering project.

Concrete strengths in practice. A salesperson can ask Copilot to draft a follow-up email referencing the last three meetings with a customer, and it works because both the email history and the meeting transcripts are in the tenant. A finance lead can summarise a long Excel model into talking points for a board meeting without copy-pasting between tools. An operations manager can search across thousands of SharePoint documents in natural language and get reasonable answers with citations.

The deployment side matters too. If you have already paid for the SharePoint and information governance work that Copilot needs to be safe — labelled documents, sensible permissions, clean tenant structure — you can be live within weeks. The work to get there is documented in the Copilot readiness checklist for SharePoint deployments, and most SMBs underestimate it the first time round.

For a knowledge worker spending six hours a day in M365 apps, Copilot can reasonably save 30 to 60 minutes a day on drafting, summarising, and information retrieval. At £24.70 per user per month, that is a strong line item for the right roles.

Where Copilot stops at the boundary edge

Three points:

• Copilot can read the content in your tenant, but it cannot reliably trigger actions in non-Microsoft systems without significant additional engineering.
• It works well for human-in-the-loop drafting and analysis. It does not run on a schedule, react to external events, or execute multi-step workflows autonomously.
• The “Copilot extends to your business systems” pitch usually means Power Platform connectors plus custom development, which carries a separate cost, complexity, and maintenance footprint that is rarely included in the per-user price.

What this means in practice. A salesperson can ask Copilot to draft a follow-up email, but Copilot will not write back to your CRM, update the deal stage, schedule the next touch, or trigger a sequence in a marketing tool unless you have built that integration through Power Automate or a third-party connector layer. Each of those builds is its own project.

The same boundary problem appears elsewhere. Customer support tickets in Zendesk are invisible to Copilot unless integrated. Invoices in Xero are invisible. Lead form submissions hitting a website outside the tenant are invisible. WhatsApp messages, SMS, and customer-facing email channels are invisible. The list is long, and it is the list that most SMBs spend a lot of their working day on.

There is also a security dimension. Copilot’s reach inside the tenant is one of its strongest features and one of its biggest risk surfaces. The over-permissioning problems documented in the security risks that come with Copilot’s broad SharePoint access are not theoretical. They affect almost every tenant where SharePoint permissions were set up before AI was a consideration.

What custom AI automation reaches that Copilot cannot

Three points:

• Custom AI automation runs autonomously, on a schedule or in response to events, across whichever systems your work actually touches.
• It integrates deeply with non-Microsoft tools through purpose-built API workflows, not generic connectors.
• It executes complete processes end to end, with business logic, exception handling, and routing built specifically for how your business operates.

A practical example. A custom AI workflow watches a shared Outlook inbox where new supplier invoices arrive. Each invoice is extracted, validated against existing purchase orders in your accounting system, routed for approval through Teams, posted to Xero or QuickBooks, and acknowledged with the supplier. All of it happens without a person opening Outlook. This is the same family of work covered in the document processing patterns that span beyond Microsoft 365, and it is the most common first build we ship for SMBs hitting the Copilot boundary.

Copilot can help draft the supplier acknowledgement once a person opens the invoice. The custom workflow eliminates the person opening the invoice. That is a different category of value, not a better version of the same value.

The scope of work that fits this pattern is wide. Lead routing across web forms, CRM, and email. Customer support triage across Zendesk, email, and Slack. Reporting that pulls data from advertising platforms, CRM, and analytics into client-ready decks. Reconciliation alerts that span banking feeds and the accounting ledger. None of these sit inside the Microsoft 365 boundary, and none of these are Copilot’s job.

The cost picture is more complicated than the per-user line

The headline cost comparison is misleading because Copilot bills per user per month while custom automation bills as a fixed build cost plus low recurring infrastructure. The right way to compare is to look at total annual cost at your team size and ask how much of the work is actually being addressed for that money.

The table below sets out indicative annual costs for a UK SMB at three sizes, with assumptions stated below.

Assumptions. Copilot at £24.70 per user per month, applied to 60 per cent of seats (a realistic adoption rate for SMBs where not every role gains from Copilot). Custom workflow cost ranges based on agency-observed UK SMB engagements for a single well-scoped process. Run cost covers API consumption, hosting, monitoring, and basic iteration.

The crossover does not work the way the per-seat number suggests. A 75-seat business spending £13k a year on Copilot is paying for productivity gains spread thinly across 45 people. The same business spending £12k once on a custom invoice processing workflow is removing a single recurring cost centre that was eating ten or more hours per week. Both can be true at the same time. The honest comparison is in the side-by-side Microsoft Copilot vs custom AI comparison, which breaks down what each spend actually buys.

The key trap to avoid. Treating the Copilot per-user line as the total AI budget. For most UK SMBs above 50 employees, the Copilot spend addresses individual productivity, and a separate budget is needed for the workflow automation that addresses operational drag.

Three workflows where Copilot is the right answer

Three points:

• Knowledge work that sits almost entirely inside Microsoft 365.
• Roles where the productivity gain comes from faster drafting, summarising, or information retrieval rather than process execution.
• Situations where the human is meant to stay in the loop and Copilot’s job is to make them faster, not to replace them.

Three scenarios where Copilot wins clean.

A 40-person professional services firm where consultants spend most of their day in Outlook, Word, and SharePoint. Copilot drafts client emails referencing prior correspondence, summarises long client documents, and extracts key points from Teams meetings. Each consultant saves 45 to 75 minutes a day. The business case writes itself.

A 60-person finance team where analysts work primarily in Excel and SharePoint. Copilot accelerates model building, surfaces patterns in data, and produces narrative summaries for management reporting. The work was already in M365, the human stays in control, the gain is real.

A 30-person legal team using Word, SharePoint, and Outlook for the bulk of contract work. Copilot drafts initial review notes, summarises long documents, and extracts contract terms. The lawyers still own the decisions, but the time spent on first-pass work falls sharply.

In all three, the pattern is identical. Work lives inside the M365 boundary, the human stays in the loop, the gain is per-person productivity. Copilot is the right shape of investment.

Three workflows where custom automation pays back faster

Custom AI automation pays back faster than Copilot in any scenario where a defined process repeats predictably across systems that are not all Microsoft, or where the goal is to remove human effort entirely rather than accelerate it. This is the workflow automation builds we deliver for businesses outside the Microsoft boundary most often, and the pattern repeats across industries.

Three scenarios where custom wins clean.

A 50-person ecommerce business processing 200+ customer support tickets per day across email, web chat, and a Zendesk inbox. Copilot cannot triage these, route them, or auto-resolve the simple ones, because the work lives outside Microsoft 365. A custom triage workflow handles classification, routing, and first-response for the 60 to 80 per cent of tickets that follow predictable patterns. Build cost £8k to £12k, payback inside six months.

A 100-person B2B services company processing 80+ inbound leads per week through web forms, email, and CRM. Copilot can help a salesperson draft individual replies. A custom workflow enriches each lead, scores fit and intent, drafts personalised follow-ups, and books qualified discovery calls automatically. The salesperson sees only the qualified meetings, not the noise.

A 40-person agency producing 25+ monthly client reports pulling from Google Analytics, Meta Ads, HubSpot, and project tracking tools. None of those data sources live in M365. A custom reporting workflow pulls the data, generates commentary, builds the deck, and delivers it on schedule. The account managers spend their time on strategy conversations, not data assembly.

In all three, the pattern is identical. Work crosses the M365 boundary, the volume justifies removing human effort entirely, the gain is operational rather than per-person.

The realistic stack most UK SMBs end up running

Three points:

• Above roughly 50 employees, the realistic answer is rarely Copilot or custom. It is both, used for the work each is shaped for.
• Copilot serves as the productivity layer for knowledge workers operating inside M365, typically rolled out to a subset of seats rather than the whole business.
• Custom AI automation handles the two or three highest-volume workflows that span systems Copilot cannot reach.

A typical hybrid setup at a 100-person UK SMB. Copilot deployed to 50 of 100 seats at around £14,800 per year, focused on the roles where the time saving justifies the spend (sales, finance, operations, leadership). Two custom AI workflows handling the highest-volume cross-system processes, at a combined £18k to £25k build cost and £400 to £700 per month in run cost. Total first-year AI spend around £35k to £45k, addressing both per-person productivity and structural operational drag.

This is the same shape of conclusion reached in the parallel comparison covering ChatGPT Enterprise vs custom agents, and the underlying logic is the same. Productivity tools serve the long tail of varied knowledge work. Custom builds serve the high-volume processes that drag operational cost. They are complementary investments, not competing ones.

The mistake most often made by SMBs at this size. Buying Copilot for everyone, declaring AI done, and being surprised six months later that the operational bottlenecks have not moved. Copilot was never going to move them.

How to decide before your next M365 renewal

Use this matrix to think through where your money should sit. Most SMBs we audit start with one or the other and add the second layer 6 to 12 months later as the evaluation sharpens.

The honest version. If your renewal quote is the only AI line item you are looking at, you are looking at the wrong page. Open it alongside a list of your three or four most painful cross-system processes, and ask which page would do more to reduce operational drag for the same money.