Issue 008: Accessible AI — From Hype to Everyday Use

Microsoft has sold 15 million paid Copilot licences. According to Recon Analytics, barely a third of those licensees are actively using the tool. In practical terms that means in the average Copilot-enabled office, two people out of three have access to AI that they are simply not opening. They are not technophobes. They are not refusing to use it. The tool is embedded in their workflow, but it has never made it into their habits. The gap between those two sentences is the most expensive blind spot in enterprise AI today.

Most of what gets written about AI concerns the extraordinary. The mind-blowing demo. The model that passed the bar exam. The agent that booked a holiday. All of that is real, and none of it explains why your finance director still exports a pivot table to PDF the long way around. The real productivity gains of AI, the ones that compound on a quarterly basis rather than announce themselves at a keynote, come from AI that has become so ordinary that it has stopped being noticed. That is the Accessibility Dividend.

The Friction Test. Does the tool live where I already work, or do I have to open something new to use it? A tool that requires a context switch will be forgotten by 11am on a Tuesday. The best AI today is embedded in Outlook, Teams, the browser tab you already had open, the spreadsheet you already had live.

The Language Test. Can I get value from the tool in plain English, without learning a new syntax or protocol? If a tool needs its own training course before it is useful, it is failing the accessibility test. An eight year old can get useful output from a mainstream AI model. That is the bar.

The Trust Test. Are the outputs good enough that I will edit them, rather than rewrite from scratch? The difference between a 70 per cent draft and a 40 per cent draft is not 30 percentage points. It is the difference between a tool I open tomorrow and a tool I never open again.

The Gallup data behind that callout is worth dwelling on. Twenty-three thousand US employees, surveyed in February. Only about one in ten in AI-adopting firms strongly agree that AI has fundamentally changed how work gets done. Individuals are getting faster. Organisations are not being rewired. This is not because AI does not work. It is because giving people access to AI, and redesigning the work around AI, are two different projects, and most firms have only completed the first. The deeper question of what happens to the people on the wrong side of that gap is the subject of a special report I am publishing on Thursday.

If you are deciding where to invest next in your AI programme, the interesting question is not which model is best, or which vendor is ahead. It is this: what does your team actually open at nine o'clock on a Tuesday morning? The AI that wins the next twelve months is the AI that is already there when they do.

Microsoft Copilot for Microsoft 365 is the productivity layer of Microsoft's Copilot family, embedded inside Word, Excel, PowerPoint, Outlook and Teams. Context for regular readers: this review is distinct from Issue 002's coverage of Microsoft Copilot Cowork, which is the agentic background-task variant. This is the AI in your inbox, your meetings, your slide deck and your spreadsheet. Different product, different review.

The strongest features are exactly the ones an average professional uses every day. Meeting recaps in Teams produce accurate minutes and action items without human note-taking. Outlook summarises long email threads into a few sentences and drafts context-aware replies. Word turns a list of bullets into a first-draft policy update or client letter. PowerPoint generates a full slide outline from a document. Excel formula and chart generation handles the moments where most users lose ten minutes. Microsoft added Agent mode across Word, Excel and PowerPoint through Q1 and Q2 2026, shifting Copilot from single-prompt answers to multi-step work on a file. The Researcher feature now exports to PowerPoint, PDF, infographic or audio in a single click.

The most damning figure in this review is not a feature gap. It is the 35.8 per cent workplace conversion rate in Story 2. Microsoft has sold the licences. Many users do not open the tool. Hallucinations still appear in meeting transcripts, particularly around who said what. Discoverability inside each app is inconsistent. At a standard licence of £24.70 or $30 per user per month, the tool pays for itself quickly for heavy document and email users, and slowly or not at all for occasional users. The July 2026 pricing update is likely to tighten that calculation further.

Try Microsoft Copilot for Microsoft 365 →

The numbers this week tell a consistent story. 54 per cent of UK firms now using AI. 15 million paid Copilot seats. Half of US adults using AI at work. Adoption has arrived. What has not arrived, at least not yet, is transformation. Only one in ten workers in AI-adopting firms say their workplace has been fundamentally changed by it. Individuals are being rewired. Organisations largely are not.

The Accessibility Dividend does not pay itself out automatically. It requires decisions that most organisations have not yet made. Decisions about which tasks get automated and which stay human. Decisions about what to do with the hour a day workers are now saving. Decisions about whether that saved time goes into more volume, or into the work that was previously crowded out. Those are leadership decisions, and the data suggests very few leaders are making them.

A prediction for the next twelve months, and one you can hold me to. By the end of 2026, two divides will matter more than any model benchmark or vendor comparison: who has access to AI at work, and who has been supported to actually use it. Thursday's special report, AI, Your Job and the Next Five Years, goes deeper on both. If this issue gave you the framework, Thursday's report gives you the evidence and the action toolkit. Watch your inbox.