Direct answers to
the most common questions.

Microsoft 365 Copilot is priced at $30 per user per month (around £24.70 GBP) on top of an eligible Microsoft 365 subscription, with an annual commitment. Eligible plans include Microsoft 365 E3, E5, Business Standard, and Business Premium. There is no per-user setup fee, but the licence is typically sold in annual blocks rather than monthly subscriptions.

As of 2024, organisations under 300 employees can buy Copilot without minimum-seat requirements. Microsoft Copilot Chat — the included version that comes with paid commercial M365 plans — has no additional cost and offers enterprise data protection, file upload, image generation, and Copilot Pages.

Copilot Chat is the free, web-grounded chat experience included with paid commercial Microsoft 365 plans (Business Standard, Business Premium, Enterprise E3/E5). It offers enterprise data protection, file upload, image generation, Copilot Pages, and integration in Teams and Outlook as a side panel.

Microsoft 365 Copilot is the paid add-on (~£24.70/user/month) that adds in-app Copilot inside Word, Excel, PowerPoint, OneNote, Loop, and Whiteboard, plus Graph-grounded chat that can search across your tenant's email, files, and chats with appropriate permissions.

The simplest summary: Copilot Chat is the AI you already have. Microsoft 365 Copilot is what you can add.

Microsoft's own WorkLab research reports productivity gains for individuals using Copilot — typically around 20–30% time savings on tasks like drafting, summarising, and meeting recaps, with senior knowledge workers reporting the largest gains. However, these figures come from Microsoft-commissioned studies and apply to individual task time, not bottom-line ROI.

Independent ROI measurement remains scarce because outcomes depend heavily on the level of organisational adoption (most organisations report 20–40% active usage in year one), the workflows targeted, and whether time saved is reinvested productively.

The honest answer: ROI is achievable but not automatic. It depends on adoption discipline, not the licence itself.

Technical deployment is fast — Copilot licences can be activated in hours. Meaningful adoption is the slower work. A typical pattern:

Weeks 1–2: provision licences and run initial training. Weeks 3–8: embed Copilot in regular workflows with structured exercises and prompt practice. Months 3–6: usage reaches steady-state across departments, with role-specific use cases driving the most productive applications.

The biggest determinant of timeline is not technical setup but how seriously the organisation treats change management. The common mistake is treating Copilot as an IT project; it is an organisational capability project.

Microsoft 365 Copilot inherits the existing access controls in your tenant — meaning it can surface anything an individual user already has permission to see. The most common pre-deployment governance work covers:

SharePoint and OneDrive access permissions audit (oversharing is the most-cited issue); sensitivity labels and DLP policies; retention policy review; conditional access policy verification; and external-sharing audit.

Microsoft provides tools (SharePoint Advanced Management, Microsoft Purview) but the work is organisational rather than technical. A typical pre-Copilot governance review for a mid-market organisation takes 4–8 weeks. The risk of skipping it is well-documented: Copilot surfaces information users theoretically had access to but never knew existed.

For Copilot Chat (the free version): any paid commercial Microsoft 365 plan — Business Basic, Business Standard, Business Premium, or Enterprise E3/E5 — gives access.

For Microsoft 365 Copilot (the paid add-on): the prerequisite plans are Microsoft 365 E3, E5, Business Standard, or Business Premium (and the Education A3/A5 SKUs for faculty). The add-on is sold via direct, partner, or CSP channels at $30/user/month. Frontline worker plans (F1, F3) are eligible for a separate, more limited Copilot SKU.

Microsoft has progressively removed minimum-seat requirements: small organisations under 300 users can now buy single-seat Copilot.

AI readiness is the state of organisational preparation that determines whether an AI investment will deliver value or stall. The Requisite Intelligence AI Readiness Diagnostic assesses five dimensions:

Leadership Clarity (does the board have a coherent position?); Data Foundation (is the data estate ready?); Workforce Capability (can the organisation absorb the change?); Governance Posture (are risk and compliance frameworks in place?); and Use-Case Discipline (have the right problems been identified and prioritised?).

The output is a board-ready picture of readiness, exposure, and the practical sequence of work that follows. AI readiness is not a precondition you achieve before adoption begins — it develops through use, but it can stall or reverse without the leadership work.

AI ownership is one of the most common stalling points in AI adoption. The wrong answers, in our experience: "IT owns it" (turns AI into a technology project, not a business one) or "everyone owns it" (no one is accountable).

The pattern that works: a named senior owner at executive level — sometimes a CIO/CTO, sometimes a COO, sometimes a CFO depending on the organisation — with explicit board-level mandate, supported by a small cross-functional working group covering operations, risk, legal, HR, and IT.

For mid-sized organisations without the in-house capacity to carry that ownership, a fractional Virtual Chief AI Officer provides the senior accountability without the cost of a full-time appointment.

Effective AI governance is less a static policy document and more a set of operating disciplines. The common elements:

A published responsible AI policy aligned to your sector regulations; clear ownership of AI outputs (a named human is accountable for every AI-assisted decision in a regulated process); validation routines for AI-generated content used externally; data classification and sensitivity labelling so AI tools respect appropriate access; an approved-tools list and a process for adding tools to it; an AI incident register; and regular board reporting on AI usage, outcomes, and exposure.

Governance introduced too late creates anxiety. Introduced too early, it stifles experimentation. The discipline is to introduce just enough governance to make experimentation legitimate.

In the research underpinning our practice (18 UK SMEs interviewed in 2025–2026), the most common failure pattern was not technical: it was treating AI adoption as a technology decision when it was a leadership and meaning-making one.

The organisations that stalled tended to deploy AI tools, expect adoption to follow, and then be surprised when usage remained low. The organisations that progressed treated adoption as an interpretive process — leaders made meaning of AI for their teams, modelled use from the top, set provisional boundaries that evolved with learning, and treated unevenness as normal rather than as failure.

The second most common mistake was over-strategising: writing a three-year AI strategy before learning enough through experimentation to know what the strategy should be.

Adoption can be measured at three levels:

Level 1 — activation: how many licensed users have signed in and used the tool. Microsoft 365 Copilot Admin Center and the Anthropic Console provide these metrics natively.
Level 2 — engagement: frequency of use, depth of use (which features), and breadth of use (which departments). This is where most adoption programmes plateau.
Level 3 — outcomes: time saved, output quality, or business metrics tied to specific use cases. Outcome measurement is hardest and most valuable.

The common failure is to measure only Level 1 (high activation feels like success) without Level 3 (no business impact).

Agentic AI refers to AI systems that can take actions on a user's behalf — not just answer questions, but plan multi-step tasks, call tools and APIs, navigate software interfaces, and complete work autonomously.

Microsoft offers Copilot Studio for building business agents; Anthropic offers Claude with computer use and tool calling; Google offers Gemini-based agents through Vertex AI.

Maturity varies. Limited agentic use cases are production-ready today: structured data extraction, scheduling, ticket triage, document classification. Complex agentic workflows remain experimental. The risk profile is also different: an agent that takes actions can make consequential mistakes faster than a chat assistant can.

Vendor selection should follow use case definition, not vice versa. The process we use:

Identify three to five priority use cases (where AI would add measurable value); for each, identify the assistant capabilities required (long-context reasoning, in-app integration, agentic action, data sovereignty); evaluate the candidate vendors against those capabilities, not against marketing material; review commercial terms (especially data handling, retention, training opt-out); pilot one or two vendors against the priority use cases before committing.

The common failure is to select a vendor on brand or relationship and then look for use cases to justify the spend. Vendor-neutral evaluation matters: Requisite Intelligence does not take referral fees or carry reseller margin on any AI vendor.