Microsoft 365 Copilot
The deepest enterprise AI stack you can buy into your existing Office estate — but licensed seats are not used seats, and the ROI math only clears if you do the governance and adoption work Microsoft won't do for you.
Element scores
Strengths
Covers more of the periodic table than any workspace rival: Work IQ context over your whole tenant, multi-model chat with GPT-5.x and Claude Sonnet 5, Researcher/Analyst reasoning agents, Agent Mode default in Word/Excel/PowerPoint, Copilot Studio as a real harness with evals, computer use and MCP, and governance (Purview DLP, Agent 365, audit) nobody else matches. Seat growth is real: 20M+ paid seats, +250% YoY adds, Accenture at 740k seats.
Honest dings
The adoption gap is the story: independent data puts weekly active usage at 20-30% of purchased seats, and when employees can also pick ChatGPT or Gemini, Copilot's voluntary preference share drops to single digits (Recon Analytics). Output quality inherits your tenant's mess — stale SharePoint gets cited with confidence, over-permissioned content gets amplified. Pricing is a maze (add-on + bundles + Studio + consumption credits on top of $30 seats), and Gartner found only 5% of pilot orgs moved to larger deployments.
Sources
Every audit lists the research it rests on — transparency and traceability are the product. Tools evolve: each audit is a snapshot of its audit date, and re-audits supersede older versions (kept below for reference).
- microsoft.com/en-us/investor/events/fy-2026/earning… — Official Q3 FY2026 earnings (29/04/2026): 20M+ paid seats, +250% YoY seat adds, Agent Mode default, Cowork, Critique & Council, Copilot Credits ~2x QoQ, ~90% of Fortune 500 building agents (accessed 2026-07-06)
- microsoft.com/en-us/investor/events/fy-2026/earning… — Official Q1 FY2026 earnings: 150M MAU across the Copilot family, 900M MAU of AI features, Agent Framework (accessed 2026-07-06)
- techcommunity.microsoft.com/blog/microsoft365copilo… — Official: federated Copilot connectors via MCP GA (May 2026), custom federated connectors, watermarking (accessed 2026-07-06)
- microsoft.com/en-us/microsoft-copilot/blog/copilot-… — Official Copilot Studio April 2026: MCP server tools in workflows, Agent 365 integration, DLP-governed workflow environments, cost visibility (accessed 2026-07-06)
- support.microsoft.com/en-us/microsoft-365-copilot/g… — Official: Researcher reasoning agent, Claude models in Researcher (phased rollout to end March 2026), licensing scope (accessed 2026-07-06)
- adoption.microsoft.com/en-us/ai-agents/copilot-stud… — Official: Claude Sonnet 5 in M365 Copilot (02/07/2026), Copilot Studio evals, computer use, GPT-5 family, voice agents (accessed 2026-07-06)
- velosio.com/blog/m365-copilot-pricing-calculator — Independent pricing verification (June 2026): $30 add-on, Copilot Business $18 promo/$21 standard, bundles $23.50/$32 permanent from 01/07/2026, July 2026 base-plan increases (accessed 2026-07-06)
- neuralwired.com/2026/06/02/microsoft-copilot-review… — Independent review: 35.8% active of provisioned seats, 8% voluntary preference vs ChatGPT/Gemini, Gartner 5% pilot-to-scale, Forrester TEI caveats (accessed 2026-07-06)
- perspectives.plus/p/microsoft-ai-numbers-good-bad-u… — Independent analysis (13/02/2026): 15M seats = 3.3% of 450M commercial base at Q2 FY26, Recon Analytics retention data, metrics Microsoft stopped reporting (accessed 2026-07-06)
- valueaddvc.com/blog/microsoft-copilot-enterprise-ad… — Independent (19/06/2026): 20-30% weekly usage of purchased seats, effective cost per active user math, adoption blockers (accessed 2026-07-06)
- gestisoft.com/en/blog/copilot-ai-review — Independent deployment review (18/06/2026): confident stale-content citations, adoption plateau ~30-40%, environment-readiness as the ROI driver (accessed 2026-07-06)