Microsoft Copilot ROI: the math that doesn't survive 12 months.

Microsoft's Copilot for M365 priced at $30/user/month is a deceptively simple-looking line item. The Forrester TEI report (commissioned by Microsoft) projects 286% ROI. Most real deployments measure something between -10% and +25% in year one. The gap isn't accounting fraud — it's that the productivity gains accrue unevenly across roles, and the per-user pricing rolls all of them into one bill.

The real math

Microsoft pitches Copilot as a productivity multiplier on knowledge work. The implicit math: each user saves N hours/week × fully-loaded hourly rate × 52 weeks > $360/year. At $100/hr loaded cost, you need 3.6 saved hours per year per user to break even. That's 4 minutes a week. Should be a no-brainer.

Reality: usage distribution is power-law, not uniform. Microsoft's own Work Trend Index shows ~20-25% of seats become "heavy users" (≥5 sessions/week), ~30-40% are occasional (1-3/month), and ~30-50% effectively don't use it after the first 30 days. The heavy 22% capture 70%+ of the productivity benefit; the long tail of seats pay $30/month and barely log in.

Per-role math: a financial analyst writing 4 board memos per quarter saves 4-8 hours each — easy ROI. An accounts-payable clerk processing invoices in a workflow Copilot doesn't speed up sees zero benefit. Same $30/month bill.

3 roles where Copilot consistently pays back

1. Senior analysts / consultants who write deliverables. First-draft generation of memos, decks, exec summaries. Time savings: 30-50% on document creation. Real payback in month 1.
2. Engineering managers running 1:1s + perf reviews. Meeting recap, action-item extraction, perf-review draft generation. Time savings: 1-2 hours/week. Payback by month 2.
3. Sales engineers + solutions architects. Email drafting, RFP response, technical-doc generation against M365 corpus. Time savings: 3-5 hours/week. Payback in weeks.

Outside those roles, the math gets murky fast. For high-volume operational roles (AP, support, ops), Copilot for M365 is rarely the right tool — purpose-built agents (Copilot Studio bots, third-party automations) usually win.

The licensing traps

1. The Microsoft 365 E3/E5 prerequisite. Copilot requires an underlying E3 or E5 license. If you're on Business Premium, you can't add Copilot — you have to migrate, often at +$10/user/month before the Copilot $30 lands. Real all-in cost: $40-50/user/month, not $30.
2. Annual commitment, monthly publishing. The list price is monthly but contracts typically commit annually. You can't downgrade 30% of seats in month 4 when you realize they don't use it.
3. Pay-per-use Copilot Studio adds. Once teams start building custom Copilot agents (which Microsoft pushes), Message-Unit costs stack on top of seat licensing. 1,000 active Studio bots × moderate use ≈ $40-100K/yr extra.
4. Data-readiness prerequisite. Copilot is only as good as the M365 corpus it reads. Most orgs have SharePoint sprawl + permissions chaos. Implementations that skip data hygiene return garbage answers and adoption tanks. Add $50-200K in remediation work that wasn't in the original ROI calc.

How we'd decide

Three honest conditions where Copilot for M365 pays back:

• You can identify a target population of 20-30% of seats that are heavy-document-writers, and you license only those seats (per-user assignment is allowed). $360K → $80-100K.
• Your M365 data hygiene is already strong (mature SharePoint governance, clear permissions, working metadata). Adoption needs ~3 months of clean data to be meaningful.
• You measure usage by role for 90 days before scaling. Microsoft Adoption Score + 3rd-party telemetry give you the seat-level reality before you commit to year-2 renewal.

The mistake we see weekly: rolling out Copilot org-wide because the CEO read about it in WSJ, then quietly de-licensing 70% of seats 9 months later. The pattern wastes $200-300K and trains the org that AI tools are overhyped — making the next, better tool a harder sell.