← All entries

Copilot Studio vs Microsoft 365 Copilot — licensing + credits

Praxisis D365 Wiki · Public reference · Client-specific details redacted

Source: Claude.ai chat (original: learning-about-copilot-studio__250e3333.md) Added: 2026-10-02

Summary

  • M365 Copilot is the finished product (you license, users use). Copilot Studio is the maker/builder tool for custom agents (standalone or as M365 Copilot extensions).
  • Convergence (Sept 2025+): Copilot Studio tools (agent flows, AI tools, deep reasoning) became included in the M365 Copilot license for many employee-facing scenarios. Role-based Copilots (Sales, Service, Finance) folded into M365 Copilot in Oct 2025.
  • Copilot Studio maker license is FREE — your whole dev team builds without seat costs. You pay for what agents DO, metered in Copilot Credits.

Key takeaways

  • Two pricing paths: (a) Capacity packs $200/month per pack = 25,000 Copilot Credits prepaid (~$0.008/credit); (b) Pay-as-you-go $0.01/credit, no commitment. Breakeven = 20,000 credits/month. Below → PAYG wins. Above → buy packs. Packs pool tenant-wide, can be stacked.
  • Zero-rating exemption: if users already hold M365 Copilot licenses, internal employee-facing agent interactions by those licensed users do NOT consume credits — classic answers, generative answers, grounding, actions, flows in Teams/SharePoint/Copilot Chat (subject to fair use). External-facing agents on websites, Power Pages portals, or WhatsApp ALWAYS consume credits regardless of licensing.
  • This is the single biggest cost lever: a staff-facing intranet assistant can be effectively free on an existing Copilot tenant. The same agent published to a Power Pages portal is fully metered.
  • Pack credits reset monthly with no carry-over. Annual pre-purchase credits expire at term end. Over-buying is a real loss, not a deferred cost.
  • Agent action rate card (rough spread): classic/scripted = 1 credit · generative answer = 2 · agent action = 5 · Microsoft Graph tenant grounding = 10 · voice = up to 75/min · premium AI tools = 100 per 10 interactions. Reasoning models bill twice (feature rate + premium tools charge for reasoning tokens) — most likely to blow past estimates.
  • Three exceptions: (1) new GitHub Copilot harness does not honour the licensed-user zero-rating — credits burn in both building and runtime regardless of license. (2) Computer Use (CUA) bills at agent-action rate but is not zero-rated for licensed users. (3) Downstream Power Automate / Dataverse / Azure costs are separate from credits.
  • Billing terminology changed from "messages" to Copilot Credits on 1 Sept 2025 — any older client deck / estimate uses the wrong unit. Copilot Studio moved to usage-based Credit billing on 3 August 2026; agents built before that date were grandfathered on the old model only until 1 Sept 2026 — this transition hit production right as the chat was written. Pull consumption reports after any client-tenant transition rather than discovering on the invoice.
  • Governance to set day 1: exceeding the purchased pool triggers enforcement that can disable agents — one team's experimental agent exhausting a shared pool shouldn't take down another team's production chatbot. Power Platform Admin Center supports monthly consumption limits per agent; enable PAYG alongside packs so overage bills through rather than cut off.

Learnings / how to apply

  • Lead every Copilot engagement with the internal-vs-external billing split — it's the single biggest cost lever and clients rarely realise it before signing.
  • Build a Praxisis "Copilot Credit Estimator" sheet modelling by interaction type (not headcount) — the rate card spread (1 → 100 credits per interaction) makes headcount models close to meaningless.
  • NGO / nonprofit pricing is a separate track worth checking — Copilot grant terms differ from the standard M365 nonprofit offers.
  • Verify against Microsoft's own pricing page + Credit Estimator before quoting — third-party licensing analysts disagree on specifics (e.g. premium AI tools "100 per 10 responses" vs "100 per 10 tokens" — wildly different cost models).
  • Pre-deployment tenant hygiene (oversharing, stale permissions, abandoned SharePoint sites) is where Copilot engagements actually land — surfaces as "Copilot showed me something I shouldn't have seen." That's a remediation project, not a licensing decision.

Need this applied to your own tenant?

Praxisis Sustain Self-Serve builds a private AI Wiki trained on your documents and tickets, with free Tier-2 support when the AI cannot answer. Book a 30-min overview.