Copilot Studio starts at $200 per tenant per month for 25,000 messages, with overage at $0.01 per standard message and $0.02 per generative message. Premium message types invoked by Dataverse, third-party connectors, and document grounding cost five to thirty times the standard rate. A bot that touches Dynamics 365 and a third-party CRM at 10,000 daily interactions can run $90,000 to $300,000 per month at list, not the $200 the starter pack suggests. The pricing is unforgiving on use cases that span line-of-business systems.
This page is the 2026 reference for what Copilot Studio actually costs, the message types and their multipliers, the capacity reservation lever, and the build-versus-buy decision against first-party Copilot SKUs. Sourced from the Copilot Studio licensing guide (May 2026 edition) and advisor-led Copilot Studio deployments during 2025 and 2026.
Message pack pricing
Copilot Studio is licensed by message consumption. The starter pack delivers 25,000 messages per tenant per month at $200. Additional capacity is purchased in additional packs or as overage at the per-message rate.
| Pack tier | Monthly list | Included messages | Effective rate |
|---|---|---|---|
| Starter (per tenant) | $200 | 25,000 | $0.008 per message |
| Additional pack | $200 | 25,000 | $0.008 per message |
| Standard overage | Variable | n/a | $0.01 per message |
| Generative overage | Variable | n/a | $0.02 per message |
| Capacity reservation (committed) | Negotiated | 1M+ monthly | $0.005 to $0.008 per message |
The starter pack is sufficient for a single low-volume internal use case. A 200-user departmental Q&A bot consuming 4 to 6 messages per user per day across 20 working days will run 16,000 to 24,000 messages per month. Two such bots exhaust the starter pack and push the tenant into overage.
Premium message types
The pricing dimension that surprises most enterprise deployments is the premium message category. Microsoft documents premium message types in the Copilot Studio licensing guide, but the categorisation is not surfaced in the public pricing summary or the Power Platform admin centre. The premium rate applies when the agent invokes a connector or operation that consumes additional compute, calls an external system, or grounds a generative response in a long document.
| Premium message type | Effective rate | Typical trigger |
|---|---|---|
| Dataverse read/write | $0.05 to $0.10 | Custom Dataverse table interactions |
| Third-party connector (premium tier) | $0.10 to $0.30 | Salesforce, ServiceNow, SAP, Oracle, custom REST |
| Document grounding (long context) | $0.05 to $0.20 | Retrieval over large documents or sites |
| Power Automate flow invocation | $0.05 to $0.15 | Triggering a Power Automate flow from the agent |
| External knowledge source (Bing, Tavily) | $0.05 to $0.15 | Web-grounded answers from external indexes |
The cost multiplier on a single bot interaction can be substantial. A user asks a question, the agent retrieves a customer record from Dynamics 365 (premium), generates a response with LLM call (generative), looks up an order in SAP (premium), and writes a follow-up note in a third-party CRM (premium). Total: one standard, one generative, two premium messages. Effective cost at list: $0.20 to $0.70 per interaction, not the $0.02 a back-of-envelope suggests.
The premium message blind spot: Procurement teams routinely build the Copilot Studio business case using the $0.01 standard rate uniformly. Real deployments that touch any line-of-business system run 30 to 60 percent of message volume at premium rates. The realised cost is five to fifteen times the back-of-envelope projection. Build the forecast at a 40 percent premium mix as the planning baseline, then verify against pilot telemetry before signing a capacity reservation.
Capacity reservation
For tenants with predictable monthly message volume above 1M, a capacity reservation reduces the effective per-message rate by 15 to 30 percent. The reservation is a committed monthly message volume at a fixed monthly fee, with overage at the standard rate. The lever is similar to an Azure Reserved Instance: trade flexibility for discount.
| Committed monthly volume | Typical effective rate | Reservation profile |
|---|---|---|
| 1M to 5M messages | $0.008 (20 percent off) | Mid-size enterprise rollouts |
| 5M to 25M messages | $0.007 (25 to 28 percent off) | Large enterprise, multi-department |
| 25M+ messages | $0.005 to $0.006 (30 to 40 percent off) | Strategic accounts, custom reservation |
The risk of capacity reservation is under-utilisation. Committed message volume that goes unused is wasted. Microsoft does not credit unused capacity. The defensible position for a procurement team is to size the reservation against the P50 (median) forecast, not the P90 (high) forecast, and to handle the variance via pack overage. Over-commitment is the most common Copilot Studio procurement error in 2026 reviews.
Build versus buy decision
The choice between building an agent in Copilot Studio, buying a first-party Microsoft Copilot SKU, or building outside the Microsoft 365 stack depends on three factors: where the data lives, how deep the Microsoft 365 integration needs to be, and the message volume per interaction.
| Use case archetype | Best fit | Reason |
|---|---|---|
| HR policy Q&A over SharePoint | Copilot Studio agent | SharePoint native, low premium mix |
| Sales call summarisation | Copilot for Sales ($50 SKU) | Pre-built, no message economics to manage |
| Customer service triage with CRM lookup | Copilot for Service ($50 SKU) | CRM native, supported channel |
| Cross-system order status (CRM + ERP + fulfillment) | Build outside (Azure OpenAI direct) | Premium mix makes Studio expensive at scale |
| External-facing customer agent | Dynamics 365 Customer Service + Copilot | Licensed for external users |
| Power Automate-led automation with conversational interface | Copilot Studio agent | Native Power Automate integration |
The cost-tipping use case is the one that touches multiple line-of-business systems at high volume. A customer service agent answering 5,000 daily questions, each requiring two premium messages, runs $1,000 per day at list. Annualised that is $365,000 for a single use case. A custom agent built outside Copilot Studio against Azure OpenAI directly, with custom connectors to the same systems, typically runs 40 to 70 percent less at scale because the premium-message multipliers do not apply.
Power Automate as an alternative
For deterministic, workflow-driven automation that does not require a conversational interface or natural-language interpretation, Power Automate is materially cheaper than Copilot Studio. Power Automate is licensed per user ($15) or per flow ($100), with no per-message metering. A flow that pulls data from Dynamics 365, formats it, and posts to Teams runs at the per-flow rate, not at the Studio premium-message rate.
The pattern most enterprises miss is that 30 to 50 percent of candidate Copilot Studio use cases are actually Power Automate use cases dressed up as agents. The conversational interface is not required. The deterministic workflow is. The decision should explicitly compare both options before committing to the Studio licensing model.
External users and customer-facing agents
Copilot Studio agents distributed to external users (customers, partners, prospective buyers) follow a different licensing model. The internal-message metering applies to employee interactions; external user interactions are governed by separate Power Pages or Power Apps external user licensing. The pricing is typically per-authenticated-user or per-anonymous-session, with material implications for high-volume customer-facing deployments.
The most common procurement error in this category is assuming external usage falls under the internal message pack. It does not. A customer service bot deployed to a public website that serves 50,000 anonymous sessions per month consumes external session licensing in addition to the underlying message consumption. The compounded cost at scale is significantly higher than the equivalent internal deployment. For external-facing high-volume use cases, Dynamics 365 Customer Service with Copilot is usually a more cost-effective licensing path than building in Copilot Studio.
Negotiation levers
Five levers move money in Copilot Studio negotiations. First, the capacity reservation, where committed monthly volume earns 15 to 30 percent off the message rate. Second, bundling with the broader Power Platform commit, which adds Power Apps and Power Automate to the reservation and earns additional discount. Third, multi-year price protection, which removes the annual 4 to 7 percent escalation from the message rates. Fourth, deployment funding via Microsoft's Power Platform Accelerator programmes, which can deliver $50,000 to $500,000 in deployment credits. Fifth, Q4 timing (Microsoft fiscal year ends 30 June), which delivers materially better terms in May and June.
The most consistent lever is the capacity reservation. The most missed lever is the Power Platform bundle. Most procurement teams negotiate Copilot Studio in isolation. Microsoft offers materially better unit economics when Studio, Power Apps, and Power Automate are bundled into a single commit, because the platform-level economics favour the bundled customer.
Governance and tenant-level admin cost
Copilot Studio inherits the Power Platform governance model. Every published agent is registered in the Power Platform admin centre, governed by environment policies, and subject to data loss prevention rules. The governance is functional but has three cost surfaces that are not immediately visible at procurement.
First, the Dataverse capacity required to store agent knowledge sources, conversation history, and operational data. The default Power Platform environment includes a baseline Dataverse capacity, but production agents at scale consume capacity beyond the baseline and trigger additional storage charges. Second, the connector licensing for any premium connector used by the agent, which is charged per user per connector at $10 to $40 per month depending on the connector. Third, the developer licensing if a Copilot Studio agent uses Power Fx logic or custom plugins, which can require Power Apps premium user licences for the agent author.
| Hidden cost | Typical magnitude | Trigger |
|---|---|---|
| Dataverse storage overage | $50 to $400 per GB per month | Agent conversation history retention beyond default capacity |
| Premium connector licensing | $10 to $40 per user per month | Agent uses Salesforce, ServiceNow, SAP, or custom REST connector |
| Power Apps premium for authors | $20 per user per month | Agent uses Power Fx logic or custom plugins |
| API call overage (Power Platform) | Variable | High-volume agents in low-tier environment |
The cumulative effect on a department-level Copilot Studio deployment can be material. A 500-user deployment that adds two premium connectors at $15 per user per month is $180,000 per year in connector licensing alone, on top of the message consumption cost.
Pilot-based forecasting
The defensible way to size a Copilot Studio reservation is a 60-day pilot at small scale (5 to 20 agents) with full telemetry capture from the Power Platform admin centre. The pilot data establishes the messages-per-interaction multiplier for each use case archetype, the premium-message mix for each archetype, and the daily volume curve.
The two pilot errors that destroy the forecast are: assuming the pilot user population represents the broader user base, and assuming the pilot premium mix represents the deployed premium mix. Pilot users tend to use the agent more carefully. Deployed users invoke premium-message-heavy operations more aggressively. Build a 30 to 50 percent uplift on the pilot baseline before committing to a reservation tier.
For the broader Copilot reference, see Microsoft 365 Copilot pricing 2026. For agent licensing specifically, see Copilot agents licensing. For the EA-level commercial framework, see Microsoft EA complete guide, Copilot licensing strategy, and the Microsoft vendor hub. For Copilot Studio procurement engagement, see our AI procurement advisory and software licensing advisory.