How can we calculate the actual cost per conversation for each Copilot Studio agent?

Copilot Studio does not currently expose one supported API or Dataverse billing table that maps every Conversation ID to Copilot Credits, tokens, tools, and currency cost; the closest supported capability is session-level credit visibility in Monitor for Microsoft Copilot harness agents, while tenant-wide exports remain aggregated by agent, environment, feature, and date.

What is available

The Power Platform admin center reports explicitly omit single-conversation/message usage and input, output, or cached token counts. Their detailed export contains Agent name, Product / AI feature, Channel, LLM model, Knowledge sources / Tool used, and billed/non-billed credits.^1^

Recommended dashboard design

1. Collect authoritative agent-level consumption

In the Power Platform admin center:

1. Go to Licensing → Copilot Studio → Environments.
2. Select an environment.
3. Open Copilot Credits capacity.
4. In Copilot credit consumption details, select Download.
5. Load the report into the dashboard and retain:
   • environment
   • agent name
   • product/AI feature
   • channel
   • LLM model
   • knowledge source/tool
   • billed and non-billed credits
   • reporting period

Use the Summary → Download report exports for historical environment-, agent-, and user-level trends. Built-in monitoring provides daily current-month and previous-month data, plus monthly data for the past 12 months.^1^

2. Collect session telemetry separately

For Microsoft Copilot harness agents:

1. Open the agent in Copilot Studio.
2. Select Monitor.
3. Set the reporting date range.
4. Review Sessions and Credits used.
5. Download session data before the retention window expires.

Monitor exposes individual sessions and their consumed credits, while transcripts can be downloaded for periods within the last 29 days.^2^

For agents whose transcripts are stored in Dataverse, export ConversationTranscript. It provides the conversation identifier, agent ID/name, timestamps, channel, and transcript content, but no documented Copilot Credit or currency-cost field.^3^

Important exceptions:

• Transcripts aren't written for Dataverse for Teams environments.
• The general Copilot Studio transcript export documentation also states that transcripts aren't written for Microsoft Copilot agents.
• Copilot Studio session storage is limited to approximately 28 days.^4^

For longer retention where Dataverse transcripts exist, Microsoft recommends exporting ConversationTranscript incrementally through Azure Synapse Link to Data Lake Storage Gen2 and using append-only storage or snapshots.^5^

3. Calculate monetary cost only where a valid rate exists

For pay-as-you-go, the current documented example rate is USD $0.01 per Copilot Credit.^6^

conversation_cost =
    billed_credits_for_conversation
    × contracted_payg_price_per_credit

For example, 141 billed credits at $0.01 per credit produces an attributed cost of $1.41.

Do not apply that formula as an “actual” cost for every licensing arrangement:

• Prepaid credits are pooled across the tenant.
• Microsoft 365 Copilot-licensed, authenticated employee usage can be non-billed under specified conditions.
• Pre-purchase discounts and negotiated pricing change the effective rate.
• Azure Cost Management shows actual pay-as-you-go charges by subscription/resource/meter, not Conversation ID.^7^

For prepaid capacity, label the result allocated cost, using an accounting rate such as:

effective_credit_rate =
    prepaid_period_cost / credits_available_or_consumed

allocated_conversation_cost =
    conversation_credits × effective_credit_rate

That is an internal allocation, not a Microsoft-provided conversation-level invoice amount.

Tokens, tools, and flows

Reasoning-model premium usage is billed at 10 Copilot Credits per 1,000 billed tokens. Therefore:

estimated_billed_tokens = premium_credits × 100

This is approximate when the same report line also includes prompt-tool usage. Cached input read from the prompt cache isn't billed; input written to cache, output, and reasoning tokens are billed.^8^

Tool and flow costs should be retained as separate feature rows. Current standard-harness rates include 5 credits per agent action and 13 credits per 100 agent-flow actions; AI-tool rates vary by model tier.^8^

API conclusion

There is no documented supported API that returns the requested complete record keyed by Conversation ID. The Microsoft 365 Copilot Cost Management API reports AI-credit usage but explicitly excludes prices, invoices, currency costs, and forecasts.^9^

Consequently, exact per-conversation cost is supported only where the applicable agent’s Monitor experience exposes final session credits. Otherwise, the defensible dashboard granularity is environment + agent + date + billing feature, with conversation metrics maintained as a separate telemetry dataset rather than joined as exact billing records.

References

1. Manage Copilot Credits and capacity for Copilot Studio
2. Monitor agent performance overview (Microsoft Copilot)
3. Understand downloaded conversation transcripts from Power Apps
4. Understand downloaded session data from Copilot Studio
5. Develop a custom analytics strategy
6. Optimize Copilot Credit costs with a pre-purchase plan
7. Usage-based billing: What you see in the Microsoft 365 admin center vs. your Azure bill
8. Billing rates and management
9. Understand usage-based billing and cost management for Copilot Credits