Problem with triggers in Copilot Studio

Question

Hi,

I built an agent in Copilot Studio that is supposed to trigger when an email is received and then create a draft reply. My problem is with the triggers—they seem to work randomly. Sometimes they work, and other times they don’t.

When I check the Power Automate flow for a specific trigger, it looks like it executed successfully, but Copilot Studio didn’t receive any message. From the Copilot Studio side, it appears that nothing happened, even though Power Automate shows that everything ran correctly.

I’ve tried republishing, but that only seems to work sometimes. After one or two responses, it breaks again and stops responding to incoming emails. Power Automate still shows no errors and indicates that it executed properly.

Could someone please help me understand what might be causing this issue? Aren’t triggers supposed to work deterministically? I would like to rely on them working for every incoming email.

Thanks for reading, and have a good day.

Answer 1

Hello Matej Jakubčík,
Power Automate flows can appear “successful” because they often only confirm that a request was accepted (e.g., HTTP 202) rather than fully processed by Copilot Studio, leading to scenarios where the agent never executes despite green flow runs. The primary root cause is that Copilot Studio event triggers are still in preview in many environments, making them non-deterministic and unreliable for production use—often working intermittently and failing silently. Additional contributing factors include connection or identity mismatches, environment misalignment between the flow and agent, throttling under load, and Outlook trigger batching behavior. To achieve reliability, the recommended architecture is to decouple ingestion from processing using a durable queue (e.g., Dataverse or Service Bus) and a worker pattern with retries, ensuring no events are lost. If event triggers must be used, they should be hardened with retries, logging, strict environment alignment, payload control, and burst handling. Ultimately, production-grade solutions should avoid relying on preview-trigger pipelines and instead adopt deterministic, retry-safe designs.

also give the below details:

  1. Is this agent running in **Teams** or just **Copilot Studio / web**?  
  2. Are you using **Copilot Studio event triggers** or are you calling the agent from a normal flow?  
  3. Is the mailbox **shared** or **user mailbox**, and do you see bursts (many emails in short time)?