Regarding Fallback Mechanism in Copilot Studio

(0) Share

Report

Posted on by PG-13041328-0

I have an orchestrator agent and 5 sub agents attached to in copilot studio . I want that whenever an user query is asked , it should trigger the best matched agent and if not found then trigger the MCP attached to only orchestrator agent and if MCP cant find answer , send a message to teams channel with the user query .

My agents are currently only triggered by their trigger description , not by topics of orchestrator agent .

How i can implement this kind of fallback mechanism ?

Also in most of the sub agents I have topics which generates KQL query from the user query and it runs that query inside Azure Monitor Logs(run query and list results connector) .

SO sometimes the KQL query generation(Currently handled using Customize response node available in topics) is incorrect so the connector fails because the input (KQL query) is wrong.

How to improve this also ??

Categories:

Calling actions from Copilot Studio

I have the same question (0)

All responses (3)

Answers (0)

Sort by

Suggested answer

11manish 2,376 on at

Like (0)

Report
Copy link

Link copied!
Yes — this kind of fallback mechanism is possible in Microsoft Copilot Studio, but it is better implemented through a centralized orchestration pattern rather than relying only on sub-agent trigger descriptions.

My recommendation is:

treat the orchestrator as the “decision engine”

treat sub-agents as specialized tools

implement explicit fallback and retry logic

avoid relying only on trigger descriptions for routing

That architecture is much more stable, scalable, and production-ready for enterprise Copilot Studio implementations.

Was this reply helpful? Yes No
Nivedipa-MSFT Microsoft Employee on at

Like (0)

Report
Copy link

Link copied!
Hello PG-13041328-0 ,
1. Fallback path (sub-agent → MCP → Teams)

Improve routing first: make each sub-agent's description intent-specific with 3–5 example utterances, and keep them non-overlapping.

Make "no answer" handling explicit: tell every sub-agent and the MCP step "If you can't answer, reply exactly NO_ANSWER."

Add a clear fallback path in the orchestrator:

Run the sub-agents first (generative orchestration).

If the response is empty / NO_ANSWER → call MCP.

If MCP also fails → trigger a Power Automate flow → Post message in Teams channel with the user query + user info.

Reply to user: "I've forwarded this to the team."

2. Reliable KQL generation

Swap out Customize Response for a Prompt action (Power Platform prompts):

Include the table/column schema + few-shot examples (question → KQL).

Instruction: "Output ONLY valid KQL, no markdown."

Validate before execution: strip code fences, check that it starts with an allowed table, and enforce | take N plus a default time filter.

Retry on connector error: send the failed KQL + error message back to the prompt and ask it to correct it, retrying 1–2x.

On final failure → reuse the same Teams fallback flow.

If you found the information above helpful, I would appreciate it if you could share your feedback.
Your feedback is important to us. Please rate us:
🤩 Excellent 🙂 Good 😐 Average 🙁 Needs Improvement 😠 Poor

Was this reply helpful? Yes No
PG-13041328-0 22 on at

Like (0)

Report
Copy link

Link copied!
Hii @Nivedipa-MSFT , thanks for your response .

I have some questions :

Make "no answer" handling explicit: tell every sub-agent and the MCP step "If you can't answer, reply exactly NO_ANSWER." - this type of lines i should add in all attached sub-agents instructions or somewhere else ??

Regarding Add a clear fallback path point :

Can you explain how to do this in detail : via orchestrator agent instructions or modifying the system fallback topic in orchestrator or creating a custom fallback topic in orchestrator.

2. Reliable KQL Generation :
I really liked your answer , thanks very much for answering

Was this reply helpful? Yes No