Power Platform Community Forum Thread Details

I created an agent that answers questions about company metrics based off an Excel spreadsheet that I uploaded as knowledge. It's fairly straight forward with metrics and values by quarter. The metrics are hierarchical with sub metrics offset by one row and column. There are some queries where it just doesn't find the answer where it should be straight forward.

It will return the correct answer for one metric and then not find anything for another. Is there a way to correct bad answers? In my instructions I tell it what tab to use, the columns the metric names are in and what columns contain the quarterly results.

Sample table:

1	A	B	D	E	F	G
2	Metric		2023 Q4	2024 Q1	2024 Q2	2024 Q3
3	Metric 1		100	150	200	150
4		Metric 1.1	60	100	150	100
5		Metric 1.2	40	50	50	50
7	Metric 2		345	350	340	360

Categories:

General topics

Hello!

What I’d suggest you is to switch the data source. Usually when the model is searching for the row, it’d just randomly find the answer based on the prompt. To avoid this randomness you can use a Sharepoint List as a data source. How to achieve this?

1. Define a list which contains the same data structure as your Excel file.

If you’re creating a chatbot with generative orchestration enabled, then:

2. Define an action Get items from a Sharepoint connector.

3. Make the model create the correct ODATA query filter by manipulating the instructions (in the general tab).

4. Try to ask about the data and see how it performs.

If you’re creating a chatbot with a classic approach (topic based):

2. Define a topic with trigger phrases that you like.

3. In the topic, invoke Get items action based on the user prompt - you’d need to guide it by questions, for example.

You should be able to receive more predictable answers based on the filters.