Skip to main content

Hi there

I’m testing Copilot to understand how it behaves and where it gets data from and just had an interesting experience with a very simple prompt: 

“Was [Company X] ever a customer?” 

It told me it couldn’t answer because “no timeline data”. 

When I asked it if there are any customers containing the name [Company X], it gave me a list of such customers with their status. When I asked it if we have any active customers by the name of [Company X], it listed the corresponding customers. 

It looks like the first prompt called the Timeline Agent, the second and third ones called the Company Agent. I suppose the word “ever” awoke the Timeline Agent because of the notion of time in that word.

My question is: why was a split model implemented (one object = one agent)? I’m not certain I see the point in this. I am certain however is our data is such (and I think many other customers’ data) that looking at both objects would have produced a better answer: “It looks like you have active customers by the name of [Company X] but the lack of interaction suggests this account may be stale”.

Keen on understanding this further. 

cc: ​@shantan_reddy 

Thanks

A

 

​@alizee While the agents are 2, the copilot does use both of them for a single query when needed. We have separate services since structured and unstructured querying involve different tech architectures.
 

Your insight is correct - the word “ever” likely threw it off. It should have looked at both sources ideally for this query. 
In the future, we’ll give admins provisions to configure these behaviors and should tackle this and more such issues.

 

Hope this answers your question.

 

Regards,

Shantan

Hi ​@shantan_reddy 

Thanks for taking the time to respond to this question so quickly. I’ll be exploring the tool further on that basis to form some best practices / library of prompts users would benefit from to get unified responses.

A

 

Terms & ConditionsCookie settingsAccessibility statement