Predictive Intelligence and a Flow

[Sky4]

I'm using a classification definition to predict the Supporting Service of all incidents newly opened via email or self-service. I'm also using a Flow to get this prediction and assign the Supporting Service for the incident accordingly. When I test the Flow within Flow Designer, everything seems to work perfectly fine. However when I create a new incident, for instance via self-service, it appears that my model returns an error. I'm also using the exact same incident to do the test inside Flow Designer, so I'm really baffled. Here's what I see in the test results / flow context when I look at the outcome of the classification prediction action:

Test result:

Flow execusion context:

Can anyone tell me how it is that I get these two different outcomes, even though it's the exact same incident? I'd really appreciate any tips!

[nschneid]

Why is my flow able to find the ml solution when I manually run the flow with an already created incident but my flow cannot find the ml solution when the flow runs automatically via a newly created incident? 

I am in the exact same boat as Sky4 and unfortunately none the comments here are helpful.

[HeatherSN]

It is possible that this issue is related to ACLs being enforced due to Query String hardening initiatives.  
When a user without ML-related roles creates an incident, the following error might occur:
"Part of the query on ml_capability_definition_base has been ignored because of insufficient access for 'query_match' operation on ml_capability_definition_base.table."

Try adding ACLs granting the necessary access (for ML) to the users who are creating incidents to see if that addresses the issue. 
https://www.servicenow.com/docs/bundle/yokohama-intelligent-experiences/page/administer/predictive-i...