Can NowLLM Analyse Multiple Records to Decide Incident Priority?

[Amitoj Wadhera]

How can we provide multiple records—such as a list of past incidents—as input to a Now Assist skill set, so that the content generated is based on analysing those records? For instance, instead of setting the priority of an incident based only on its short description or description field, I want the LLM to study previous incidents and determine the priority accordingly for the newly created ones. Is this possible? 

[Dr Atul G- LNG · Accepted solution]

Hi @Amitoj Wadhera 

I'm not sure about Now LLM, but I have seen something similar in PI. I can say that Now LLM has used machine learning (ML) and PI in the backend to some extent. This concept is called similarity and classification. Have a look from PI side and see what we can do in Now LLM. 

[Dr Atul G- LNG · Accepted solution]

Hi @Amitoj Wadhera 

I'm not sure about Now LLM, but I have seen something similar in PI. I can say that Now LLM has used machine learning (ML) and PI in the backend to some extent. This concept is called similarity and classification. Have a look from PI side and see what we can do in Now LLM. 

[PaulSylo · Accepted solution]

Hi @Amitoj Wadhera - Now assist is to provide deterministic output . this can be well complete using predictive intelligence as mentioned by Atul below. I see this can be done very well using Predictive intelligence, you can use Now assist  along with AI agents ( with proper tuning)for your use case). i don't think you get value if you do this try with now assist, it will be expensive soluiton

[Jayden4]

Just a note that if you are using Predictive Intelligence, and you do not have any instances of a P1, the Predictive Intelligence will never predict a P1, as it does not have any reference data in the trained model for what a P1 looks like. 

For what you are asking for in particular, you will need to retrieve the incident records with a RAG or subflow to parse into the prompt, but there is a cap of 15,000 words/tokens per prompt I think? I have just received the same sort of request.