At the time of this writing, AI Prompts are in preview. If you want to play around with AI Prompts, you’ll need to go to make.preview.powerapps.com to be able to use them in your environment. If you want to find out more about the Prompt column, you can go here.
The prompt column is a new AI-driven column type in PowerApps that lets you pose queries to PowerApps and let AI do the heavy lifting to determine things such as sentiment and categorization.
Creating an AI prompt field is as easy as clicking New column and then selecting the prompt type. One thing I hope that gets changed in the GA release is to allow the prompts to exist on multi-line text formats; right now, we are limited to single-line inputs.
When you click “Edit”, you are brought to the following page, where you can work with your prompt and merge it with Dataverse data by clicking “Add Content” and/or using the “/” character to open the field chooser window. The field chooser window is quite extensive, it could be a big bigger and it would be great to have a search field at the top as we all have many, many, many fields and relations with sometimes lengthy names. The model selection has an interesting twist where you can actually direct it to your own Azure AI Foundry model – I have yet to do this, but it’s a great way to bring your own model into the system.
I love the test prompt feature, but I am still trying to get the hang of it. Right now, I think my prompts work well in this mode, but don’t translate to the individual records. I.e., I like the idea of being able to understand ranks between my data. The above is great, but when I accessed the record, the results would evaluate against that specific record only. It could be my limited understanding as well 🙂
With that in mind, I changed up my prompt to focus on deriving data solely from the record I was on, and I intentionally left it pretty wide open because I wanted to see what the system would come up with.
When rendered to the screen (now you see why I want a multi-text field), you can see the results of my prompt executed for this particular account.
Fully expanded, the text reads as below.
Based on the provided data for the account “City Power & Light (sample),” here is an overall sentiment assessment considering the key factors you mentioned:
1. Number of Cases (Account.Customer): The data does not explicitly provide the number of cases associated with this account, so this factor cannot be directly evaluated.2. Actual Revenue from Opportunities (Account.Potential Customer (Opportunity).Actual Revenue): The account shows a revenue figure of 75,000,000.0 (currency in Canadian Dollar), which indicates a substantial revenue generation. However, the “openrevenue” and “openrevenue_base” values are 0.0, and “opendeals” is 0, suggesting no current open deals or pending revenue.
3. Posts or Notes on the Account: There is no explicit data on the number of posts or notes related to this account in the provided information.
4. Account Status and Activity: The account status is “Active,” with no credit on hold, and marketing communications are allowed. The account is not marked as private, and contact methods are open, indicating good engagement potential.
5. Other Indicators: The account has 600 employees, which suggests a medium to large-sized customer. The account is not merged, and there are no restrictions on communication channels.
Sentiment Summary: The account “City Power & Light (sample)” appears to be a well-established and active customer with significant historical revenue (75 million CAD). However, the lack of current open deals or pending revenue and no data on cases or engagement posts/notes limits the assessment of ongoing activity. Overall, the sentiment is cautiously positive, reflecting a strong revenue base and active status but with limited visible current engagement or pipeline activity.
Actionable Suggestion:To improve the sentiment and engagement assessment, it would be beneficial to gather data on the number of customer cases, current opportunities, and account-related posts or notes. This would provide a more comprehensive view of the account’s health and activity.
Pretty detailed to say the least, and yes in a real-world example, I would definitely shrink and set the sentiment parameters, but I wanted to see what the system could come up with.
I guess the question is – if I’m already evaluating the above information via Power Automate AI Builder, do I need these prompt columns? And/Or do they exist for those that might not be leveraging Power Automate, in which case it’s duplication?