Skip to main content
Story solution recommendation

ElementsGPT, GPT, How might it be implemented?

Ksawery Lisinski avatar
Written by Ksawery Lisinski
Updated over a week ago

You can ask ElementsGPT to suggest how a story might be implemented in your Salesforce Org. The recommendation will list metadata that can be re-used and elements that need to be created to support the user story.

Article outline


How to ask for a solution recommendation?

To ask ElementsGPT to generate a solution recommendation, open a story - preferably generated with ElementsGPT.

Scroll down to 'How it might be implemented' field and click Ask Elements.

You can open a story from a diagram, from a metadata dictionary, or from the change grid.

How solution recommendation works

The diagram below captures the overall flow of how solution recommendation is generated: (click on the diagram to see a larger version)

  • First, the story's description and acceptance criteria are used to find metadata in the semantic database with names and descriptions that most closely match the provided string. Top results are brought back.

For instance, if you have a story that mentions 'I need to be able to capture customer problems...', the semantic search might bring back objects "Case", "Issue__c", "Feedback__c", "Incident__c", record types "Case.technical_support" etc. with their descriptions as possible matches.

  • The retrieved list of metadata, the descriptions, and the story's description and acceptance criteria are then merged into a proprietary prompt template. That prompt is then sent to the OpenAI API. (In the future we plan to enable the Salesforce AI Cloud and/or other providers.)

  • The AI model interprets the data based on the prompt command and determines if metadata from the provided list can be re-used to support the story. For any missing metadata, it generates a recommendation to create new elements based on the provided instructions.

  • The response consisting of the list of metadata suggested for implementation with a step-by-step process for creating them is then pasted onto the story. You can accept the recommendation, amend the details, or delete it if you don't find it satisfactory.

Ways to improve the quality of the suggested recommendations:

  • improve story definition, directly or with enhanced process detail

  • improve acceptance criteria on the story

  • enhance Salesforce metadata descriptions using Elements (description field)


There are three main limitations in the current pilot:

  • Solution recommendation is focused on listing metadata required to support the story but it does not offer specific recommendations for how to build automation. ElementsGPT may recommend building a record triggered flow but it will not specify how it is supposed to work.

  • In the pilot phase, ElementsGPT has shown itself to be very prone to recommending apex triggers and classes for automation when flows or other low-code automation would suffice. We are working on making it follow the well-architected framework to improve the recommendations.

  • When you are creating a new story, the 'Ask Elements' button will not appear. This is because until you save the story, it hasn't yet been committed to our database. Once a story is created, it can be re-opened to generate a recommendation.

Did this answer your question?