Summarization and Escalation of Zendesk Tickets with ChatGPT

Make sure you have all of your tools lined up and sharpened, this checklist will help ensure you have all the ingredients necessary to make the final product that will leave your agents begging for more.

Check out some of the resources for more information about what we're going to be covering at the surface level while building out the solution.  Having a firm grasp on the different parts needed for the final product will help you better understand the solution and how you can customize and maintain it for your team for years to come.

Why do we want to have standards on best practices when using Power Automate?

Enabling experimental features affords us much more 'real estate' when writing formulas in Power Automate + a few more bells and whistles

The importance of renaming variables and actions to items carrying meaning to help readability and maintainability against your flows

Optimize your flow to avoid unnecessary actions and resources that count against your service level

Avoid issues that can come about by having a single owner for flows

Errors happen, sometimes on purpose and sometimes on accident.  Anticipate them and add logic to account for them to avoid flows terminating unexpectedly.

There are known issues in Microsoft Power Automate.  Learn how to make peace with them until they're formally resolved.

Even with low-code/no-code, assembling a flow takes time.  Get some back by using the clipboard and SCOPE for common routines reused between flows.

Save yourself time and frustration by making copies of your flow prior to making significant changes.  You're welcome.

As an administrator in Zendesk Support, we create our new fields and selectively add them to our forms where we want to make the options visible.

As an administrator for Zendesk Support, we add the triggers and webhook call placeholders for calling the summarizer and escalation flows.

Optional section on an analogy on how technology is changing rapidly with AI and how it became a moving target for this course.

Incremental build of the flow to introduce key variables and perform proof of concept calling and handling the Zendesk Comment API endpoint

Add looping logic to capture when there are more than 100 comments on a given ticket.

Find today's date and use it for determining milestones.  Filter out unnecessary items from the JSON array and rename some elements more meaningfully for later use.

Develop the prompt and pass the filtered and renamed content to the completions endpoint.

Pass the returned information from Chat GPT back to Zendesk, Outlook and if necessary, Teams.

The output doesn't look right, can you figure out why?

We're going from crawling to walking by taking away our instant flow trigger and replacing it with the webhook call from Zendesk.

Extract the parameters needed for completing the flow as passed from Zendesk - agent name, email, ticket number, subject and initiator.

Execute an example that touches on all of the positive logic of the flow, summarizing the ticket back to the ticket as a private comment, sending an email and performing an escalation to Teams.