Eric

Big +1 to this

Hey Arotar,

Are you building an app to live in the zendesk ecosystem or a web-application that you're wanting Zendesk to interact with via webhooks / api?

Happy Wednesday folks! Our second A.I. User Group session was a success! Despite some technical difficulties, we were able to walk users through a tutorial on how to write a great prompt in the context of the Zendesk ecosystem. Our session went through a common use case that we all have for A.I. - summarizing and concisely providing the context of a ticket to another team during a handoff / escalation. As always our session was accompanied by an example app that showcases this functionality which can be found here.

During the session, several key best practices were shared for writing successful prompts:

• Clarity and Precision
• Ensure prompts are clear and directly address the user's needs.
• Avoid ambiguity to prevent confusion and misinterpretation.
• Always be Succinct
• Keep prompts concise with clear instructions to keep the LLM on track.
• Avoid unnecessary details that can dilute the main message.
• User Intent Understanding
• Design prompts that capture and reflect your intent accurately.
• Tailor the language and structure to match your perspective.
• Structured Format
• Use a well-organized structure to facilitate easy comprehension.
• Include headings, bullet points, or numbered lists where appropriate.
• Contextual Relevance
• Ensure prompts are relevant to the specific context or situation.
• Adapt the language and content to fit your current needs.
• Actionability
• Include clear calls-to-action or next steps.
• Make it easy for the user to know what to do after reading the prompt.
• Feedback Incorporation
• Regularly update and refine prompts based on user feedback.
• Be adaptable to changing user needs and preferences.
• Testing and Iteration
• Continuously test prompts with real users and iterate based on performance.
• Use A/B testing to determine the most effective prompt variations.
• Alignment with Goals
• Ensure each prompt aligns with the overarching goals of the service or product.
• Avoid prompts that could lead to user frustration or goal diversion.
  
  

Analysis of the projects’ provided prompts:

In the session we walked attendees through 3 different prompts and why they are well written and the best practices that they followed. One of these can be found here:

You will be analyzing a customer service conversation to prepare a ticket for escalation to a specialized team. The escalation message should be structured as follows:


Reason for Escalation: Clearly specify why this ticket is being escalated. This could be due to:

Customer Sentiment: Indicate if the customer’s mood (e.g., frustration, urgency) warrants escalation for better handling.

Specialization Requirement: Explain if the issue falls outside the current team's expertise or is specific to another team's domain.

Technical Complexity: Describe if the issue's technical nature exceeds the current team's capabilities and requires more specialized knowledge or resources.


Recommended Team for Escalation: Suggest the most appropriate team or department to handle the escalated issue, based on its nature and complexity. Teams available for escalation include:

- Sales

- Product

- Tier 3 technical support

Action Items and Pending Questions: List any immediate actions that the receiving team needs to undertake and any questions that remain unanswered, specifying who (the customer or the new team) should address them.

The provided prompt serves as an excellent example of the best practices discussed. Let's break it down:

1. Ticket Summary

• Best Practice Followed: Conciseness and clarity. The summary is limited to 100 words, forcing focus on the core issue.

2. Reason for Escalation

• Best Practice Followed: These categories are aligned with understanding user intent and issue complexity.
• Structural Clarity: The clear division into categories follows a key best practice of structured prompt engineering.

3. Recommended Team for Escalation

• Best Practice Followed: Direct and actionable. It asks for a specific recommendation, aligning with the actionable nature of prompts that keeps them on track.

4. Action Items and Pending Questions

• Best Practice Followed: Calls for specific action items and questions, demonstrating the emphasis on prompt clarity and directness.

 As you can see this session provided some valuable insights into the art of prompt writing and engineering! As we continue to engage with our customers, let's apply these best practices to enhance our efficiency and effectiveness. Have any feedback or questions? Feel free to join in on the conversation.

I transitioned to working in the A.I. / LLM space about 6 months ago and it was incredibly daunting getting up to speed on the technology. I found some great success by going 'straight to the source' and found a variety of resources created by experts in the space.

General Learning resources:

- Andrej Karpathy is a researcher at OpenAI and has put out some phenomenal papers on his website and educational webinars on his youtube channel. The material ranges from non-technical to highly technical, so a little something for everyone.

- A beginner's guide to large language models 

Research and white papers that I found incredibly useful:

- The Research index by OpenAI

- Pythia: A suite for analyzing large language models across training and scaling

- QLoRA: Efficient Finetuning of Quantized LLMs

Hello Zendesk Community!

For those unfamiliar, each month the A.I. User Group hosts two events. One which is an open discussion around a topic that the group has decided to discuss in depth. Users are also able to use this session to bring up any questions or problems that they are currently facing with utilizing or implementing A.I. into their customer support workstreams. Our second session is education focused. These can be anything from a speaker joining to present on a certain facet of A.I., a technical tutorial, or a deep dive into Zendesk functionality that can be enhanced with the proper usage of Artificial intelligence.

We kicked off our 2024 A.I. User Group sessions with a fantastic start this January, and it was terrific to see so many familiar faces alongside new ones!

The focus of our discussion was on setting ambitious yet achievable goals for utilizing A.I. in our customer support strategies this year. Here's a brief recap for those who couldn't attend or want to revisit the conversations had:

• Enhancing Knowledge Bases with A.I.: This was a hot topic and continued the discussion from our December session on configuring your Help Center for use with A.I. (find the slide deck here). We delved into best practices in setting up a fantastic Knowledge base that can be leveraged as a data source for A.I.
• A.I. for Low Touch Tickets: We explored the potential of A.I. in handling simple support tickets along with tools available in the Zendesk Marketplace that can assist with this task. The consensus was that by using A.I. to tackle these 'low touch' issues, we free up valuable time for our agents to focus on more complex queries. This approach isn't just about efficiency; it's about enhancing the quality of your agents' work, allowing them to be impactful where it matters most.
• Understanding Customer Sentiments Through A.I.: We discussed leveraging A.I. to gauge customer sentiments, aiming to understand their feelings even before the customer satisfaction survey (CSAT) rolls out. This proactive approach is geared towards delivering a more personalized and responsive customer experience.
  
  

Each of these goals ties into a broader vision of making our customer support more intuitive, efficient, and empathetic through the power of A.I.

Now, we'd love to hear from you! What are your plans for using A.I. in your customer support strategy this year? Your insights and plans could spark new ideas to foster collaboration within the community, or even new features or products to help us all serve our customers better.

Looking forward to our next session! In which there will be a tutorial on how to optimize handoff, escalation, and “out of office” processes by leveraging A.I.

You can also click here and join our AI User Group to get notifications of our upcoming events.

Eric Nelson

askstylo.com

Hey Jim,

I think you could start the pipeline by calling a service which provides you the JWT Token then use that to update the `zcli.apps.config.json` before updating the app.

Hey there! There are some of Zendesk's services that use web sockets like (chat conversations). However the action of opening and updating tickets isn't done via this.

Hi Tami,

Admins are only allowed to create api tokens as the tokens themselves have the elevated permissions of an admin. I'm not familiar with Boomi, but you should see if they support OAuth as this would provide you the flexibility that you need.

Hey there!

Are these values using the secure flag? If so once set they won't be visible to the UI/ Are they set to required? If not I'd recommend doing this so that users can't move past the installation without filling them out. 

Gulzar Shikalgar,

Thanks for the info, this is expected behavior if your Webhook trigger is the last one. Essentially this comes down to the trigger cycle which you can read about here. The easiest way to resolve this is to put the webhook trigger higher in order so that it is one of the first to fire.