Best Practices: Get the most of AI

Here are some best practices on how to get the most out of AI

Better Questions:

  • Longer and more expressive questions are better than shorter ones

    • 👎 Bad: "New Cluster"

    • 👍 Good: "How do I set-up a new cluster"

  • Avoid single word questions

    • 👎 Bad: "Cluster"

    • 👍 Good: "How do I set-up a new cluster"

  • Try to reword a question, if you don't get an answer

    • 👎 Bad: "Unable to provision a new user"

    • 👍 Good: "How do I add a new user"

  • Avoid asking questions based on transactional data

    • 👎 Bad: "What is the total revenue of SR231 Campaign in North America"

    • 👍 Good: "How do I look-up the revenue of a campaign based on filters"

Better Data:

  • Is your data source being referred:

    • Check if the answer included some of sources you expected the answers to come from. For instance if you expected an answer to be found in your help center. Check if the citations had any document reference from the help center. If not, please reach out to the Twig team. We have a few ways to dive into this scenario.

  • Knowledge Base:

    • Twig offers a built-in knowledge base. Questions and answers in the knowledge base are added directly to the AI memory. This ensures that the AI answer is provided exactly the way the knowledge base article is written. here are some scenarios to consider using the Knowledge base

      • Frequently Asked Questions

      • Email Macros / Canned Answers

  • Synthetic Questions and Answers:

    • Conversation streams like ticketing data can be very noisy. This results on poor data quality for AI and bad answers. If you want to ingest conversation streams consider using Twig's synthetic QnA engine. This method extracts questions and answers from the data source. The AI generated QnA is used while the original data source may be discarded.

  • Word/Url Replacements

    • Sometimes companies use new terminology instead of previously used ones. Twig is able to replace old terminology and URLs while ingesting data. This gives you more control over answers

Refine Answers:

  • Answer Refine shortcuts: When you review an answer you are presented with a few tools to refine the response. Twig provides 6 short customer to Shorten, Lengthen, Make Empathetic, Make Fun and Make Bulleted responses.

  • Use your own prompt: By using your own prompt you may refine Twig answers. This gives you better control over how you can use the AI.

Ask for help:

There are several scenarios where the Twig team and help quickly resolve an issue. Please feel free to reach out, Here are a few common scenarios.

  • New Data: When a data source has been made redundant and a new data sources is used in its place

  • Data not being referred: If you know an answer should have been found in a data source but its not being used.

  • Change in business: When a business or taxonomy/naming changes in your business and documents still refer to the old names.

  • Different Truths: When the same question needs to provide different answers depending on who is asking the question, this applies to product versions, licensing levels etc.

  • Regional Needs: Twig supports multi lingual AI, if you have the need to region specific needs, let the team know.

PreviousOutlook App Installation

Last updated