AI Agents

Conversational agents that can deflect customer questions

Conversational agents are only as good as your documentation! If the answer to a question is not in your documentation, do not expect your AI agents to know the answer.

AI Agents can automatically answer customer questions and carry out a support workflow, including escalating a question to a team member.

Expectation Setting

AI agents are only as good as the content you provide them. This means having lots of specific knowledge base content is the best predictor of whether your AI agent can deflect customer questions. If you often find yourself answering customer questions with links to your knowledge base, you will likely see good results with using AI agents.

Principles we follow:

  • If your AI agent is not confident in an answer to it's question, it will not attempt to answer it.

  • Your AI agent will only attempt to address tickets that is has been assigned to. We provide you full control in determining what questions should be attempted.

  • Your AI agent will always cite sources directly to the customer, so that the base knowledge can be accessed directly.

Setup

Reach out to the Pylon team to see if you are a good candidate for using AI agents in your workflow. A good candidate is someone who has or is invested in creating a lot of documentation (Pylon has features to help make this easy 🙂). A good candidate is also someone who often gets a lot of questions about documentable product features or behavior.

Persona

We recommend making it clear to your customer that their question is being answered by an AI agent with the way your agent is named or their avatar.

Select a name and avatar for your AI agent. This is how they will appear to customers. Examples of common name formats are your company mascot/avatar or a name like "Ashley the AI" or "Ashley the Bot".

Knowledge Sources

  • Once the Pylon team has enabled this feature for you, go to Settings > Training Data

  • Determine the knowledge sources you want available

    • Your Pylon knowledge base will be enabled by default and reindexed live

    • Provide urls to individual webpages or a base url to crawl (ex. the home page of your documentation). These sources will be reindexed every 7 days.

  • Give Pylon a few moments to automatically scrape and index all your content

Response Configuration

  • For each source of issue (slack, in-app chat, etc.) configure the behavior at various stages in the conversation with our messaging builder

  • Define behavior for when the conversation starts, when the agent doesn't know the answer, when the agent needs to hand off to a human, and more.

Testing

Understand exactly how your AI agent will respond to customer questions by simulating questions in our testing playground. See what knowledge sources it will use, what it will do when it can't answer, and how it structures its response.

Usage

Consider your new AI agent as a member of your team. You can assign them to issues you'd like them to take a look at. This means you have full control of which customer issues your AI agent attempts to answer.

Manual Assignment

You can also assign an individual issue to your AI agent, the same way you might normally assign an issue to a team member.

Automated Assignment

You can use a feature like triggers to granularly scope which issues your AI agent attempts. For example, you can have it only attempt to answer

As defined by your setup, if your AI agent is unable to answer a question, it will follow your escalation workflow and reassign the issue.

Last updated