AI Playbooks Best Practices

Prev Next

Crafting the right definitions and topics for your AI Playbooks ensures effective evaluation and accurate analytics.

Let's go over some tips and tricks on how to create effective topics.

Topic-definition relationship

The topic name and definition must be related. The topic name should summarize the definition, and the definition must explain the topic name.

Good Topic-Definition relationship

Bad Topic-Definition relationship

Topic: Customer Budget

Definition: The budget the customer has allocated for this project.

Topic: Step 1

Definition: The budget the customer has allocated for this project.

Single topic pairing

A topic-definition pair should focus on a single subject.

If the topic description contains more than one topic, we recommend separating them into multiple topics.

Good topic pairing

Bad topic pairing

Topic: Main Pain Points

Definition: The most significant challenges or obstacles currently being faced.

Topic: Customer Budget

Definition: The budget the customer has allocated for this project

Topic: Main Pain Points and Budget

Definition: the most significant challenges or obstacles currently being faced, and customer budget

Clear and concise

The topic name and definition should be clear, concise, and objectively measurable.

Avoid subjective or immeasurable assessments like “friendliness” or “politeness”.

If a topic name or definition is too vague, Dialpad AI will not have enough detail to accurately detect the topic in real-time, which can result in missed topics (incomplete analytics) or even AI hallucinations.

Long topics or definitions can also result in missed detections and inaccurate evaluations because the focus is too narrow. Topics should be detailed for accuracy, yet broad enough to cover various instances.

Recommended

Too vague or long

Topic: General Customer Satisfaction

Definition: How happy the customer is with Aerolabs’ service after interacting with the company representative.

Topic: Expected Outcomes After Choosing Aerolabs

Definition: Customer expectations and desired outcomes after migrating their shipping services to Aerolabs. Understanding these expectations can help providers facilitate the transition smoothly and ensure customer satisfaction, whether the focus is on enhanced packing, a simpler user experience, faster shipping times, or cost efficiency.

Topic: Additional Solutions

Definition: Additional service options or solutions to resolve customer concerns.

Topic: Post-Migration Expectations

Definition: Customer expectations and desired outcomes after migrating to Aerolabs.

Tips

An ideal topic name is 4-40 characters.

The ideal topic goal is 25-150 characters.

Avoid biases

If a topic includes gender, racial, or other group stereotypes, misinformation, or societal biases, it can lead to biased or false detections by focusing on specific language patterns, resulting in inaccurate and unfair outcomes. Remember, the “questions” should assist the agent by providing practical hints or examples.

Bias Examples

  • “How old is the candidate?”

    • Using age as a topic or description can lead to biased data collection, potentially discriminating against various populations.

  • “What is the candidate's nationality?”

    • Using nationality as a topic or description can lead to data collection based on harmful stereotypes.

  • “What was the overall tone of the conversation?”

    • Tracking conversation tone through "friendliness" or "politeness" topics results in biased evaluations due to the narrow set of available linguistic patterns.

Avoid personally identifiable information

Topic titles and descriptions should not contain personally identifiable information (PII) or cause the collection of sensitive information like credit card details, social security numbers, passwords, or email and home addresses.

PII Examples

  • “What is the candidate's complete birthday?”

  • “What is the candidate's address?”

  • “What is their CVV”

Titles shouldn't directly address agents

To keep topics consistent throughout your playbooks, avoid addressing agents directly.

Choose straightforward topic titles like "Customer's ideal price point" over phrases like "You should discuss the ideal price point".

Use the "Questions" field for agent guidance or hints.