QA Playbook Cheatsheet: Elaborating 'Standard' Criteria

Best practices for writing AI-friendly descriptions of standards criteria in QA Playbooks

Core Principles

While setting up your AI QA Playbook, you can use the core principles mentioned below for accurate and consistent AI results.

Keep it simple

Each requirement of an elaboration should be written as a short, clear statement focusing on one specific behavior or action. Avoiding filler words and vague phrasing ensures the AI can accurately identify and evaluate the described behaviors in transcripts.

Use correct grammar

Proper grammar, complete sentences, and clean punctuation are essential for accurate AI-understanding. Inconsistent formatting or grammatical errors can lead to misinterpretation and scoring inaccuracies.

Eliminate ambiguity and subjective language

Minimize ambiguity by replacing subjective descriptors with observable behaviors. Terms like "professional," "helpful," or "clear" are subjective unless specifically defined.

Criteria Example

Bad Elaboration ❌

Good Elaboration ✅

Professional

Rep should be professional.

Rep uses polite greetings, avoids slang, and maintains respectful tone.

Signposting

Rep performs helpful signposting where necessary.

Rep provides step-by-step instructions with confirmation checks when the customer asks how to perform some task on the application.

Use statements, not questions

All elaboration items should be written as declarative statements rather than questions.

Criteria Example

Bad Elaboration ❌

Good Elaboration ✅

Probing

Was the customer's phone number and delivery address verified for a delivery issue?

  1. Delivery Issues:

    1. Must verify customer's phone number

    2. Must confirm complete delivery address

  2. (… other scenario-based cases)

Introduction

Did the agent mention "(the brand name)" and their name?

Both "(the brand name)" and the agent's name must be present in the opening greeting.

Avoid contradictory requirements

When describing requirements in the elaboration of criteria, the description should define what would satisfy the criteria if it were applicable to the conversation.

This point is especially important for criteria which are expected to fail on "Yes" evaluations or be marked as Fatal on "Yes".

Criteria Example

Bad Elaboration ❌

Good Elaboration ✅

Was jargon used in the conversation?

(Fail on "Yes")

Jargon terms:

  1. (term 1)

  2. (term 2)

  3. … and so on

—————

(This may result in the AI marking the criteria as "Yes" if they did not use jargon)

The rep must avoid using the following prohibited terms:

  1. (term 1)

  2. (term 2)

  3. … and so on

Was the rep rude?

(Fail on "Yes")

Language and behavior should be checked.

—————

(This may result in the AI marking the criteria as "Yes" if language and behavior are appropriate)

The rep must not use language or behavior that could be interpreted as rude by the customer.

Handle conditional scenarios and logical relationships clearly

For situations that may or may not occur during interactions, use structured "If/Then" statements. For example, "If [event/condition], then [required action]." When connecting multiple requirements use "AND/OR" relationships appropriately.

  1. AND Relationships require all specified actions to occur:

    • "The rep must share the Help Center for self-serve support." AND

    • "The rep must log a support ticket." AND

    • "The rep must confirm the customer receives the ticket reference."

  2. OR Relationships require only one of the specified actions:

    • "The rep must verify the customer via their email address." OR

    • "The rep must verify the customer via their account ID and phone number."

  3. AND-OR Relationships accept one or more of the listed actions:

    • "The rep provides a product guide." AND/OR

    • "The rep shares a tutorial video." AND/OR

    • "The rep schedules a follow-up session."


Bad Elaboration ❌

Good Elaboration ✅

Conditional Scenario 1

Rep handles technical problems as needed.

If technical issue occurs, then rep offers callback within 2 hours.

Conditional Scenario 2

Rep should handle frustrated customers appropriately.

If customer expresses frustration, then rep acknowledges emotion before proceeding.

Logical Relation: AND

Rep gets email and maybe confirms it.

Rep collects email address AND confirms spelling AND repeats back to customer.

Logical Relation: OR

Rep verifies customer using email or account or phone.

Rep verifies via email address OR account ID with phone number.

Logical Relation: AND/OR

Rep gives some help materials.

Rep provides user guide AND/OR tutorial video AND/OR schedules training.