1) Keep it simple!
Write short, clear statements.
One behavior per line.
Avoid filler words and vague phrasing.
2) Match detail to scoring intent
Generic ⇒ catch-all scoring: “Rep explains the warranty terms.”
Specific ⇒ pointed, exact-match scoring: “Rep states the warranty covers parts and labor for 12 months.”
The more specific you are, the more the AI will check if the transcript exactly matches what’s written.
3) Use statements, not questions
Do: “Rep confirms the customer’s problem before proposing a fix.”
Avoid: “Did the rep confirm the customer’s problem?”
4) Use correct grammar
Complete sentences, proper capitalization, clean punctuation.
5) Use “If / Then” for edge cases
Best for events that may or may not occur; define the required response.
Examples:
“If the customer expresses frustration, then the rep acknowledges the emotion before proceeding.”
“If the customer mentions a competitor, then the rep follows the competitive handling script.”
Template: “If
<event/condition>then<required action>.”
6) Use “AND / OR / AND-OR” between statements
AND — all must happen
“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.”
OR — only one must happen
“The rep must verify the customer via their email address.”
OR“The rep must verify the customer via their account ID and phone number.”
AND-OR — one or more acceptable
“The rep provides a product guide.”
AND/OR“The rep shares a tutorial video.”
AND/OR“The rep schedules a follow-up session.”
7) Separate content into distinct paragraphs
Break criteria into smaller, clearly defined paragraphs or bullet points.
Easier for the AI to parse, reducing the chance of misinterpreting combined actions.
Improves scoring accuracy by letting the AI evaluate each requirement individually.
8) Use graded marking for soft skills and pass/fail for technical skills
Soft skills (e.g., empathy, tone, rapport) often benefit from graded marking — for example, 1–5 scale for “how well” the rep performed the skill.
Technical skills (e.g., verifying account, providing correct product info) should be pass/fail — either the step was completed correctly, or it wasn’t.
Mixing graded and pass/fail scoring gives a balanced evaluation:
Objective accuracy for technical elements.
Nuanced assessment for interpersonal performance.
9) Minimize ambiguity by avoiding subjective words
Words like professional, helpful, clear are subjective unless defined.
Replace them with observable behaviors the AI can detect.
Instead of “Rep was professional,” write “Rep uses polite greetings, avoids slang, and maintains respectful tone.”
10) One action per scoring item
Break complex behaviors into separate items so each can be scored independently.
Bad: “Rep confirms the customer’s details, explains the product, and offers a discount.”
Better:
“Rep confirms the customer’s details.”
“Rep explains the product.”
“Rep offers a discount.”
11) Use consistent structure
Start each scoring item with “Rep must…” or another fixed phrase.
Consistency helps the AI learn patterns and reduces scoring errors.
12) Use AI-friendly keywords
Include key terms exactly as you expect them to appear in transcripts.
If synonyms are acceptable, include them:
“Rep must say ‘shipping date’ or ‘delivery date’ when confirming the timeline.”
13) Order items logically
Arrange items in the sequence they are likely to appear in the interaction.
This makes parsing easier and prevents context confusion in scoring.
14) For graded items, anchor scores to real examples
When possible, add examples of transcript excerpts for poor, average, and strong performance.
Example for empathy:
Poor: “That’s not my problem.”
Average: “I understand that’s frustrating.”
Strong: “I understand that’s frustrating, and I’ll take care of it for you right now.”