Meet the Smart Checklist

What Is the AI Checklist Trigger?

Balto’s new AI-based Checklist trigger transforms how Checklist items are completed on calls. Previously, Balto used phrase matching: the agent had to say most of what was scripted in the Checklist item for it to trigger. Now, with the LLM-based AI trigger, the mechanism that makes an item trigger is decoupled from the words the agent actually sees.

This means you can define why an item triggers and what counts as success without being locked into specific scripted language. The AI evaluates the full context of the conversation, not just keywords.

Why It Matters

Phrase matching works well in many situations, but it has some limitations:

Overlap: Intros and closes often sound similar (“Thank you for calling…”), causing items to trigger at the wrong time.
Rigidity: Wording varies call to call, but phrase matching requires consistency.
One-sided: Only agent language could trigger items, until now.

The AI Checklist trigger solves all three of these problems.

When to Use the AI Checklist Prompt

The AI Checklist Prompt is the right choice when:

Wording varies naturally from call to call
Two or more checklist items sound similar and overlap
You want a customer’s response (not just the agent’s words) to trigger an item
You’re tracking outcomes or confirmations (e.g., appointment booked, decision-maker confirmed)
You’re building in flexibility for rapport, recaps, or relationship-building moments

Tip: Not every item needs to switch. If you have specific verbatim phrases you want your agents to use, phrase matching is the way to go. Reserve the AI trigger for your trickiest items.

The 4 Steps to Writing a Strong AI Checklist Prompt

These four steps are your framework every time you write an AI Checklist prompt.

Step 1: Be Literal. Identify Exactly What You’re Trying to Do.

Write exactly what must happen for the item to be complete. Avoid vague language like “appropriately,” “properly,” or “clearly.” Instead, describe the specific behavior or outcome.

Instead of…	Write…
“Agent builds rapport appropriately.”	“The agent engages in friendly small talk before beginning the transaction.”
“Agent asks about project details.”	“The agent asks the customer to share specifics about what they want remodeled.”

Step 2: Identify Requirements or Exclusions. Specify Who Says What.

Clarify what counts and what doesn’t. Are there specific details that must be covered? Does it only count if the agent says the phrase, or can the customer trigger this item too? Being explicit here prevents false positives and missed triggers.

“The agent must ask how many windows are to be replaced.”
“Thanking the customer in response to information they provided (e.g., ‘Thank you, how can I help?’) is not sufficient to complete this item.”
“‘Thank you for calling’ alone is not sufficient.”

Step 3: Add Examples if Needed

Pull one or two real examples straight from your call transcripts. You can also include negative examples to show what doesn’t count. Examples help the AI calibrate accurately to your specific use case.

Must determine the scope of the project. May be described by the customer or asked for by the agent. Examples: “How many windows are you looking to get replaced?” (agent-initiated) or “I want to replace my carpet in my basement.” (customer-initiated).
The agent must say the company name when beginning the call. Example: “Thank you for calling. How can I help you?”should not trigger this checklist item.

Step 4: Define Who Can Trigger This Item

Ask yourself: does this item require the agent to say or do something, or could the customer also trigger it by volunteering the information first?

Agent only: Required for compliance disclosures or regulated language.
Agent or customer: Great for discovery questions. If the customer shares the info before being asked, it still counts.
Customer only: Use when you want to capture what the customer confirms or states.

Example prompt (insurance enrollment): “The agent must state this phrase verbatim “To complete this enrollment I need you to re-state your name and date of birth and that you agree.” Once the agent has completed the sentence, only trigger this checklist item if the customer repeats their name, date of birth, and says “yes,” “I agree” or another affirmative phrase.”

Example Prompts by Use Case

Prevent Overlap (Intro vs. Close)

“The agent should thank the customer for being a valued customer. This represents a successful sign-off at the end of the interaction. Any reasonable verbiage to accomplish this objective is OK. Thanking the customer as a response to information they provide (e.g., their account number) is not sufficient. ‘Thank you for calling’ alone is also not sufficient.”

Flexible Discovery Question

“The agent should ask the customer to share specifics about the type of trip they are looking to take, including location(s), date(s), and number of travelers.”

Verbatim Requirement

“The agent must say something similar with minimal deviation to: ‘Are you looking to cover anyone else in your household, like a spouse or child?'”

Win Marker / Appointment Confirmation

“Trigger when the agent confirms the date and time of the appointment. Must also include mentioning how long the appointment will be (usually 60–90 minutes) and that all decision makers will be present.”

Recap / Summary

“Trigger when the agent summarizes actions taken on the call as well as any next steps needed. Must include next steps.”

Getting Started

Pick 1–2 high-frustration checklist items that overlap, mis-trigger, or require too much scripting.
Write your prompt using the 4 steps above.
Test it using the Agent App.
Refine and expand. Like QA scoring, AI Checklist improves with tuning. It costs nothing to iterate.

Questions? Reach out to your Balto Customer Success Manager.