Kaliper Documentation
Setting up Detectors
Detectors are instructions that tell the AI what information to identify in a call transcript.
To configure, go to Scoring setup → Detectors tab.
Detectors are reusable — build a library once, then assign the same Detector to as many Scorecards as you need.
Detector types
Type | What the AI does | Best for |
|---|---|---|
Yes/No | Returns YES or NO | Compliance checks, confirming a phrase was said, any binary question |
Extract | Pulls a specific value from the transcript | Capturing dates, amounts, names, plan types, or any stated fact |
Disposition | Classifies the call into one of your defined categories | Call outcome buckets, lead quality buckets, intent classification |
Managing the Detectors library
Using templates
Kaliper provides a library of pre-built Detectors for common use cases. Toggle Show templates ON to browse them.
Templates are read-only. To use one, open it and click Add to Scorecard — this creates an editable copy in your org.
Limits
Maximum 500 Detectors per organization
Maximum 50 Detectors per Scorecard
Maximum 50 Scorecards per organization
Organizing with categories
Categories are folders — they appear as collapsible rows in the table with their Detectors nested underneath. Useful once your library grows large. Create a category with New category. Categories have no effect on scoring.
Version history
Every time you save a Detector, Kaliper stores a full snapshot. These snapshots appear in the Change log at the bottom of the editor.
Browsing and filtering
The collapsible filter panel (labeled "Search" in the UI) lets you narrow by type, category, and whether templates are shown.
The table has a Quick search icon for instant name filtering.
Table columns:
Column | What it shows |
|---|---|
Detector | Name, with a Template badge for pre-built Detectors. Grouped under collapsible category rows if categories exist. |
Category | Which category the Detector belongs to. |
Type | Yes/No, Extract, or Disposition. |
Target Answer | The target answer set at the Detector level, if configured. |
Used | How many Scorecards this Detector is assigned to. |
Updated | When the Detector was last saved. |
Manually creating a Detector
Click Add Detector at the top of the Detectors tab.
Fields
Name The display name for this Detector. Also used as the column header when the Detector is added as a column in the Phone calls table.
Category An optional folder for organizing your library.
Type Select Yes/No, Extract, or Disposition. When you select a type, the relevant fields appear — they're not visible until a type is chosen.
Scorecards A shortcut to add this Detector to Scorecards without navigating into each one separately. Adding a Scorecard here applies default settings:
Score of 10
Not Critical
Not Conversion Related
The Score, and Critical/Conversion Related flags can be customized in the Scorecard editor — see Setting up Scorecards.
Yes/No and Extract Detectors
For a Yes/No Detector, a Target Answer field appears (YES or NO). You can override it per Scorecard in the Scorecard editor later.
The Prompt field is the exact instruction sent to the AI. Write it as a clear, specific question or extraction instruction — vague prompts produce inconsistent outputs.
Good Yes/No prompts:
Name | Prompt |
|---|---|
Primary Language English | Based on the transcript, was the primary language spoken by the caller English? |
Treat as YES if the majority of the caller's speech is in English and there is no explicit request to speak with an agent in another language. Treat as NO if the majority of the caller's speech is not in English, or if they explicitly request another language. | |
Caller is Homeowner | Based on the transcript, did the caller explicitly confirm they are the homeowner of the property where the security system is to be installed? Treat as YES for direct confirmations like "I own my home," "this is my house," or "I am the homeowner" with high confidence in firm, unambiguous responses. Treat as NO if caller states they are a renter, live in an apartment, are calling for a rental property they do not own, or if there's any ambiguity/hesitancy about homeowner status (low confidence). |
Active Bank Account for Payments | Based on the transcript, does the caller indicate they possess an active checking or savings account for premium payments? |
Treat as YES if the caller makes explicit statements like "I pay with my checking account" or confirms having a bank account when asked about payment methods (high confidence for firm, direct responses). Treat as NO if the caller states they only use cash/money orders, lack a suitable bank account, or express hesitation/inability to provide banking details (low confidence undermines YES). |
Good Extract prompts:
Name | Prompt |
|---|---|
Property Type | During our agent's portion of the call before transfer, what type of property did the caller indicate? |
Extract terms like "single-family home," "apartment," "commercial property." | |
Partner Name | Extract the exact name of the partner company that the agent says the caller will be transferred or connected to. |
Return the company name exactly as spoken; if no partner name is given, return "NA" | |
Zip Code | What zip code did the caller provide? |
Extract the 5-digit U.S. zip code; NA if it was not provided. |
Disposition Detectors
Disposition Detectors work differently. Instead of writing one prompt, you define items — the categories the AI chooses between. The final AI prompt is generated automatically from your items and definitions — you don't write it manually.
Click Add new to create an item. Each item has two fields:
Value The label the AI outputs when it selects this category. This is what appears in the Phone calls table and call detail window. Keep it short and descriptive: "Converted", "Not Interested", "Callback Requested", "Wrong Number".
Definition A description of what qualifies as this category. The AI uses this to decide which item fits a given call. The more precise the definition, the more accurately the AI classifies.
Good Disposition items:
Item Value | Item Definition |
|---|---|
Dropped | The call ended abruptly due to disconnection, technical issues, mid-sentence cutoff, mid-enrollment without resolution, no agent acknowledgment of the last customer statement, no proper farewell/conclusion, or extremely short/incomplete/nonsensical transcripts lacking meaningful progress. |
Agent Hang Up | Agent collects information and quickly ends the call. Agent gathered details but did not engage further before disconnecting. |
Solicitor | Customer is attempting to sell a product or service. |
Customer was a salesperson or marketer trying to promote a product or service. |
Reusing existing Detector text
The Select existing button lets you pick any existing Detector in your org. It copies that Detector's name into the Value field and its prompt into the Definition field. This is a text shortcut only — it creates no structural link.
Detector-level context
The Advanced section contains a single field: Context prompt.
This is background information sent to the AI alongside every Detector in this Scorecard — the campaign type, what the agent is required to do, compliance rules, or domain-specific terminology.
The context prompt supplements the main Detector prompts — it doesn't replace them.
© 2026 A-Launch. All rights reserved.