- Post Call Strategy Model: calls come in different lengths, so transcripts come in different sizes. Since LLMs work on token count, you can use different models for different call-duration segments. For very short calls, or calls with no user message, you can skip post-call analysis entirely and assign a default value to each variable.
- Conversion Reason: the condition that makes a call count as a successful conversion. Example: if the goal is to book a demo, then “Demo Booked” is the conversion reason. Once the demo is booked, the call is marked converted.
- Post Call Analysis Prompt: instructions that tell the AI how to analyse the call and what to look for in the transcript.
- Post Call Data Retrieval: define the information you want to extract from the call. You add variables based on what you want to pull out. There are 4 types — Text (for words, sentences, names, addresses), Selector (when the answer must be chosen from a fixed list), Boolean (for Yes/No or True/False), and Number (for values like age, budget, quantity, or phone numbers).
- Call Drop-off Reason: define why a call was unsuccessful or ended without converting. This helps you understand why customers could not be converted. Example: Customer Busy, Requested Callback, Budget Issues. You can add your own reasons in the “Add Call Drop-off Reason” section.
Audit post-call accuracy
Set up an Audit Strategy
Choose post-call metrics and add audit variables for the values you want to check.
Review Audit Results
Review Postcall Accuracy and inspect expected and actual values for each issue.