S
member badge
Smith.ai Product
How it currently works:
  • After a VA call, an LLM classifies the caller disposition and status based on the content of the transcript.
Problem:
  • This classification is done independently of the playbook used, which creates the possibility of the disposition and the playbook not being aligned.
  • Additionally, we've observed the same caller receiving different dispositions for similar types of calls.
Solution:
  • CF2 has the ability to set disposition data directly within playbooks (similar to setting a data key to a specific value)
  • We need to update the CF2 templates for SSVA to use these special play steps so that the dispositions are set deterministically according to the playbook used.