Understanding Agent Transfer Count Discrepancies in Analytics Summary

Overview

This article explains why customers may observe a mismatch between the total Agent Transfer count displayed on the Analytics Summary page and the language-wise breakdown (e.g., English, Arabic) in the Kore.ai platform.

Below is one such example where:

  • Summary page total Agent Transfers: 35
  • English: 21
  • Arabic: 24
  • Language-wise total: 45

Since 45 ≠ 35, users may assume the totals are inconsistent.
This Behavior, however, is expected based on how multilingual conversations are classified within analytics.

Root Cause: How Multilingual Conversations Are Counted

Kore.ai Analytics categorizes conversations based on the languages used during a session. The counting behavior works as follows:

1. Single-Language Conversations

If a user interacts in only one language from start to end, the conversation is counted once under that language.

2. Multilingual Conversations (Language Switching)

If a user switches from one language to another in the same session e.g., starts in English and later switches to Arabic the conversation becomes a multilingual conversation.

When language filters are applied:

  • The English filter includes:
    :heavy_check_mark: Pure English conversations
    :heavy_check_mark: Multilingual conversations that include English
  • The Arabic filter includes:
    :heavy_check_mark: Pure Arabic conversations
    :heavy_check_mark: Multilingual conversations that include Arabic

Impact on Totals

Because the same multilingual session appears in both language filters:

  • The sum of language-wise counts will be higher
  • The Summary page displays the count of unique conversations, which will be lower

This is why in the reported case:

  • Summary unique count = 35
  • Language-wise filter totals = 21 + 24 = 45
  • The additional 10 represent multilingual conversations counted twice (once per language)

This is expected behavior, not an error.

Example Scenario

Session ID English Arabic Summary Count English Filter Arabic Filter
C101 :heavy_check_mark: :heavy_multiplication_x: 1 1 0
C102 :heavy_check_mark: :heavy_check_mark: 1 1 1
C103 :heavy_multiplication_x: :heavy_check_mark: 1 0 1

A multilingual session like C102 contributes:

  • 1 count in Summary
  • 1 count under English
  • 1 count under Arabic

Hence, totals can differ.

Customer Guidance

If you notice similar discrepancies:

  • Be aware that language filters reflect all conversations that include that language, including multilingual sessions.
  • Always refer to the Summary tab for accurate overall counts.
  • Use language filters only when analyzing language-specific engagement, not totals.

Content by : @sravan.dara