Agent Assist - Google AI

How to set up Google AI Agent Assist

Talkative's integration with Google CC AI Agent Assist allows smart suggestions to show to the advisor based on the customer's messages.

Pre-requisites:

• Enterprise Licence - please contact us to discuss pricing for Google Agent Assist integration

• New agent console

Ideally (but optionally) you will also have:

• A Google Dialogflow chatbot

• A public-facing FAQ knowledgebase

• At least 30,000 chat interactions in your Talkative account with meaningful data i.e. have messages from both agent & customer

Set Up

Google Agent Assist requires input from Talkative support. Please contact us to provision your Google Cloud account and perform the set up steps.

Once this has been done, when you accept interactions in the console, you will be given message suggestions from Google:

Suggestions are in several categories:

• FAQ answers - pulled from your knowledge base

• Chatbot responses - pulled from Dialogflow

• Smart suggestions - managed in Google Agent Assist console

Data

Only the message transcript is sent to Google, pre-chat form data and other interaction metadata is not used. You are able to select a regional data centre within Google when setting up Agent Assist.

Mid-interaction flow:

Customer, agent and chatbot messages are fed into Google Agent Assist. This is data that does not appear to be retained by Google past the duration of the interaction. Based on that context and based on data sources (below) google agent assist will provide suggestions for agents to respond with

Data sources

So that Agent Assist has context to respond with, it needs to pull data in from somewhere. The first two sources are:

Public

• knowledgebase
• chatbot

The third source is:

Private

if we want the chatbot to learn from past transcripts these need to be fed to it. This is optional. This means that Google will retain copies of your transcripts. They are retained indefinitely. While Google itself has the full transcript, PII and all, it attempts to censor out the PII before feeding it to the AI.

If you would like to send this data to Google, this does not necessitate increasing the data retention policy time in Talkative as it can be sent out immediately after the interaction (the transfer of transcripts is Talkative -> Google). However, we cannot retrospectively send transcripts that we do not have access to. This means that, if the Talkative data collection policy is not increased until we start sending transcripts to Google, we will not be able to meet the training requirements as quickly.

Training

The training for Agent Assist smart replies (suggesting things based on previous messages by agents) needs at least 30k transcripts. Sending the transcripts from Talkative->Google if being sent at the end of an interaction is negligible but sending 30k interaction transcripts at once will take up to a couple of days. After that, the model needs to be trained on the data. In our experience, it took about 2 days to learn off 30k transcripts. Google do not provide an estimate. This is optional and the bot is still usable without the data set, just no user data suggestions.