TABLE OF CONTENTS
- Introduction
- How it works
- Privacy
- How to get started
- Language interaction
- Important to keep in mind
- Topics & Sentiment Reporting
- API Support
Introduction
The Feedback AI Assistant is our market leading text and sentiment analysis tool powered by Gaugie AI, it is a more powerful and reliable way to derive structured data from comments without the need to maintain a training library.
How it works
The assistant will analyze the comment and look for any of the topics you have set up in System Admin. If a topic is mentioned the assistant will then assign a sentiment to the detected topic. A topic can only be assigned one sentiment: positive, negative, neutral or mixed.
Topic analysis
Detects any relevant topics mentioned in the comment based on the list of topics managed in Feedback AI Topics.
For example: “our account manager is really supportive and the support desk is very fast to solve any of our issues” this comment would have two topics detected - "Account Manager" and "Support Speed".
Sentiment analysis
Assigns a sentiment (positive, negative, neutral or mixed) to the detected topics in a customer comment based on the feeling that the topic was mentioned.
One of the best uses of sentiment analysis is the ability to get the “feeling of a topic” without having to read every single feedback. You can leverage this to pick up on trends and where improvements are needed.
Privacy
This feature by default redacts any personal identifiable information (PII) and business identifiable information (BII) before sending the comment to be analyzed by OpenAI and no additional data is sent other than the comment.
The redaction of PII, BII or both settings can be enabled or disabled by System Admins, check the step by step guide below.
How redaction works
- Step 1: The original comment is first sent to be processed by AWS Comprehend tool to find any mentions of identifiable information such as company names, people names, addresses, usernames, passwords, etc.
- The comment then has all those identifiable information replaced by a tag such as <ORGANIZATION> for company name or <NAME> for a person, so even though the sensitive information has been redacted the Feedback AI assistant can still generally have a sense of what is being communicated.
- Click here to read through how the feature used for Person Identifiable Information works.
- We utilise this AWS Comprehend feature to detect COMMERCIAL_ITEM and ORGANIZATION entities, and redact these entries from the unstructured text. Click here to read through how the feature used for Business Identifiable Information works.
- The comment then has all those identifiable information replaced by a tag such as <ORGANIZATION> for company name or <NAME> for a person, so even though the sensitive information has been redacted the Feedback AI assistant can still generally have a sense of what is being communicated.
- Step 2: Only after both these redactions have been applied to the texts used in the prompt, CustomerGauge will send the information to OpenAI for analysis.
Enabling and disabling redaction
1. Navigate to Settings → System Admin → System Settings → Gaugie AI → Privacy Settings
2. Click on the toggle for the setting you would like to enable/disable.
Personal identifiable information is any information that can be used to identify a person.
Business identifiable information is any information that can be used to identify a company.
3. Click on Proceed to save the change. It will take effect immediately. Please be aware that disabling these settings will cause personal/business identifiable information to be shared with OpenAI.
How to get started
- The first step is to get all the relevant topics for your business added in the Feedback AI Topics management page and make sure only the topics you'd like to be detected are activated.
- With your topics all ready, contact your account manager and request the Feedback AI Assistant to be turned on.
Language interaction
The Feedback AI Assistant is able to comprehend multiple languages and will detect the topics even if the comment made is in a different language than the topic was set up.
For example: if you set up your topics in english and you receive some comments in spanish, the AI will still be able to detect the topics and they will be shown in the platform with the same Reporting Label that you have set them up with, regardless of the language in the comment.
Important to keep in mind
- When setting up your topics, avoid having multiple topics with a similar meaning.
- Although the limit of topics is 250, try to maintain only topics that are useful and relevant activated.
- Only comments that has at least 10 characters and less than 2000 characters are analyzed.
- Only responses received after the feature was turned on are analyzed.
- Gaugie AI may still sometimes inaccurately detect Topics.
Topics & Sentiment Reporting
Currently the following widgets are able to report on Topics and/or Sentiments:
API Support
Our GET Responses API can be used to retrieve the topics and sentiment of records when sending a request with the tags parameter, for more detailed information, refer to the support page linked above.