GPT-Powered Answers
Onboarding guide to use Clearfeed GPT-Powered Answers
Before starting, ensure that you have integrated your Slack Workspace with ClearFeed and check that your product edition is set to "Internal" on Product Editions.
Once done, you can create your first collection - a group of Slack channels that have similar helpdesk settings
Note: GPT-Powered Answers are available only for Professional and Enterprise accounts. You can also access this feature during the free trial.
Create Collection
Navigate to Collections from the side navigation bar.
Click on the
Add new Collection
button on the bottom-left of the page.Choose support team members as Responders
By default, all users that are agents will be considered as Responders
Go to user section to add your support team members as users and make them Agents
Only Agents can reply to the tickets from the triage channel and Clearfeed web application and can be assigned tickets
Non-Agents can still view tickets and make private comments on tickets
Add Request channels on which you plan to provide support to your users
All public Slack channels can be added from the web app
Private Slack channels can be added by typing
/invite @Clearfeed
on the Slack channels. They can only be added after a Collection is created
Add a Triage channel for the support team to work from on Slack
You can select
Create new channel
orSelect existing
optionYour support team can also work directly from Request channels, in which case you can opt in
Proceed without triage
Under Ticketing, select ClearFeed ticketing and choose the trigger mode for ticket creation
Trigger mode for the ticket can be an emoji (eg. 🎟️) applied to the thread or tickets can be auto-created on every new message on the request channel
You can edit an existing workflow to send a notification to the corresponding triage channel if there is no response to a request for 60 mins
Add Knowledge Source
In the Clearfeed dashboard, go to the 'Answers' section in the left navigation bar and click on the 'Add Knowledge Source' button.
The Knowledge Source will either start indexing immediately (reflected by an Indexing status) or will be picked soon (reflected with an Initiated status). Get more details on Managing Knowledge Sources here.
Testing Answers
Confirm that all necessary Knowledge Sources are added and their current status is Indexed. You cannot test a Knowledge Source having any other status.
Navigate to Answers from the navigation bar and click on the
Test
tab to enter the testing environment.Use the dropdown menu on the right to select the appropriate tags for the Knowledge Sources you want to test.
Enter your question into the text box and click the "Run" button to begin the test.
The generated answer and the search results (snippets) will be displayed in a text box below your question.
Collection Settings
After adding the knowledge sources, go to the Collection Settings of the desired collection to enable 'Answers.'
Once the 'Answers' toggle is enabled, select from the following options: -
Virtual Agent - Virtual Agent responses are visible to the requester as well as responders.
Agent Assistant - Agent Assistant responses are visible only to responders.
Select and add the appropriate Knowledge Source from the dropdown menu.
After saving the configuration, you are all set to use GPT Powered Answers.
Try on Slack Channel
A request is received on the request channel.
This triggers the Virtual Agent, which starts searching within your indexed Knowledge Sources for snippets containing information pertaining to the request.
On finding the snippets, the Virtual Agent responds to the request. If the answering mode configured for this request channel's Collection is:
Search Results
: The Virtual Agent simply posts these snippets (as Search Results) in a threaded reply to the request.Generated Answers and Search Results
: The Virtual Agent uses GPT to generate an answer using the most relevant snippet. Then, it posts this Generated Answer, along with the snippets (as Search Results) in a threaded reply to the request.
A feedback question is present at the footer of the posted answer.
If the answer is helpful, users (requesters/responders) can click on the
Yes, mark as solved
button and solve the request. Else, if users do not find the answer helpful, they should click on theNo
button. This marks the answer as not helpful and keeps the request open.Get more details on the working of GPT Powered Answers in the Slack Channel here.
Last updated