Its depends on your product.
Our product is very big and can have a lot of various use cases to be handled.
We are already implemented ChatGPT. How it works?
Read the customer email message.
Check in knowledgebase, system, other sources for possible solutions.
check is there older conversations for the customer (for past 3 days) do they maches.
Sumarizes all requests, and pick all needed info.
provide a reply as an Agent in ticket as private note.
If the answer fufill the request, agent just copy and paste the text.
At the moment we launched it in English, russian, latvian, estonian, lithuanian languages.
As extra - we use GPT to define a theme of a ticket (on what question customer is contacting us).
Also we detect is the ticket is autoreply - when customer support center replies to our requests: thank you for you message, we reply in 24 hours and similar.
also we use it cover some use cases which are straight forward and have exact workflow what need to be done.
In general - its hard to implement it. You have to have very good knowledgebase. If your product is easy to handle, like e-shop - implementation would be way faster. If your product is more complex - then prepare for inifinitive work
We are working with it already for almost a year. ANd still many improvements need to be made.
In reply what you want to have, so summarization is easy to use.
Grammar - as i understand you want to check grammar of Your agent, so this need to be done by chrome plugins (its way better way )
The reply part is hardest. As i said before - depends on your product ant knowledgebase you have.
To be honest, it can be worth thinking about let Chat GPT create a “shared draft” within the ticket instead of a private note. Then the agents do not need to copy&paste, they can simply use the existing draft and modify before sending.
Yes.
Or under private note should be a button: “Create a reply” where by hitting it, all private note text moved as a reply with ability to edit it before sending.
Having a private note as history entry also very useful. Due you need to see how GPT generated a reply, and what was the real reply to the customer. So you could check/validate/match and solve the problems why GPT didnt answered what it need to be answered