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
Hi Leonardas, im Marcel from germany and i have to implement ChatGPT into our Zammad. Is there any chance to get in touch with you to discuss yor approach. Or can you tell me wich tools you are using? Do you use a middleware or python, JSON etc? What about Zammad Triggers, Webhooks?
I just need a thing to start with. That would be very nice. Thanks in advance.
In our org, we’ve only just started using text modules to draft replies to customers and I believe that they’re well-defined enough in advance that there really isn’t a need for us to resort to using Chat GPT to figure out how to reply to a customer. Disclaimer: I’m the person doing the initial drafts on all the text modules.
But if there’s ever a time where I’m stuck on how to reply to a customer, this is the process I’d use:
Begin draft of reply, putting a summary of the stuff I think should go into a paragraph in all caps.
Send Slack message to supervisor asking them to review the draft and come up with suitable reply.
Review supervisor’s reply, c/p’ing verbiage as appropriate as a new response to the customer. If it’s been less than 10 minutes since the Slack message went out, delete the draft response.
Update ticket, assigning it back to supervisor for final review and ticket closure, along with an emoji reply to the supervisor that confirms that the revised reply was sent.
I think that’s a less risky method for helping agents compose responses than using Chat GPT. And besides, you also want to be employing agents who know how to communicate well in text, right? Which means that they wouldn’t really need to use Chat GPT in the first place, right?
What I intended to say is that I agree that the ability to highlight text in an internal note and and reply to it could be useful towards answering OP’s questions about knowing how easy it would be to implement/incorporate Chat GPT into their operations and how they use Zammad. That’s all.