Do you use AI/LLMs to improve your answers?

Lobby
1

If you are working within Zammad as an agent, you probably write a lot of mails to your customers. Do you have a habit of letting classic AI/LLMs assist you in writing your answers? Do you copy & paste an answer into Ollama, ChatGPT, Claude, Gemini and ask it to help you with the text? What do you ask your AI to do with it?

I would be glad, if you could share some uses cases.

I sometimes use a local ollama, if I have the feeling that a text I wrote feels “clunky” or over complicated. Then I ask ollama to “make it more enjoyable to read without adding unnecessary humor or jokes”.

2

Nope. We’re a two-person organization and our writing skills are very strong. If we were to expand, any additional agent we hire would need to have strong customer service skills, especially with writing email responses.

In our two-person org, whatever the boss decides is the best way to phrase something is what eventually will be sent to the customer or published in any self-help docs we end up putting on our website. We would not trust an AI to write things for our customers.

1 Like
3

We are three people company and respond to about 500 different people and three different languages. I personally use ChatGPT a lot to rephrase my answers before I send them through Zammad so I write a text and then put this text into ChatGPT to improve the quality of the written text. Especially to improve the “tone to the customer” and also spell everything correctly. This saves me a lot of time, to not have to rephrase every sentence perfectly. With this the LLM can help me a lot. In general, it would be great to have such a function built in with a click of a button to set the tone correctly. Because sometimes especially in IT agents (in this case I mean real people) do not find the correct tone when corresponding with a customer and it would help a lot to have this built in with some kind of API.

1 Like
4

We integrated Zammad with N8N. When we put an internal note containing “CallAI” a trigger is send to N8N to create the answer.

4 Likes
5

Good morning! Could you explain better how you do this?

6

Mount an n8n workflow that fires with a webhook trigger, an has an AI agent in the end. Give that last AI agent HTTPRequest tools connected to you Zammad API instance, so the agent can consume those API and update you ticket or send a reply (endpoints: ticket_articles and tickets).

Then go to Zammad and create a webhook that points to said workflow. Create a Trigger that fires when: article is created AND article in internal note AND article contains “CallAI”, connect this trigger to the webhook. Thats it.

If you want you can create a modelo text to “CallAI” so you can call it quicker

1 Like
7

To the initial question. I often refine my complex replies by copy-pasting my draft to various tools : grammarly, translators back and forth, chatgpt. I ask to correct my grammar, improve my wording, or simply ask ChatGPT with short prompts like improve/simplify/concise my text. Sometimes I only need to improve just a single phrase or a paragraph, not the entire answer. Still, none of these tools helps alone, that’s why I often use several.

Thinking about what could help to simplify my workflow, I can only imagine a UI with three text fields like production/stage/test, where test is filled by AI with an input from stage, to which I can place short excerpts from the prod, while prod is my final full version. Having a short prompt field above the test field could help to customise the AI to play around.

1 Like
8

@aliaksei I agree and like your ideas for zammad displaying more stages or versions or variations of responses, perhaps from different llm sources or using different prompts and context, during work in progress.

9

I respectfully disagree. I think that any person who works at a customer service help desk for a small team/org needs to either already have impeccable communication skills or be someone who can and is willing to quickly adapt their communication style to mesh with the tone the team wants to project and the team’s values.

Also, for most people, a Web browser can correct for spelling errors. I do not know if there are any grammar-checkers built into browsers, though. So perhaps a grammar tool would be a better tool to have built into Zammad than an AI tool if folks are just looking to use AI for that.

Before I get too into the weeds, why are you asking?

10

In August we were still in a phase of “what problems of Zammad users can be solved using this new technology called AI”.

If you are interested in our thoughts there are some blog posts, you might find interesting:

1 Like