Documentaal, ChatGPT and Python

ChatGPT for your help desk?

The AI chatbot ChatGPT needs no introduction by now. Almost every Internet user has tried it out at some point. It seems like pure magic: you ask it to explain Einstein's theory of relativity to a 12-year-old child and you get a clear and, on the face of it, correct answer in flawless Dutch or English.

Would it be feasible to build a chatbot based on ChatGPT to replace a staffed help desk? Help desk staff regularly get the idea that the answer to a customer question was crystal clear on the help pages, but that a customer was stupidly too convenient to look it up themselves.

Pure magic, but what are the snags?

As much as ChatGPT seems to know, the model cannot contain everything. Specialist knowledge about the application your company offers will not be found in it. We have to put that knowledge into it ourselves.

We could start training an existing model. However, this may cause the model to produce untruths or outright nonsense, or "hallucinate," Further, you would have to retrain the model if the documentation is changed or expanded. Therefore, instead, for each question we will make a selection from the entire knowledge base for our application. We will send this selection along with each question. This way you avoid hallucinating and the answers will automatically stay up to date.

Kadd knowledgeadd to ChatGPT

So to get started. It still requires some programming to get everything running, but for an experienced Python developer, it's perfectly doable on its own.

The first results are encouraging, but also no more than that: answers, while correct, are not complete. To the question "How can I create a model?" the chatbot describes the creation of a static model, but about dynamic models, which are also there, it says nothing.

OK, but can ChatGPT still a solution?

Yes, fortunately we do. We pre-instruct ChatGPT's system to list ALL options, if applicable. We can also have the same instruction checked by the user (see first image).

In addition, we create another field where the user can ask follow-up questions, such as about certain details in the given explanation. The second image illustrates how that works out.

What's funny is that the answers are not always the same: sometimes you get a numbered list of actions indicating how to insert a table, for example. Other times it's just a straightforward description. Content-wise, though, everything is correct, and that's what matters in the end.

ChatGPT help desk ChatGPT Helpdesk

Summary

To summarize all the experiences: ChatGPT, in the form of the OpenAI API driven by Python, can certainly add value to your help desk. As long as the provided knowledge base is correct and complete, many users will indeed find the answer they are looking for. This leads to less workload for the helpdesk staff and an improvement in the quality of the answers, response times and customer satisfaction

Mshould your organization also get started with ChatGPT?

In conclusion, we have found that integrating ChatGPT into your help desk can be a great solution. Help desk staff can focus their attention on more complex issues, while ChatGPT can handle the simpler questions independently. It takes the simple but time-consuming calls away from your help desk, leaving more capacity for important customer service issues.

If you want to know exactly what ChatGPT can do for your help desk, contact Documentaal and one of our experts will be happy to help you. Documentaal has extensive knowledge and experience in ChatGPT, customer service processes and integration into your existing help desk. Let us demonstrate to you how easy it is to use this advanced technology to support your help desk by using your capacity more efficiently.