Best Practices in Creating and Maintaining Content
Keep the content base with less than 20% N2 sections
Understand the composition of the content base
The content base is composed of N1, N2, and Interactive sections (when Eddies are being used). The ideal content base in
terms of retention is one that contains only N1 and/or Interactive sections and does not escalate requests to humans,
making ClaudIA more effective and providing faster responses to customers (compared to humans).
Consider the quality of service
However, the quality of service may be affected if we try to automate 100% of requests. There are situations where a
human agent needs to intervene. For these cases, we create N2 sections.
Avoid creating a queue of requests
If all sections were N2, we would have no level of automated service. This would leave the retention level at 0% and
could generate a queue of requests, also affecting the quality of service.
Seek balance between N1 and N2 sections
To maintain good levels of retention and quality, it is essential to balance the sections that automate service (N1) and
those that escalate to human assistance (N2).
Follow the recommended maximum percentage of N2 sections
We recommend that the maximum percentage of N2 sections in a knowledge base be 20% of the total. For example, in a
knowledge base with 1000 sections, ideally, up to 200 should be N2.
Write the section title clearly and without spelling errors
Imagine that the user sent a message that ClaudIA could not respond to correctly because there was no content in its
base. You know the answer to the customer's question, which leads you to create a new section in the content base.
In situations like this, or when there is a need to create or change some content, it is crucial that the texts are
clear and free of spelling errors. This increases the likelihood of ClaudIA understanding the user's question and
locating the correct section to respond.
Tip: Do not insert the customer's message in the section exactly as they wrote it. Correct the spelling and, if
necessary, make the question clearer.
See an example of a section with an inadequate title:
Now, observe an example of a session with an appropriate title:
Remove unnecessary sections
After identifying that a section is being frequently used incorrectly (in this video, we teach how to map these cases),
delete it.
In many cases, these sections were created deviating from best practices, and there is no problem in deleting them,
especially if they are N2.
Constant maintenance of content is a lever to drastically improve the quality and retention capacity of tickets by
ClaudIA.
Video on how to map retention offenders.
Template to map N1 opportunities based on N2 sections and for ticket analysis (section was used/section was used
correctly).
Delete duplicate sections
Duplicate Sections: What Are They and Why Avoid Them?
Duplicate sections are those that have identical titles but present different answers or forwarding. In addition,
sections with similar titles that should be activated in the same situations but contain distinct answers are also
considered duplicates.
Why avoid duplicate content?
NEVER create duplicate content, especially with different types (N1, N2, or INTERACTIVE). This practice can cause
various problems, such as:
- Difficulty in content maintenance: repeated content generates constant rework.
- Risk of incorrect answers: increases the chance of ClaudIA selecting the wrong section to respond to the user's
query.
Example of Duplicate Sections
In the examples above, in addition to the sections having different responses, they also imply distinct directions: the
first is an N1 case, while the second is N2. Note that in these cases, ClaudIA tends to prefer to escalate the ticket to
N2.
Example of the Impact of Duplicate Sections
How to identify duplicate sections?
There are two tools in the Hub that can be used to identify duplicate content:
- Content screen, using the "semantic" filter:
- Playground
In both cases, the suggested step-by-step is:
1. Access the tool;
2. Add a question with possibly duplicate content;
3. On the content screen: check if there are two or more sections with similar titles - and, mainly, different
responses;
4. In the Playground: check if the answer was correct and click on the "i" to check if another similar section was
present in the topK;
Both tools work like a semantic search mechanism, that is, they search for content similar to the added question.
Do not include greetings in IDS sections
Avoid using greetings like 'Hello! How are you?' along with responses in the sections. ClaudIA already greets the user
at the beginning of the conversation or when necessary.
Understand that this practice can intensify the feeling of "robotization" of service for the customer. It can also
increase the chance of ClaudIA triggering an incorrect section when the user greets. This also harms the results of
ClaudIA's tagging tool, the Tagger.
See below an example of an inadequate section:
Do not create "no content" sections in IDS
Similar to the above item, avoid creating sections that do not have content. Like sections that try to adapt ClaudIA's
tone of voice:
- Title: "Okay, thank you"
- Response: "You're welcome, I'm here to help you!"
Sections like this hinder ClaudIA's performance, adding noise to its knowledge base.
