View source for QNA/Current/User/writegoodFAQ

{{Article
|Standalone=No
|DisplayName=How to write a good FAQ
|TocName=
|Context=Shows how to write questions and answers that Knowledge Search Service can understand, so it can give your customers the answers they need.
|ComingSoon=No
|Platform=GenesysCloud
|Role=Administrator, Developer
|Application=Platform Administration
|Prereq=
|Section={{Section
|sectionHeading=What do you need to know?
|anchor=
|alignment=Vertical
|Media=
|image=
|structuredtext=When a customer asks a question, they need information:

*''What kind of mortgages do you offer?''
*''How much is an annual subscription?''
*''Are you open on Saturdays?''

In order to provide that information, your bot needs to be able to figure out what they are really asking for. That's where training comes in.

After you finish writing your Frequently Asked Questions (FAQs), Knowledge Search Service uses machine learning to analyze these questions—along with the answers that go with them—and "remembers" what it learned during this ''training''.

The better job you do in writing your questions and answers, the more likely it is that Knowledge Search Service will give your customers the right answers.
|structuredtextwide=
|FAQHeading=
|Status=No
}}{{Section
|sectionHeading=Using the right words
|anchor=
|alignment=Vertical
|Media=
|image=
|structuredtext=Knowledge Search Service uses the training data to build a natural language understanding model that can detect what your customers are likely to ask for. A big factor in that is its ability to recognize the most significant words, concepts or phrases within your questions and answers, so it can correlate them with the questions it receives from your customers.

This means you need a wide enough range of vocabulary to give your bot the information it requires to recognize ''a varied set of questions'' that all have a similar answer. The questions and answers you provide must not be vague; they must also be distinct from one another.

To make this work, you must provide at least five complete question and answer pairs that include the key words that a customer is likely to use.

'''Good questions:'''

*''What are your hours?''
*''When are you open on Saturdays?''
*''How late do you close today?''
*''What are your hours on Wednesdays?''
*''How early do you open on Tuesday?''

Notice that each question uses a somewhat different set of critical words: "hours," "open," "close," "Saturday," "today."

Compare these questions to ones that don't really give the AI what it needs.

'''Bad questions:'''

*''When are you?''
*''How much hours?''
*''Test question''
|structuredtextwide=
|FAQHeading=
|Status=No
}}{{Section
|sectionHeading=Building blocks
|anchor=
|alignment=Vertical
|Media=
|image=
|structuredtext=You can think of the critical words used in the good questions above—such as "open," "hours," and so on—as building blocks.

These building blocks will ultimately be used to construct the relevant question. But first the software has to decode—to ''parse'', if you will—the input, looking for those individual elements that can be used as building blocks. Once it has found them, it can put together the answers your customers need—with a fairly high level of confidence.

Your training data will also include words that are irrelevant to the intent of your questions. These words are discarded during the training, just as they are discarded when your bot is fielding questions from customers.

'''Note:''' It's important to make sure that your training questions and answers do not just contain irrelevant words.
|structuredtextwide=
|FAQHeading=
|Status=No
}}{{Section
|sectionHeading=Best practices
|anchor=
|alignment=Vertical
|Media=
|image=
|structuredtext=As you work on your questions, keep the following best practices in mind:

*'''Focus on the most common questions:''' something like eighty percent of customer inquiries revolve around a relatively small set of questions. For example, a knowledge base might have higher top three recall with a hundred FAQs that correspond to eighty percent of customer questions rather than a thousand FAQs that correspond to ninety percent of customer questions.
*'''Keep questions short'''
*'''Don't cross-reference other FAQs within a single answer'''
|structuredtextwide=
|FAQHeading=
|Status=No
}}{{Section
|sectionHeading=Error messages
|anchor=
|alignment=Vertical
|Media=
|image=
|structuredtext=The following errors can occur during training:
<br />
{{{!}} class="wikitable"
{{!}}+
!Error message
!User action
{{!}}-
{{!}}The minimum number of FAQs required to train the model is 5. Only ''x'' FAQs were provided.
{{!}}Make sure your training input contains at least five questions and answers.
{{!}}-
{{!}}This question: "{question}" does not provide sufficient information for the machine learning model to be trained effectively. Refer to the Genesys documentation for information on how to write an FAQ that can be used by the training software.
{{!}}Add more usable words to the question.
{{!}}-
{{!}}This question: "{question}" includes an answer that does not provide sufficient information for the Machine learning model to be trained effectively. Refer to the Genesys documentation for information on how to write FAQs that can be used by the training software.
{{!}}Add more usable words to the answer.
{{!}}-
{{!}}The FAQs provided for training do not contain enough unique words. Refer to the Genesys documentation for information on how to write an FAQ that can be used by the training software.
{{!}}Add more questions and answers with different words. Make sure that at least some of these words are relevant to the questions and answers.
{{!}}-
{{!}}Internal error.  If the problem persists, please contact support.
{{!}}Contact Genesys Customer Care.
{{!}}}
|structuredtextwide=
|FAQHeading=
|Status=No
}}
}}