If Ezra is not responding with the correct answer/phrasing you have entered for him in the Knowledge Base, the first step is to check the answer tags.
Firstly, make sure the tags are of a normal length. Given the fuzzy-matching of tags is determined by the length of the tag, overly long single tags may match in odd circumstances. Fuzzy matching refers to Ezra’s ability to understand a word even if spelled incorrectly, depending on Ezra’s certainty in identifying the correct word.
Unless the tags need to be in the exact order, longer multi-word tags should be grouped in an ‘AND’ set. While you can have more, it is best to tag two words unless absolutely required.
“Enterprise BI Portal” as a tag is quite long, especially as they are also all words in the dictionary. “Enterprise” & “BI” & “Portal” is a better tag set.
Because of Ezra’s scoring, ‘AND’ sets only are necessary where an idea needs to be represented by all tags in the set. If separate entities together strongly imply an answer, but separately, also imply the answer, keep them as ‘OR’ tags.
“What do you eat?” and “What do you like to eat” could return different answers from your Knowledge Base, despite the two being essentially the same question. Even though the second answer is defined as separate from the first by the word “like”, the rest of the tags do imply this information without the word “like”, so an ‘AND’ set is unnecessary. The more generic answer will have a higher score when the user has not used the word “like” in their question.
Try not to use multi-word dictionary words if you can help it.
Use ‘AND’ sets to break up longer phrases, but only if they are a part of the same idea.
Tag answers with queries and verbs sparingly.
Consider Ezra’s scoring algorithm, and try to separate generic responses from more specific answers by not using ‘AND’ sets unnecessarily, or to group separate ideas.