Skip to main content

Training Rules

Apart from the knowledge, sometimes you may want to provide the AI Assistant with specific instructions on what actions to carry out based on specific conditions. For that purpose, you can use special condition-action instructions called rules, where the condition part must relate to the user, and the action part must relate to the AI Assistant.

Example of a rule

“If a user asks for help with resetting the password, the AI Assistant must connect the user to the real agent support.”

Important information about rules:

  • The AI Assistant pays extra attention to the rules, but there is no guarantee that the AI Assistant will always execute the action.

  • The AI Assistant can execute only those rules that are located close to the information (in training notes) relevant to the user query. That is why, instead of listing all rules together in a single document, rules should be placed in relevant documents near contextually relevant information.

  • Rules don’t have strict format, but require that condition was user related and action was AI Assistant related.

Valid Rules

”If user ..., then AI Assistant must ... .”
“When person ..., bot has to ... “
“Whenever human ..., ai bot needs to ...”

Invalid Rules

“If AI Assistant ..., then ...”

  • Avoid using conditions based on exact phrases.
Good rule

”If user says that he/she forgot the password, then AI Assistant must ... .”

Bad Rule

“If user writes “i forgot the password”, then AI Assistant must ... .”

  • Avoid using action based on exact phrases.
Good rule

”..., then AI Assistant must inform the user that it can’t help with that and apologize for inconvenience.”

Bad Rule

“..., then AI Assistant must say “Sorry, I cant help you with that.”