Advanced NLP (AI Intent Assist)

what is ai intent assist ai intent assist improves nlp intent detection using ai it helps the bot understand user queries more accurately when exact keyword or phrase matches are not found this feature will be available for all versions ⚠️ ai intent assist does not generate responses it only improves intent identification how to enable ai intent assist open your bot go to settings → solution settings enable ai intent assist confirm the behavior in the modal note if the feature might not be available on your bot, please reach out to your respective account manager to get it enabled how ai intent assist works (behind the scenes) ai intent assist enhances the existing nlp intent detection by adding an ai based intent matching layer it does not replace nlp; instead, it acts as a secondary detection step when standard nlp matching fails step by step flow standard nlp intent detection the bot first attempts to detect an intent using the existing nlp engine this is based on keyword and phrase matching from trained intent examples nlp match fails if the user query does not contain keywords or phrases similar to the training data nlp intent detection fails without ai intent assist, the bot would return a fallback or error message ai intent assist re attempts detection when ai intent assist is enabled, the system re analyzes the user query using ai the ai compares the user query against all existing intents to identify the closest match ai based intent matching the ai evaluates each intent using intent name training questions intent description it determines which intent best matches the meaning of the user query, even if the wording is different intent resolution if a close match is found the corresponding nlp intent is triggered if no suitable match is found the bot sends a fallback message, or routes the query to ai agent flows (if enabled) what data ai intent assist uses for accuracy ai intent assist relies on existing intent metadata to determine the closest match the quality of this data directly impacts accuracy 1\ intent name represents the high level purpose of the intent used by ai to understand what the intent is meant to handle 2\ training questions examples of how users might phrase queries for this intent used to understand variations in user language and phrasing 3\ ai description (ai assisted field) a descriptive summary of what the intent covers this field is auto generated by ai using the intent name and training questions users can review and edit this description before saving ⚠️ this description field is critical for ai intent assist accuracy , as it helps ai understand the intent’s scope beyond keywords best practices for using ai intent assist effectively to get the best results from ai intent assist, follow these guidelines 1\ use clear and meaningful intent names avoid vague names like intent1 or generalquery use descriptive names that reflect the user’s goal (e g , check order status , apply for credit card ) 2\ add diverse and relevant training questions include different ways users might ask the same question cover common paraphrases, informal language, and partial questions try to cover synonyms 3\ review and edit the ai generated description read the auto generated description carefully ensure it accurately represents what the intent should handle what it should not handle update it to clarify the scope where needed do not write more than 2 lines; the more rules and checks in place, the more chances of hallucinations 4\ test and iterate test with real world user queries if the wrong intent is matched add more training questions, or refine the intent description to narrow or expand scope re test after every update to improve detection accuracy 5\ combine with ai agent for best results use ai intent assist to improve intent detection use ai agent as a fallback where ai will intervene and respond to the user queries unstructured queries long tail or complex questions this provides the most reliable and scalable setup how to test ai intent assist test exact intent phrases test paraphrased queries or more complex queries test long or ambiguous questions validate correct intent routing test fallback scenarios when no intent matches current limitations ai agent first with nlp fallback is not supported rich ui elements are currently handled better in nlp flows