Taxpayer Services Chatbot
Description
This project implemented the Taxpayer Services chatbot, which provides self-service Frequently Asked Questions (FAQ) to taxpayers and international taxpayers regarding refunds, status of refunds, amended returns, Employer Identification Number (EIN), Earned Income Tax Credit (EITC), change of address, identity theft, extension to file topics. The Intent Engine, using Natural Language Processing (NLP) and AI algorithms classifies taxpayers' utterances (questions) into intents (requests). The answers provided to the taxpayers are not created by Generative AI. The responses of the chatbot and business logic is predetermined by content owners. This AI tool aims to classify/navigate to a correct predetermined response.
Detailed example
The NLP algorithm functions by classifying user inputs into predefined topics using advanced natural language processing techniques. It evaluates the input against a set of trained models and assigns a response based on a confidence score calculated by the algorithm. Only responses associated with these pre-established topics are provided, ensuring accuracy and consistency. The algorithm does not generate new or ad-hoc answers outside the defined topics, making it reliable for controlled environments where precision and adherence to predetermined content are critical.
AI / analytics pattern
Natural Language Processing: AI that processes, interprets, and shares information in human language.
Automation level / stage
c) Deployed – The use case is being actively authorized or utilized to support the functions or mission of an agency.
Expected benefit
This chatbot provides means for the taxpayers to receive information without need to speak with an Agent. The benefits include reduced taxpayer burden as well as increased internal IRS operational efficiencies.
Audit / financial statement impact
The output is not presumed to be high-impact and is not used as the principal basis for significant decisions/actions
Controls / human review
ATO: Yes; PIA: Not published
Data needed
Learning set of utterances-to-intent maps.