What is TaxSaver.Club?
TaxSaver.club uses natural language processing and deep machine learning to understand the highly complex tax code and answer practical day-to-day financial issues.
Filing taxes takes too long, costs too much money and is far too overwhelming a process for taxpayers. Americans spend 6.1 billion hours each year doing their taxes and complying with tax laws. According to a General Accounting Office (GAO) study, failure to itemize deductions is costing people nearly $1 billion in overpaid income taxes. The growing complexity of the tax code (4M words as of 2012) makes it almost impossible for any human to completely understand and find tax breaks for which they are eligible. 68.5% of taxpayers take the more straightforward standard deduction even though they would have been able to deduct more had they itemized. The reason people don’t itemize deductions is because it takes too much effort to find the information needed to take itemize deductions. This results in overpaying on an average of $438 per taxpayer.
How we built it
We have used IBM Watson Natural Language Classifier to analyze a string of text, and assigned predefined categories to it by applying deep-learning technology. Deep learning is based on a relatively recent set of approaches that use algorithms with similarities to the way the human brain works and we have applied to text classification.
We created a slack bot so that you can ask tax questions from within slack and have them answered. We have 2 backend servers - Watson Natural Language Classifier on IBM Bluemix and the Slack bot.
Challenges we ran into
Getting the Watson sample code working was a challenge as the code out of the box kept crashing with a cryptic error message in Java (which our team had very little background on). Majority of our time was spent on training the Watson Natural Language Classifier using various questions and answers.
Accomplishments that we're proud of
Early on we decided to limit the scope of work as what we are building was too complex. We decided to just use slack and not have to worry about the front-end. This saved some development time.
There are plenty of tax preparation services like: Intuit®, H&R Block, Jackson Hewitt Tax Service, etc. They can only work with information provided by their clients. Currently no automated solution exists to provide intelligent and accurate tax advise to consumers. Consulting companies like Deloitte, Ernst & Young, PwC, KPMG etc. provide tax services that are typically targeted at high net worth individuals. TaxSaver.club’s vision is to democratize access to high quality tax advice for the common man.
What we learned
We learned how to:
- Setup Natural language classifier
- Train the classifier for tax information
- Formulate training data for deep learning in the format that IBM expects.
What's next for TaxSaver.Club
We are planning to start a company and build the AI technology for TaxSaver.club. We will launch the service by offering a free tax return audit. This service will be focused on analyzing expenses to claim itemized deductions and will be augmented by human assistance allowing us to get to market early. IRS allows amending tax returns for the previous 3 years. We will only charge on a contingency basis based on helping taxpayers get a refund.