Confidence in the details

Go to [[Watson AI is changing how business is done]] or [[Ideas for Newsrooms (LIST)]].

Also see the [[Master of Philosophy - Main Page]]

Company: KPMG

Industry: Tax consulting

This is what I want to get into - deep diving the tax code with AI to find the best solutions for customers.

“The government stands to give back $148 billion in the next 10 years, but getting those rewards isn’t easy. It’s hard to build up detailed evidence to qualify,” said Brad Brown, Chief Innovation Officer, Tax, KPMG LLP.

On the nature of AI in Q&A and exploration applications:

“Watson isn't a PhD grad out of the gate. It starts off as a kindergartner and works its way up,”

Results

From IBM:

Using 10,634 documents, Watson was trained on five different R&D models by 13 tax professionals. After training, Watson recommended the correct tax treatment about three out of four times. 

How Watson learned, in five easy steps:

1. Watson was trained

Watson was trained on five different R&D models by a tax team of 13 with more than 10,000 documents.

2. Watson was tested

End users tested Watson early and often to develop a system that worked the way they did.

3. Watson was launched

Watson began helping with research credit services and now helps five more service lines transform their offerings.

4. Watson got results

Watson recommended the correct tax treatment about three out of four times.

5. Watson keeps learning

Watson learns from each interaction and gets better at generating more detailed results every time.