When I was a young child, I was fascinated by a Bagatelle board. Firing small metal balls to the top of a ramp, I could watch them fall, bouncing of metal pins, to ultimately end up in a number of different scoring zones. Now, many years on, I like to think about Machine Learning in a similar way. The scoring zones are the outputs of the Machine, the place you drop the ball, and the nature of the ball, represent the different inputs, and the ‘randomly arranged’ pins which deflect the balls on their journey to the scoring zone are the inner workings of Machine Learning. Whilst the pins of themselves do not have any symbolic representation of the connection between the inputs and outputs, by constantly refining their position within the board, you can alter the journeys & effectively create a machine which for a given input, will correctly predict the desired output. All you need is a lot of balls to pour into the board whilst you adjust the pins. Of course, in Machine Learning there are many more ‘pins’ than on a Bagatelle Board, and many more balls, but the principal and concept works nonetheless.
Now let’s turn our thoughts to Indirect Tax. Imagine if you had lots of data, instead of balls – perhaps a year or two of ITX transactions from an ERP system. Then imagine you had a board with a lot of pins, such as a custom Machine Learning environment. Then finally, imagine that you have all of the ‘correct’ tax coding decisions from that data; the Tax Coding from the ERP system for those transactions. As long as the mistakes in the data are sufficiently few (the data has been ‘cleaned’ as a training set of data) then the Machine Learning can be taught (moving the pins) to get the same outputs (Tax Codes) for the same inputs (Tax relevant data) as the people & rules that originally coded the transactions. With a technology such as this, you could review your ERP rules, your manual over-rides, and every other facet of your ITX coding. In fact, you could set it up to monitor your ERP system, and maybe one day replace it in your coding decisions.
Of course, if this all sounds like science fiction to you, then you can always go and play on an old Bagatelle Board, but if you think this is an exciting future, then why not join me on the PwC ITX Data Analytics Webcast this coming Wednesday 8th March, & I will tell you how soon you can have one of these PwC Machines learn Your ITX Coding:
I hope you can join me.