Author: Yogi Schulz

Neural nets are to identify “non-compliant” tax returns. Revenue Canada is applying its talents, energy and resources to a high-tech adventure to improve tax compliance. I hope for their sake, that they’re not embarking on a high-tech boondoggle.

In the August 1, 1996, edition of Computing Canada, Revenue Canada staff proudly describe their initiative which uses artificial intelligence pattern recognition techniques and neural network learning concepts to identify “non-compliant” tax returns.

To me the Computing Canada article indicates that Revenue Canada knows full well that it has dug itself into a very deep hole with the complexity of the current tax code. The hole has become so deep that there is considerable doubt that genuine mistakes and deliberate fraud are being caught. In an effort to create the appearance of control, Revenue Canada is taking a flying leap at a piece of technology. The hope that the technology will bail it out of a tight corner may be easily dashed. It feels like trying to fix a broken leg with an extra large Band-Aid.

I wonder how you teach a neural network to spot anomalies among “Federal forward averaging tax credit” and “Additional federal foreign tax credit from Part II of Form T2209” and “Provincial foreign tax credit from calculation on Form T2036”. I am not making this up; these phrases are copied directly from the T1 General form for 1995. If you sense my skepticism about whether the well-meaning bureaucrats and their IT staff will achieve a positive outcome from their leap into artificial intelligence, you’ve caught my point. A dose of business process reengineering can be much more effective in helping us to climb out of the deep hole the current tax code has placed us in.

I believe that the management focus at Revenue Canada is on entirely the wrong priorities with this technology fling. Instead, I would ask: How much energy is Revenue Canada pouring into supporting tax payers who do their best to be compliant? How much leadership is Revenue Canada applying to drafting legislation which will simplify tax collection, reduce costs at Revenue Canada, and reduce costs for me to be fully compliant with the laws of the land? These ideas will get us to more stable government finances and deficit control.

I’m appreciative of Revenue Canada’s recent efforts to be more responsive and provide faster, better service to Canadian taxpayers. However, taxpayer compliance hardly seems like the place to apply neural network technology.

Let’s apply the neural network technology to genuine research, product development and commercial processing opportunities, not as a giant Band-Aid for an out-of-control tax code. Let’s apply the well-understood process of business process reengineering to make life simpler for every taxpayer and Revenue Canada employee.