This blog was last updated on September 24, 2019
No one really knows what percentage of today’s invoices are noncompliant with VAT law. Given the complexity and rate of change in VAT law, there is an almost unlimited set of possible nonconformities on the scale between outright fraud and clerical errors. One big problem is that it is often very difficult to gather all the data that is necessary to make fully informed tax determination decisions for invoices in more advanced business scenarios.
Artificial Intelligence (AI) – a promising avenue for Tax & VAT compliance
The fact that most companies don’t like being out of compliance has given rise to a thriving sector using automation to arrive at better decisions. This industry has thus far relied on algorithms which facilitate decision making but cannot overcome the incomplete invoice data problem. Against this backdrop, it should not surprise anyone that artificial intelligence is considered a promising avenue for overcoming these weaknesses.
Now let’s fast forward to 2025, when AI-based tax determination will for sure be used for tax determination by some larger enterprises. Among other things, the use of AI can be expected to complement algorithms to fill in data gaps based on correlations the AI systems has learned from trading patterns.
Artificial Intelligence (AI) can help reduce invoice error rates
One can expect such systems to be very effective and greatly reduce the error rate in invoices – which will be very welcome because by 2025 most invoices are likely to be exposed to tax administration ‘clearance’ engines in real time. So what, now, if that invoice turns out to be incorrect? The tax administration may not find such errors because they’ll not often possess better information than the business itself. However, such errors may be found in many different ways: a competition law problem, commercial litigation, even a labour dispute…
Most legal systems have some standard of reasonableness, care or proportionality that would protect someone who has taken all possible measures one can expect from a responsible businessman from excessive tax penalties. Surely, using state-of-the-industry AI to further perfect tax determination in one’s invoices should be viewed as going to great lengths to comply.Or will it?
Machine Learning – not a substitute!
A lot has been written about the fundamental problem that these advances in machine learning pose to applying ageold legal rules. AI can get very close to perfecting data, but it does so by identifying correlations without semantically being able to ‘explain’ them. If you invoke a standard of reasonableness, this implies you were using the best of your skills to create and maintain circumstances within your control to avoid negligent behaviour, and the best of your cognitive abilities to do the same for what you cannot control. We believe AI can be an extremely helpful tool in pointing out ways forward where a business cannot reasonably be expected to have the right information for a decision. However, it cannot be a substitute for making decisions that keep humans in control wherever reasonable possible. AI is not here to clean up behind bad managers!
See our previous piece on AI and B2B automation/compliance here.
