Reinforcing Tax Compliance with Machine Learning and Blockchain

In a previous post, I pointed out that tax authorities’ efforts to improve collection based on traditional approach of acquiring more data, faster, have met with mixed success and resulted in the accumulation of substantial amounts of data to store, manage and protect from cyberattacks. What would a smarter approach look like?

Two technologies come to mind, and both can bolster tax compliance and fraud detection: machine learning (ML) and blockchain – or more specifically, blockchain cryptography. 

Machine learning, a subset of artificial intelligence, is an increasingly important technology for big data analytics. Supervised learning, for example, is a form of ML that enables you to uncover patterns in large datasets by capturing correlations between inputs and outputs and creating an algorithm which can make predictions. Chile used this approach in a national e-invoicing system that analyzes the patterns between invoice item descriptions and tax rates. When assessing a new invoice, the algorithm predicts whether the tax is correct and provides a probability score to indicate how much confidence should be placed in that prediction. Low-probability invoices can be routed to staffers for manual screening.

Social network analysis is another form of ML, one that measures the strength of association between two parties. It can help authorities detect networks of actors working together to accomplish a variety of frauds such as missing trader and fake-invoice schemes.

Recent advances in ML have enabled “federated learning,” which allows multiple parties to collaborate and build machine learning algorithms without having to expose their data to each other. This opens the possibility of future joint data contributions from taxpayers and tax administrators that could be used to create new algorithms.

Blockchain technology, as a free-entry, borderless, globally networked set of systems, presents all kinds of challenges for national tax administrations. But recent advances in blockchain cryptography may remove a major roadblock to fully digital tax compliance – the issue of how administrators can establish trust for a taxpayer’s system output. Even if a machine learning algorithm were created jointly by the tax administration and the taxpayer, how can tax officials be confident it was actually used in that form by taxpayers?  How can they ensure it was not modified prior to use?

There’s a parallel here with the early days of online commerce. E-commerce only took off once consumers trusted it – in large part, because they were reassured by SSL encryption and security certifications. Blockchain cryptography innovations extend trust from people to systems; both the data and the algorithms can now be secured. This means that tax administrations can remotely monitor taxpayer systems without the need to extract data or protect it, because it always remains under the taxpayers’ control. Systems can be synchronized and then kept coordinated automatically.

In short, blockchain cryptography creates the possibility of a continuous monitoring capability for tax authorities. Together with ML applications, it holds out the promise of a very different kind of response to the twin challenges of compliance improvement and fraud prevention. These developments make it imperative for taxpayers and tax administrations to work together to develop fair and suitable ways to leverage machine learning, blockchain and other rapidly emerging technologies.