Opposition to Automation at the CRA Misses the Point

Unions are pushing back against AI automation at the Canada Revenue Agency (CRA), arguing that job cuts will make it harder to go after tax cheats. The CRA, like any public agency, does not exist to employ people; it exists to deliver outcomes as efficiently as possible with the resources it has. That means generating the most compliance for each dollar spent.

The primary focus of the CRA is to maximize the amount of tax compliance it can generate for each dollar it spends. Its focus is not to maximize audits or staff.

But for years, the CRA scaled enforcement mainly by adding staff, not by upgrading the systems that make each auditor more productive. That wasn’t irrational. Hiring is fast, visible, and easy to deploy under pressure. Investing in systems is slower, harder to implement, and politically less salient.

During COVID, those constraints tightened. Demand surged as new programs were rolled out and compliance pressure increased. The CRA had two options: build systems or add people. It chose to hire.

That choice solved the immediate problem. It also locked in a labour-intensive model of enforcement, thereby raising costs.

But compliance comes from auditing the right files, not simply auditing more of them. Two thousand high-yield audits will recover more revenue than ten thousand random ones. Enforcement is fundamentally a targeting problem, one that depends on information—primarily knowing which filings are most likely to be non-compliant.

AI-based systems can analyze patterns across millions of filings at once, identifying anomalies no human team could reliably detect. Humans work through subsets and rely on rules of thumb. These systems evaluate the full distribution, incorporate more variables, and continuously update their predictions.

Even modest improvements in targeting materially increase recovery per audit. AI improves detection, prioritization, and case selection, as has been seen across tax administrations globally. The result is straightforward: Each auditor becomes more effective, and fewer are needed to achieve the same level of enforcement.

The CRA expects automation to cut repetitive tasks by half, with savings reinvested into higher-value compliance work and an estimated $1.1 billion in annual gains. The intent to shift resources away from routine processing and toward targeted enforcement is clear.

That changes the work itself: Less time spent on low-value, repetitive tasks and more focus on complex cases that actually drive revenue.

Of course, not every application of AI will work. Early efforts at the CRA have been less than impressive, and implementation inside government is uneven. But the comparison is not between AI and a perfect system. It is between AI and the current model of human-led selection, with all its own limits and blind spots.

If better systems increase enforcement output, then fewer workers won’t weaken enforcement. The CRA isn’t there to preserve jobs. It is there to deliver results as efficiently as possible.