TLDR: AI experts are urging the Canada Revenue Agency (CRA) to significantly improve the accuracy and efficiency of its human-led call centre operations before fully embracing artificial intelligence solutions. A recent auditor general’s report highlighted that CRA human agents provide accurate tax information in less than 20% of calls, raising concerns about the foundational data for AI systems.
OTTAWA – Artificial intelligence specialists are cautioning the Canada Revenue Agency (CRA) against a premature pivot to AI-driven solutions for taxpayer assistance, advocating instead for a fundamental overhaul of its existing human response systems. This advice comes in the wake of a critical federal auditor general’s report, released on Tuesday, October 25, 2025, which revealed significant shortcomings in the CRA’s current service delivery.
The report, authored by federal auditor general Karen Hogan, painted a stark picture of the CRA’s call centre performance. It found that the agency not only struggled to answer a substantial portion of incoming calls but, more alarmingly, when live agents did connect with callers regarding personal income taxes, accurate information was provided in fewer than one in five instances.
Despite these challenges, the CRA is reportedly moving forward with AI initiatives, including piloting a generative AI version and expanding the operational hours and question-answering capabilities of its existing chatbots.
However, experts like Anatoliy Gruzd, the Canada Research Chair in Privacy Preserving Digital Technologies, argue that this approach carries considerable risk. Gruzd emphasized the critical need for the CRA to ‘get it right with the people’ before relying on AI. He explained that the effectiveness of chatbots is inherently dependent on the accuracy and quality of data derived from human interactions. ‘If you’re a government agent, and you’re launching a chatbot, you want to make sure that you have the process nailed down with the human agent,’ Gruzd stated, adding, ‘I think going back to the basics and trying to figure out why human agents get some of the answers wrong will be critical before launching any major AI initiatives.’
Adegboyega Ojo, a Carleton University professor specializing in the governance of AI, echoed these sentiments while also highlighting the potential benefits of AI when implemented thoughtfully. Ojo, whose research focuses on technical mechanisms for safe and beneficial AI use in government, suggested a hybrid human-AI model. This approach would leverage machines to efficiently handle frequently asked questions, thereby easing the burden of high call volumes on CRA staff, while reserving more complex and nuanced inquiries for human specialists.
Also Read:
- Navigating the AI Landscape: Why Businesses Must Diversify Beyond Single Models and Avoid Over-Reliance
- Ensuring the Secure and Reliable Deployment of AI Agents: A BCG Perspective
Ojo stressed that while AI can significantly enhance efficiency, human workers must remain central to the process and should not be replaced outright. He warned that even the most sophisticated chatbots, capable of advanced planning, can ‘get tripped up because they miss the context.’ He concluded, ‘I think the ideal way to deploy an AI system is to always have the human oversight, and that’s where the accountability comes in.’ The consensus among experts is clear: a robust, accurate human foundation is paramount for any successful and accountable AI integration within the Canada Revenue Agency.
