Could AI’s greatest value to tax advisers lie in challenging arguments they have already dismissed?
Four senior barristers said a tax argument was unworkable. An AI model disagreed. The taxpayer won at the tribunal. Whose argument do you think they used?
Earlier this month, a taxpayer won his land tax case against an Australian tax authority, the Queensland Revenue Office (Dempsey v The Commissioner of State Revenue [2026] QCAT 424). The facts of the case were simple, but didn’t fit neatly into the tax law. The taxpayer and his wife demolished their home to rebuild it and lived elsewhere while the work was carried out. The tax authority said the property didn’t qualify for the principal place of residence exemption because the owners didn’t physically occupy the property during the year.
The taxpayer’s lawyers had been told by senior barristers (plural!) that this case was unwinnable, because the definition of ‘usage’ of a principal residence in the land tax statute was well established. The lawyers asked their in-house AI model to have another look. The AI found a creative way through the legislation, that a principal residence is defined by more than just where you live. It’s also where you leave your things, where you register to vote and whether you go and find another permanent residence. The tribunal agreed with the taxpayer. Success!
AI doesn’t know which arguments have been deemed by the profession to be impossible, nor does it know how much deference it’s supposed to have for senior members of a profession (unless you tell it). It just keeps trying arguments, again and again, without any need for professional consensus or approval of others.
Lots of people lambast AI for reinforcing what the user wants to hear. Sometimes, the more useful thing it can do is challenge an assumption or conclusion that you’ve already accepted.
If you’re an AI sceptic, I suspect this is a far more compelling use case than time and cost efficiencies.
If you’re a keen adopter of AI, how would you feel if AI told you that your merits analysis had missed a trick?
Could AI’s greatest value to tax advisers lie in challenging arguments they have already dismissed?
Four senior barristers said a tax argument was unworkable. An AI model disagreed. The taxpayer won at the tribunal. Whose argument do you think they used?
Earlier this month, a taxpayer won his land tax case against an Australian tax authority, the Queensland Revenue Office (Dempsey v The Commissioner of State Revenue [2026] QCAT 424). The facts of the case were simple, but didn’t fit neatly into the tax law. The taxpayer and his wife demolished their home to rebuild it and lived elsewhere while the work was carried out. The tax authority said the property didn’t qualify for the principal place of residence exemption because the owners didn’t physically occupy the property during the year.
The taxpayer’s lawyers had been told by senior barristers (plural!) that this case was unwinnable, because the definition of ‘usage’ of a principal residence in the land tax statute was well established. The lawyers asked their in-house AI model to have another look. The AI found a creative way through the legislation, that a principal residence is defined by more than just where you live. It’s also where you leave your things, where you register to vote and whether you go and find another permanent residence. The tribunal agreed with the taxpayer. Success!
AI doesn’t know which arguments have been deemed by the profession to be impossible, nor does it know how much deference it’s supposed to have for senior members of a profession (unless you tell it). It just keeps trying arguments, again and again, without any need for professional consensus or approval of others.
Lots of people lambast AI for reinforcing what the user wants to hear. Sometimes, the more useful thing it can do is challenge an assumption or conclusion that you’ve already accepted.
If you’re an AI sceptic, I suspect this is a far more compelling use case than time and cost efficiencies.
If you’re a keen adopter of AI, how would you feel if AI told you that your merits analysis had missed a trick?
