Tax, Regulatory Filings and Automated Preparation Risk
Tax returns and regulatory filings occupy a unique position in financial compliance. They are prepared according to strict rules, often produced under deadline pressure and subject to audit or examination. The temptation to automate tax preparation is strong: software can apply tax rules faster than humans and can process thousands of transactions. Yet tax preparation has higher stakes than most financial processes because errors can trigger penalties, corrections and audit risk.
Tax filings are high-risk applications for AI because tax law is rules-based but also contains judgment. A tax rule might state “capital expenditure is depreciated over the useful life of the asset.” This rule is deterministic: if something is capital and has a three-year life, it is depreciated over three years. Yet determining whether an outlay is capital or expense requires judgment. Is a £50,000 software licence over the assets entire useful life? Or is it an expense of the current year because the value dissipates quickly? Different companies and different tax authorities answer differently. An AI system can apply the rule once a judgment is made, but it cannot make the judgment itself.
Using AI to prepare tax filings therefore means using AI to speed up mechanical steps (data aggregation, calculation, report formatting) while reserving judgment to qualified tax professionals. This is the correct use of AI in tax. The incorrect use is to give an AI system a set of transactions and ask it to produce a completed tax return without human review of classification decisions.
Consider the specific risks. First, completeness: has the AI system captured all transactions? If your finance system does not export transactions to the tax preparation system correctly, the AI system might prepare a return based on incomplete data. This risk existed before AI (an intern could make the same export error) but is easier to miss when delegated to automation. Second, classification: have transactions been classified correctly?
A payment to a marketing agency might be deductible advertising or might be non-deductible, depending on what the agency did. An AI system cannot make this judgment reliably. Third, valuation: if you have assets that require valuation for tax purposes (inventory, property, uncertain liabilities), the AI system is unlikely to value them. This must remain with qualified staff. Fourth, regulatory change: tax rules change frequently.
If an AI system is trained on historical tax returns, it might not apply rules that changed after its training data ended.
The appropriate workflow for AI-assisted tax preparation looks like this. Finance team exports transactions and balance sheet data into the tax system. An AI engine (or a rule-based system) reads the data, applies basic classification rules and produces a draft return. Tax professionals review the draft, check classification decisions and adjust where necessary. They also review for completeness and for any regulatory changes since the system was last updated. Only after professional review is the return filed.
Documentation of this review process is essential. Keep records of: what data was provided to the AI system, what the system produced, who reviewed it, what changes were made and why. If a tax authority later challenges a return, this documentation shows that the return was prepared with appropriate professional judgment.
Specific risks in particular tax areas are worth noting. Transfer pricing (the price at which related entities sell to each other) is an area where tax authorities scrutinise closely. An AI system might underprice intercompany transactions based on historical patterns without considering whether those prices meet the arm’s length standard required by tax law. Transfer pricing is a domain where professional judgment is non-negotiable. Provisions (accruals for uncertain liabilities) are another high-risk area. A model might underestimate or overestimate a legal claim or environmental liability based on historical volatility, without understanding the specific facts of the case. Provisions require legal and factual judgment that AI cannot provide.
Tax evasion versus avoidance is also relevant. An AI system that is optimised to minimise tax without explicit constraint might recommend structures that are aggressive or that could be challenged by tax authorities. Your tax team must ensure that any AI system you use is operating within your company’s tax governance policies and risk tolerance. Some companies accept higher tax risk; others take a conservative approach. These preferences must be built into how the system operates.
Finally, audit trails. If a tax authority examines your return and questions a specific line item, you must be able to explain how that figure was derived. If it came from an AI system, you need to show the data input to the system, the system’s output and the review that was performed. This audit trail is your defence.