The Revenue Department's AI Audit Is Already Running: What Thai Accounting Firms Need to Know

For most of the history of Thai tax administration, audit selection was a human process. A Revenue Department officer reviewed a return, noticed something that did not fit, and opened an inquiry. The indicators that prompted attention were roughly predictable: unusually high expense ratios, significant revenue drops without explanation, round numbers in categories that rarely produce round numbers. A firm that kept plausible records and avoided glaring anomalies could manage its clients’ audit exposure with reasonable confidence.

That model has changed. The Revenue Department’s Risk-Based Audit system uses machine learning to score taxpayers for audit risk based on the consistency of their declared figures with data the Revenue Department receives from other sources. The system does not wait for a human auditor to notice a discrepancy. It finds them automatically, at scale, across the entire registered taxpayer population.

For accounting firms advising Thai businesses, this is not a future development to monitor. It is the current operating environment. Understanding how the system works, what it flags, and what professional exposure looks like when a client is selected changes how a competent firm designs its client workflows.

What the System Cross-References

The Revenue Department’s data position has expanded significantly over the past decade. e-Tax invoices submitted through the Revenue Department’s electronic system create a record of every declared commercial transaction between VAT-registered parties. Bank transaction reports filed by financial institutions under the Financial Institutions Business Act add another data stream. Social security contribution records from the Social Security Office indicate payroll levels. Customs data from the Department of Customs covers import and export values for businesses with international trade exposure.

The AI audit system cross-references these streams against the figures a taxpayer declared on their returns. If a business declared ฿2 million in expenses for a particular supplier category, but the supplier’s e-Tax invoice records show ฿200,000 in invoices issued to that taxpayer in the same period, the gap is not invisible. If declared payroll is inconsistent with social security contributions for the same workforce at the same time, the inconsistency is detectable without a human auditor examining either filing. If revenue declared for VAT purposes does not match revenue declared for corporate income tax, the discrepancy appears automatically.

None of this requires the Revenue Department to audit a specific business proactively. The cross-referencing runs continuously. Inconsistencies generate a risk score. Higher-scoring returns surface for review. The selection is no longer primarily intuitive; it is data-driven.

Why Old Heuristics No Longer Apply

Accounting firms that helped clients manage audit risk under the previous model developed working heuristics: avoid certain ratios, keep expense categories within expected ranges, maintain consistency across years. Some of these heuristics remain relevant because the underlying data relationships they tracked are the same ones the AI system now monitors. Others have become less reliable because the system detects patterns that human auditors were unlikely to notice unless they were already looking.

Volume triggers are one example. A human auditor working through a stack of returns prioritizes by perceived anomaly or random selection. The AI system applies consistent scrutiny to every return in scope without fatigue or prioritization bias. A return that would have received no attention simply because it was one of many now receives the same algorithmic review as every other.

Round numbers across multiple categories are another. A human auditor might note round numbers as suggestive of estimation rather than recording. The AI system can flag returns where expense estimates are internally consistent with each other in ways that actual transaction data rarely produces, particularly where the estimated figures diverge from the supplier-side e-Tax invoice data.

Lump-sum expense methods remain legally valid. The 60% deduction standard for service income and the 85% standard for certain business types are authorized by law and have not been withdrawn. But the AI system can now compare lump-sum deductions against the third-party data available for the same taxpayer. A taxpayer claiming the maximum lump-sum deduction for categories where supplier-side invoices suggest actual documented expenses are materially lower is presenting a profile that the system can identify as potentially inconsistent with available evidence, even if the deduction itself is legally permitted.

The Professional Exposure for Accounting Firms

When a client is selected for an AI-triggered audit, the inquiry is not limited to the client’s records. The accounting firm that prepared the returns and managed the client’s bookkeeping is visible. The Revenue Department can request explanation for accounting decisions. In cases where the firm’s work contributed to a material discrepancy, the firm faces reputational and financial exposure beyond the client’s tax liability.

The professional exposure is not hypothetical. An accounting firm whose client has inconsistency between declared supplier expenses and e-Tax invoice data must be able to explain why. If the explanation is that expenses were estimated using the lump-sum method and the firm’s file shows the decision was made deliberately and documented, the explanation is defensible. If the explanation is that the figures were entered without checking the underlying invoice data because the client did not provide it, the firm’s position is weaker.

The distinction matters because the AI audit system does not distinguish between a firm that made a defensible professional judgment and a firm that did not engage with the available data. The risk score is generated from the return, not from the process that produced it. The process that produced it becomes relevant when the inquiry arrives. That is when the firm’s matter record determines whether it can account for what it did and why.

Audit-Ready Records: A Workflow Design Question

The response to AI-driven audit risk is not crisis management. It is workflow design. The time to build audit-ready client records is during the engagement year, as transactions occur and decisions are made. Three weeks before a Revenue Department inquiry is not when a firm can reconstruct the rationale for an expense classification or locate the supporting invoice for a claimed deduction.

Audit-readiness for a client engagement means the firm’s file contains the documents that substantiate the declared figures, the accounting decisions are recorded with their rationale, and the correspondence with the client about the filing is dated and retrievable. This is not a higher standard than professional practice already requires. It is the standard that becomes critically important when an inquiry arrives and the firm must demonstrate it was met.

The specific risk areas the AI system has made more visible are the ones where lump-sum estimation diverges from documented transaction data. Firms that systematically collect client invoices, match them against declared figures, and note where lump-sum methods were chosen in place of documented actuals are in a materially better position than firms that rely on client-provided summaries without corroboration. The former can explain their work. The latter may not be able to.

For larger clients whose transaction volumes make invoice-by-invoice review impractical, the question shifts to sampling methodology and documentation of the sampling approach. A documented, consistent sampling methodology is defensible under inquiry. An undocumented approach that produces figures inconsistent with e-Tax invoice data is not.

The Client Conversation

Many clients do not know that the Revenue Department has this level of data integration. A client who has been comfortable with estimated expense figures because previous audits were infrequent may not understand that the data environment has changed and their filing profile now looks different to the system reviewing it.

The advisory value of explaining this is real. A client who understands that lump-sum deductions are being compared against supplier invoice data is in a position to decide whether to collect invoices more systematically going forward, or to accept that their current approach carries higher audit risk in the current environment. Either decision is valid; it should be an informed one. A firm that raises this with clients is performing a service that protects both the client and itself. A firm that does not is leaving both parties exposed to a risk neither may understand until an inquiry arrives.

The conversation is also an opportunity to review historical filing positions. If a client’s declared expenses in prior years are significantly inconsistent with what third-party data would show, the question of whether to file amended returns, or whether the firm can document a rationale for the prior positions, is better addressed proactively than reactively.

FirmFlow and the Matter Record as Professional Defence

FirmFlow’s Document Analyser and matter record create a structured, dated archive of client documents, accounting decisions, and correspondence. If a client is selected for an AI-triggered audit, the firm’s matter record is the first line of professional defence: a documented, timestamped log of what was reviewed, what was decided, and why.

The value of a structured matter record is not theoretical under the current audit environment. When a Revenue Department inquiry arrives, the firm that can pull up a dated file showing when each client document was received, what was reviewed and classified, what accounting decisions were made with their supporting rationale, and what was communicated to the client is in an entirely different position from a firm reconstructing the engagement from email threads and memory.

The AI audit system has not changed what good professional practice requires. It has changed the frequency with which that practice is tested and the speed at which inconsistencies are found. Firms that have always maintained structured matter records are better positioned than they may realize. Firms that have not are now operating with a more visible form of risk than they faced in a manual audit environment. The audit is no longer a possibility that requires bad luck to materialize. It is a systematic process that surfaces inconsistencies across the entire taxpayer population. The question for every accounting firm is whether its client files can withstand the review when it arrives.