Preparing for a China Export Tax Audit: What Your ERP Must Be Able to Produce

What an audit is actually testing

Export rebate audits are frequently misunderstood as a check on arithmetic — did you compute the refundable amount correctly. In practice the arithmetic is the least interesting part, because it is mechanical and the tax system recomputes it independently. What the audit tests is whether the transaction described in your refund claim actually happened as described. Was there really a buyer abroad. Did the goods really leave. Did the money really come back. Was the input VAT really incurred on those goods.

That framing changes what preparation means. You are not preparing an explanation of your calculations. You are preparing to demonstrate, per shipment, that four independent record sets from four different sources describe one coherent event. Companies that fail audits rarely fail because they miscalculated; they fail because the four record sets cannot be made to agree, or because the evidence for one leg was never collected at the time and cannot be reconstructed now.

The four-way match

The core of an export rebate file is a four-way match. Each leg originates outside the ERP or in a different module, which is exactly why the match is hard.

• Goods flow — evidence that physical goods moved: the customs declaration, the transport document, the packing list, and warehouse outbound records. The declared quantity and description must correspond to what was actually shipped.
• Document flow — the commercial invoice, the contract or order, and the certificate of origin where applicable. Descriptions, values and trade terms must be consistent with the declaration.
• Fund flow — the bank receipt evidencing settlement, matched to the receivable and to the declared value. Persistent shortfalls invite the question of whether the transaction was at arm's length.
• Invoice flow — the input VAT invoices from domestic suppliers or, for a producer, the input VAT on materials and services consumed. These must correspond to the goods exported, in description, quantity and value.

A well-prepared file presents these four legs together, per shipment, with the reconciliation already done. A poorly prepared file presents them in four folders, in four numbering conventions, maintained by three departments, and asks the auditor to do the matching. The auditor's conclusion in the second case is not necessarily adverse — but the review takes weeks instead of days, every inconsistency becomes a question, and the burden of proof has effectively shifted to you.

The evidence that must exist at the time, not later

Some evidence can be regenerated. Some cannot. The distinction is the single most useful thing a finance team can learn about audit preparation, because only the second category needs preventive design.

Regenerable: the refund computation, the summary schedules, the declaration listings, the ageing analysis — all reproducible from retained transaction data. Non-regenerable: the signed contract as agreed, the transport document as issued, the supplier invoice as received, the bank receipt as recorded, the customs declaration as released, and any correspondence about a discrepancy. If these were never captured, no amount of system work later will produce them, because the counterparties are not obliged to reconstruct history for you.

This is why document retention is a design requirement rather than an administrative habit. The regulatory expectation is that supporting documents for export rebate claims are retained for the prescribed period and produced on request. The practical implication is that the ERP must store the documents linked to the transaction, not merely the transaction amounts — because a schedule of numbers without the underlying documents is an assertion, not evidence.

The red flags auditors are trained to look for

Knowing what draws attention is not about hiding anything; it is about recognising where your own process is genuinely weak. These are the patterns that turn a routine review into a deep one.

• Descriptions that differ across the supplier invoice, the customs declaration and the export invoice — even when the goods are the same, inconsistent naming suggests the documents were not produced from one source.
• Concentration in a small number of trading companies with no clear commercial rationale, or buyers whose identities cannot be independently verified.
• Settlement patterns that do not match the declared terms — persistent under-collection, third-party payments, or receipts from entities unrelated to the buyer.
• Input invoices that predate or postdate the export by a wide margin without explanation, or invoices for goods that do not appear in the declared export mix.
• Sudden changes in product mix or destination that are not reflected in the business's other records.
• Refund claims that repeatedly sit just under a threshold, or filing that consistently happens at the deadline.

Each of these is a signal about data integrity, not about intent. The first is the most common and the most fixable: if product descriptions are generated from a single product master and flow into every document, cross-document inconsistency becomes structurally impossible rather than a matter of care. That is the same alignment issue discussed in export invoice versus VAT invoice in China.

Retention that is actually retrievable

Retention and retrievability are different problems. Almost every company retains; few can retrieve. The test is not whether a document exists somewhere, but whether it can be produced for a specified shipment within a working day. A retention policy that requires searching a shared drive by customer name and guessing at a filename does not meet that test in practice.

The design that works is straightforward: every document in the retention scope is linked to the transaction it supports, with the transaction as the index. Then a request for 'all documentation for shipment X' is a query, and a request for 'all shipments in Q2 with missing input invoices' is also a query. The second is more valuable than the first, because it finds problems before an auditor does. Building that list — shipments with an incomplete evidence chain — is the single highest-value audit preparation activity available to an exporter, and it is only possible if the links exist.

How a review usually starts

A rebate review rarely begins with a formal notice asking for everything. It typically begins with a single request: the documentation for one shipment, or for one supplier, or for a defined period that happens to be under scrutiny. The pattern matters because the first response sets the tone. If the requested file arrives complete, reconciled and clearly indexed, the review is likely to remain narrow. If it arrives as four documents in three formats with an apology about the rest being somewhere, the request expands — not out of hostility, but because an incomplete file is a reasonable basis for asking for more.

This is why the operational metric is time-to-produce rather than completeness-in-principle. A company that can assemble a full documentation pack for a named shipment within a working day is in a fundamentally different position from one that can do it eventually. It is worth measuring this quarterly on a randomly selected shipment, because it degrades silently: the person who knew where the transport documents were gets promoted, and nobody notices the process was dependent on them until it is tested.

The boundary problem: shipments that straddle periods

Export documentation rarely aligns with accounting periods. Goods can be declared in one month and shipped in the next, invoiced after the declaration, settled two months later, and claimed for rebate in a filing period that falls in a different quarter again. Each of those dates lives in a different system or a different module, and the rebate claim belongs to a specific period determined by rules that do not care about your reporting calendar.

The consequence is that a review which picks a period will always find shipments whose documents point to a different one, and the explanation for each is legitimate but requires the evidence to exist. Systems that store one date per shipment collapse this into confusion. Systems that store declaration date, export date, invoice date and settlement date separately can explain the boundary immediately. When you are selecting software, this is a cheap question with a revealing answer: ask how many dates the shipment record holds, and whether each is independently reportable.

People as evidence

The least discussed element of audit readiness is the human one. Auditors ask who prepared a declaration, who approved an amendment, who authorised a document release, and what the process was supposed to be. Without records, answers become recollections, and recollections under pressure are inconsistent in a way that invites further questions. The organisation is not concealing anything; it simply has no memory beyond the individuals in the room.

User-level audit trails solve part of this and are routinely under-configured. Who changed the rebate rate on a product, and when. Who edited the supplier invoice amount after it was recorded. Who reissued the export invoice and with what reason. These are not surveillance questions; they are the same questions a customs or tax reviewer will ask, and having the answer pre-built is the difference between a short verification and a long investigation.

There is a second-order benefit that finance teams rarely anticipate. When the audit trail exists, process improvement becomes possible, because defects can be traced to a step rather than to a person. A rate that was changed twice in one quarter points to a catalogue update problem. Invoices reissued repeatedly for one customer points to a master data problem with that customer's requirements. Without the trail, the same defects recur and are attributed to carelessness, which is rarely the actual cause.

Sampling: how shipments get selected

A reviewer cannot examine every shipment, so a sample is drawn — and the way a sample is drawn tells you what a company's data looks like before anyone reads a single document. If the population is a clean list of declarations with dates, values, destinations and HS codes, the sample is random and the exercise is contained. If the population has to be assembled by combining a customs export file with an internal sales report and reconciling the difference first, the reviewer has already learned something, and the reconciliation itself becomes part of the review.

This is why the practical objective is not a perfect file for the shipments you happen to care about, but a complete and queryable population of all shipments. Completeness is testable: does the total declared export value reconcile to the accounting revenue for the period, with differences explained by timing? If yes, any sample drawn from the population is answerable. If no, the unreconciled difference is the first question and the sample becomes secondary. Most companies can produce a plausible total; far fewer can explain the variance between customs value and booked revenue without a meeting.

A second selection pattern is risk-based rather than random: unusual destinations, newly added suppliers, a product category whose rates changed, or a concentration of refunds just below a threshold. Risk-based selection is where data quality becomes consequential, because the shipments chosen are precisely the ones that were unusual, and unusual shipments are the ones most likely to have an incomplete document trail. The defence is the same in every case — a link from the shipment to its full evidence set — and it is the only defence that scales.

There is a documentation angle to the human question as well. Process descriptions, approval matrices and role definitions are themselves evidence, because they show what was supposed to happen and who was responsible. A company with no written process can only argue that its people acted reasonably; a company with a documented process can show whether they did. This is the cheapest item on the entire audit readiness list — writing down what already happens — and it is the one most consistently deferred until it is needed.

What to build, in priority order

For a company preparing seriously, this is the sequence that produces the most protection for the least effort.

• Link every shipment to its customs declaration, transport document, commercial invoice and settlement receipt, at the moment each becomes available.
• Link input VAT invoices to the shipments or production orders they support, and maintain an outstanding list so missing evidence is visible weekly rather than annually.
• Store the issued version of every document as an immutable snapshot, so a later correction does not retroactively alter what was filed.
• Generate a per-shipment evidence completeness report and review it as an operating metric, not as an annual exercise.
• Keep the audit trail of who entered what and when, because the question of provenance arises in every serious review.

The sequencing matters because each step depends on the one before it. A completeness report is meaningless if documents are not linked to shipments, and linking is meaningless if the shipment record does not carry enough identifiers to match against what arrives from the broker, the carrier and the bank. Companies that attempt the report first usually end up rebuilding the links by hand, conclude the exercise is impractical, and return to the annual scramble.

The companies that come through audits comfortably are not the ones with the best accountants. They are the ones whose ordinary operating process produces evidence as a by-product, so that the audit is a matter of extracting what already exists. That property is a systems property, and it is worth more than any amount of retrospective documentation — because retrospective documentation is exactly what an auditor is trained to distrust.

The mechanism being audited is covered in export tax rebate ERP software for China, the settlement leg in FX settlement and reporting for Chinese exporters, and the document generation side in automating export documentation. If you are selecting a system and audit readiness is a criterion, ask the vendor to show the completeness report rather than describing it.