A while ago, I spent half a day at a client's financial shared services center. Twelve people, one week before the month-end close, two rows of invoice folders on the desks. An accountant who had been doing the job for eight years said something I still remember: 'What I fear most isn't having a lot of invoices; it's that the invoices all look the same.'
She was talking about three types of receipts: two accommodation invoices from the same hotel, two delivery notes from the same supplier for the same amount, and an employee submitting a taxi receipt that had already been reimbursed last month one more time. Human eyes have to distinguish them by date, license plate, and trip itinerary, and staring at them for too long makes the eyes blur. In half a day, they processed more than four hundred receipts, relying on three people's experience plus one scanner.
On November 11, the Ministry of Finance held a press conference on the high-quality development of accounting during the "15th Five-Year Plan" period. Lin Qiyun of the Accounting Department said at the meeting in very plain terms: more invoices, passenger tickets, and medical fee receipts should be given a unique "electronic ID," which would both open up the "fast lane" for reimbursement and make the use of funds traceable throughout the entire process and automatically verifiable, thereby curbing duplicate reimbursements and false accounting entries at the source.
The same 'Accounting Reform and Development "15th Five-Year Plan"' was already issued on October 23 (Caihui [2026] No. 13), and Chapter 7 is entirely about accounting digital intelligence. I read that chapter four times, and the more I read it, the more I felt: the real readers of this document are not CFOs, but the people currently working on ERP projects.
1. Those few sentences in Chapter 7 are more worth reading than interpretation articles
Policy documents are often excerpted by the media into a few slogans; what is easily overlooked are the verbs and objects. This chapter has four sentences, and the verbs are all very specific:
- “Formulate and implement accounting data standards covering the input, processing, and output stages of accounting information systems.” Note that there are three stages, not just the report submitted externally. Incoming original documents, intermediate processing rules, and outgoing reports—all three must have standards.
- “Continuously promote the application of electronic voucher accounting data standards, and steadily advance paperless, automated processing across the entire process of receiving, reimbursing, booking, and archiving electronic accounting vouchers.” The four stages are written as a single sequence; the meaning is clear: merely scanning and archiving does not count as completing the process.
- “Revise the General Taxonomy of Accounting Standards for Business Enterprises and explore promoting the establishment of a single-source system for corporate financial reporting.”This sentence is the most easily underestimated. “Single source” does not mean “using only one system”; rather, the figures in financial statements must be traceable to the same set of books and the same batch of vouchers.
- “Research and formulate measures for evaluating accounting digital intelligence, and organize pilot evaluation work.” There is another item on the management accounting side: formulate measures for evaluating management accounting application and organize evaluation pilots. Two sets of measures, two pilots.
The last sentence determines the fate of the first four sentences. If there is an evaluation method, there will be a scorecard; if there is a scorecard, someone will ask, “How far does your paperless coverage extend?”, “How many duplicate reimbursements does automated verification stop in a year?”, and “Can this report be traced back to the original vouchers?” These questions can only be answered with evidence the system accumulates in normal operation; they cannot be patched together at the last minute.
A judgment: Over the next two to three years, the dividing line in accounting informatization will no longer be “whether a system has been deployed,” but “whether the system contains a verifiable structure.” The same invoice is a record at Company A, but at Company B it is an image plus five manually typed cells. The former can be handed to an AI agent, while the latter can only be handed to a person.
II. Three stages, and where each one gets stuck
If you break “input, processing, output” apart and examine them, you will find that the failure modes at each stage are completely different, and the focus of investment is different as well.
| segment | Things to be standardized | Common real states | What happens if it's handed over to the agent? |
|---|---|---|---|
| Input | Invoice identifier, source, amount, tax amount, issuer, and corresponding business order number | Scanned documents + manual entry, with no validation relationship between fields | After reading the image, the model fills in missing items on its own; if it fills them in incorrectly, no error is reported. |
| Process | Account mapping, tax amount calculation, approval rules, related documents | Rules are scattered in people's experience and in Excel | Each judgment is not entirely the same, so it cannot be reviewed afterward |
| Output | Reporting basis, consolidation rules, and note value retrieval | Multiple algorithms for the same metric, each maintained separately | The answer you get when you ask "How much is the revenue?" depends on whom you ask. |
The input stage is the most critical, because it determines whether the later steps can be automated. Whether an invoice is recognized accurately does not depend on how powerful the model is, but on whether the invoice contains anchors that a machine can match: invoice number, issue date, amount, tax amount, and seller tax ID. If these anchors are complete, even if one field is recognized incorrectly, cross-validation can catch it; if the anchors are incomplete, the model can only guess based on semantics.
The problem in handling this step is not technology but ownership. Account mapping, tax amount calculation, approval paths—these rules usually live in the heads of veteran accountants, and newcomers rely on asking. To automate, you must first write them into tables: which business types go to which accounts, which amounts require which levels of approval, and which situations are exceptions. There is no shortcut for this; you can only sit down and check them one by one.
III. The four stages of electronic vouchers, each with its own pitfalls
The eight characters “receiving, reimbursement, posting, archiving” correspond to four completely different engineering efforts in the project. I’ll list them according to the projects I’ve done, and the order is the actual order.
| segment | What this section really needs to do | The most common breakpoints |
|---|---|---|
| Receive | Receive invoices from email, mobile, supplier portal, and tax authority, and complete authenticity verification and duplicate checking upon entry into the system | Authenticity verification is only done at the reimbursement stage; duplicate invoices have already come in. |
| Reimbursement | Receipts and expense applications are linked to budgets and projects and go through approval. | The invoice and the order do not match; rely on the remark text for explanation. |
| Post | Generate vouchers, with debit/credit accounts, tax amounts, and counterparties all in place at once | Vouchers are generated by aggregation—one voucher corresponds to dozens of invoices, so there's no way to trace it back. |
| Archive | Vouchers, documents, attachments, and approval records are packaged and retained, and can be retrieved by retention period. | What is archived are scanned copies, not structured data. |
The breakpoint in the third paragraph deserves separate discussion. To save trouble, many companies generate a single voucher at month-end summarizing expenses by type: the total travel expenses, the total business entertainment expenses. This keeps the books balanced and the reports correct, but whenever an audit asks "which invoices correspond to this travel expense," they can only go back and search through Excel. Summary vouchers were efficiency in the manual era and are a black hole in the automation era.
The fourth paragraph is most easily mistaken for IT work. The reality is: archiving determines whether you can prove yourself over the next five to ten years. When the tax bureau investigates, auditors investigate, or internal staff investigate, what they ask is not “whether you have stored it,” but “whether you can find the few records I need by invoice number, by supplier, by project, and by time, and prove that they have not been altered.”
IV. What “single source” really needs to change is the source of reports
In the plan, the sentence “explore promoting the establishment of a single-source system for corporate financial reporting”—I understand the key point to be the word “source.”
How are monthly reports produced in most companies today? Finance exports one from the general ledger, operations exports one from the order system, sales exports one from the CRM, the three are aligned in Excel, and discrepancies are discussed in meetings. In this process, reports are actually "produced" rather than "queried." Producing one takes three days, and changing a single metric definition means starting over.
The desired state of a single source is: reports are a view of the ledger. Changing the measurement basis changes the definition of the view, not re-assembling the data. To achieve this, the prerequisite is that the ledger itself is granular and complete enough: vouchers carry business document numbers, journal entries carry cost centers, counterparties carry customer and supplier master data, and attachments carry original vouchers.
The technical barrier to this is lower than one might imagine, while the organizational barrier is higher than one might imagine. Because it requires the business departments to accept one thing: the fields on a document are not for finance to read, but for the system to read. If one project number is left blank, a piece of some report will be missing in the future.
V. On the same day, MIIT set the same thing for the manufacturing floor
On November 16, the Ministry of Industry and Information Technology and the State Administration for Market Regulation issued the "Implementation Plan for Using Standards to Lead the Optimization and Upgrading of the Light Industry." The production-side section almost copied the pain points of integration projects one by one: focusing on typical scenarios such as digital R&D and design, flexible production and manufacturing, preventive equipment maintenance, online quality monitoring, and supply chain collaborative management, it is researching and formulating basic common standards such as data interfaces and communication protocols, key technical standards such as intelligent control and cloud-edge collaboration, as well as standards for application in segmented industries and effectiveness evaluation; at the same time, it is promoting the current standards for digital workshops, smart factories, and smart parks, and cultivating standard application pilots.
Looking at this plan together with that planning document from the Ministry of Finance, the logic is the same: for data to flow across systems, both sides must understand the same field in the same way. The Ministry of Finance governs how money is recorded, while the Ministry of Industry and Information Technology governs how goods are manufactured; the standards on both sides point to the same thing—interfaces and data dictionaries must be understandable to third parties.
In practical work, the thing that can be changed most immediately is one document: the interface specification. In many projects today, interface documentation is written by developers for developers to read, with only field names and types, no units, no value ranges, no failure retry rules, and no reconciliation criteria. Three years later, no one can understand this document; five years later, no one dares to touch it, and eventually the entire integration becomes a part that cannot be touched.
A reminder: Standard implementation often has a transition period, so don’t treat “standard release” as “must change tomorrow.” What you should really start doing now is completing your own data dictionary—no matter how external standards are set, a field description that people can understand is a plus under any standard.
VI. Landing in Odoo: The four stages need to be connected; what's missing is a few threads
Using Odoo to do all this, the good news is that the skeleton is already there; the bad news is that you have to thread the wires yourself.
What is the skeleton? Invoices and vouchers are two types of records in the same table (account.move's move_type distinguishes customer invoices, vendor bills, and journal entries), attachments are attached to documents (ir.attachment has res_model and res_id), approvals go through discussions and activities, archiving can go through the documents module, and reports in Enterprise are the account.report suite. None of this requires development.
What is the thread? It is what lets these four sections line up with each other. For the points most often filled in in the project, listing them in a table is clearer:
| Things that need to be supplemented | Where it lands | What happens if it is not replenished? |
|---|---|---|
| Invoice identification fields (code, number, check code, invoice date) | Custom fields or localized fields on account.move | Duplicate checking can only be based on amount plus date, so the collision rate is high. |
| Business order number backfill | Source document reference on voucher line (purchase order, sales order, expense document) | Vouchers are decoupled from business, and traceability relies on people. |
| Attachments and vouchers one-to-one | The res_model / res_id of ir.attachment point to the voucher or document | Attachments pile up in folders, and it's impossible to find their correspondence |
| Generate per ticket or in aggregate | Voucher generation strategy (recommended to generate per document whenever possible) | One voucher corresponds to dozens of invoices, making sampling impossible during audit |
| Archive Retention and Read-Only | After archiving, modification is prohibited; the retention period is configured according to policy. | Archived items are still editable; “it was never changed” proves nothing. |
| Change Audit Trail | Enable tracking on key fields, or introduce an audit log module | There is no way to find out who changed it or when. |
The third and fourth rows in the table are the two areas where I have seen the most controversy. In the third row, many implementations, in order to keep the "interface clean," attach attachments to the expense request rather than to the voucher, so when finance audits the accounts they have to jump through two layers. The fourth row is even more typical: finance is used to doing the books in a month-end summary; changing to per-document processing will produce dozens of times more journal entries, and month-end closing will take longer. This trade-off must be discussed in advance, otherwise there will definitely be rework after go-live.
One more point worth mentioning separately: multiple companies and multiple currencies. For group enterprises implementing a single source, the hardest part is often not technology, but intra-group transactions and translation exchange rates. For the same internal sale, the seller records revenue and the buyer records cost; where the elimination rules are written and which day's exchange rate is used—if these are not settled, the consolidated statements will never balance. It is recommended to run this part with one month of historical data before go-live, rather than waiting until the annual audit.
VII. Before the evaluation method arrives, first let the system prove itself
The plan says that “Measures for the Evaluation of Accounting Digital Intelligence” will be issued, and pilot programs will also be carried out. No one knows yet what the specific indicators are, but there are several types of evidence that any evaluation will require, so we can prepare them in advance:
- Coverage rate. What percentage of vouchers carry original electronic vouchers, what percentage have undergone automatic verification, and how many business types have achieved full-process paperless operations.
- Interception volume. How many duplicate reimbursements, how many verification failures, and how many amount-invoice mismatches were intercepted in a year. This number best shows that the system is working, not just storing files.
- Traceability. Pick a random figure from a report: can you trace it to the voucher within three clicks, and then to the original receipt? This is the toughest metric, and also the hardest to fake on the spot.
- Timeliness. On average, how many hours from receipt of the invoice to posting, and how many days for month-end closing. The business department also cares about these two numbers, making it easy to gain support.
These four types of evidence share one thing in common: they all require the system to record during normal operation, rather than tallying after the fact. So the action that really should be taken is to design these counts and traces into the process—each stage has a status field, each interception has a log, each archiving has a timestamp. The development effort is not large, but someone needs to think of it at the requirements stage.
8. Closing with Three Sentences
First, the value of a voucher lies not in “being stored,” but in “matching up.”A scanned copy solves the space problem, but not the verification problem. Whether a ticket can be machine-verified, traced, and sampled depends on whether it carried those anchors with it when it entered the system.
Second, standards are given externally, while the data dictionary is accumulated by ourselves. Standards from the Ministry of Finance and protocols from the Ministry of Industry and Information Technology will all have transition periods. But these tasks—interface specifications, field definitions, and business order number backfilling—if not done today, will still need to be done when the standards arrive tomorrow, and with even greater urgency.
Third, pick one line and get it running end to end first; don’t spread out.Choose one business type (for example, travel expenses, or a certain supplier’s purchase invoices), and connect all four stages—receiving, reimbursement, booking, and archiving—end to end. Use three months of data to see whether the interception rate and traceability can be achieved. Once it works, replicate it; that is much faster than rolling out ten types at once.
Return to what that accountant said, "the tickets look the same." What she was actually afraid of wasn't the tickets, but that the system was incapable of distinguishing two tickets that looked identical. This is something people can't do well, and machines can't do well either—unless there is an ID on the ticket that people cannot see but machines can.
The policy document has made this direction clear; the remaining work is up to the enterprises themselves: turn vouchers into records, connect records into a chain, and keep the chain where it can be audited.
Reference sources: Ministry of Finance, 'Notice on Issuing the “Accounting Reform and Development 15th Five-Year Plan”' (Cai Kuai [2026] No. 13, issued on 2026-10-23; released on the website of the Accounting Department of the Ministry of Finance on 2026-11-11), verbatim transcript of the press conference on high-quality accounting development during the “15th Five-Year Plan” period (2026-11-11); Ministry of Industry and Information Technology and State Administration for Market Regulation, 'Implementation Plan for Leading the Optimization and Upgrading of the Light Industry through Standard Improvement' (reported by Cailianshe on 2026-11-16). Odoo objects and fields in this article are based on Odoo 19/20; specific implementation must be assessed in conjunction with localization modules and custom fields.
Zhongdi Information · Enterprise Management Digitalization Observations · The views expressed are solely those of the author; the interpretation and implementation of policies shall be subject to the official documents of the competent authorities.
