Few questions cause as much uncertainty in mid-size companies as whether their digital document storage would hold up in a tax audit. Invoices arrive as files in a shared mailbox, delivery notes are photographed, till receipts are scanned into a folder on the server, and accounting has long since been working on screen. What is usually missing is not the technology. It is the written order behind it. The German principles for the proper keeping and retention of books, records and documents in electronic form — GoBD for short — describe what is expected of such storage. They set out six principles: traceability and verifiability, completeness, accuracy, timely posting and recording, order and immutability. On top of that comes the requirement that the records stay machine-analysable. This article translates those requirements into concrete decisions about folder structure, access rights, logging and documentation. It does not replace tax advice; how the rules apply to your own case belongs in a conversation with your tax adviser.
Key takeaways
- The GoBD do not prescribe a particular product but a set of properties: a document must be stored completely, accurately, in good time, in an orderly, traceable and unalterable way, and must remain machine-analysable throughout the entire retention period.
- Immutability does not forbid corrections, it requires that every change remains recognisable as a change: the original version is kept, and the correction is logged with time, author and reason.
- A document that exists only as an image file does not meet the analysability requirement; structured attributes such as document number, date, amount and the link to the journal entry are needed as well.
- The process documentation is not a form but a description of the workflow as it is actually lived, in four parts — general description, user documentation, technical system documentation and operations documentation — and it is itself subject to retention.
- Anyone wanting to destroy paper after scanning needs a written scanning rule that is observed in daily practice; for documents that must be kept in the original, the paper stays in the cabinet regardless.
What the GoBD govern and what they do not
Behind the abbreviation stand the principles for the proper keeping and retention of books, records and documents in electronic form and for data access. They come as a letter from the Federal Ministry of Finance, which makes them an administrative instruction: they bind the tax administration and set out how it interprets the statutory requirements of the German Fiscal Code and the Commercial Code. For a business that means two things. First, there is no standard called GoBD against which a product could be certified. Second, the requirements are binding all the same, because they derive from the underlying legislation.
Everything relevant to taxation is affected: incoming and outgoing invoices, bank statements, till receipts, delivery notes with a billing reference, contracts, payroll and travel expense records, and the records of upstream systems. Electronic mail is included where it serves the function of a commercial or business letter or carries a document. The most common misconception is that only the accounting software is concerned. In fact the requirements apply to every upstream system in which tax-relevant data arises: point of sale, time recording, inventory management, order administration, electronic mileage logs.
What is not governed is the tool you use. Whether storage sits on a server in the building or with a service provider, whether a document management system is in place or a database with attached file storage, is a business decision. What cannot be shifted is responsibility: it stays with the taxpayer, even when bookkeeping is outsourced and the technology is run by a provider. Software can support the requirements, but it is the business that has to meet them.
Certificates are no substitute for a description
Immutability: the core of digital storage
Of all the requirements listed, immutability is the one on which organically grown storage most often fails. It means that once a record has been captured it must not be altered in a way that makes the original content impossible to establish. That is not a ban on corrections. An invoice captured incorrectly may be corrected — but the original version has to be preserved, and the correction has to be traceable with time, author and reason.
The gap to common practice is considerable. In many companies documents sit in a folder structure that grew over the years on a network drive. Anyone with access can replace, rename or delete a file without leaving a trace. That nobody does so is a matter of trust, but it is not evidence. This is exactly the distinction an audit looks at: not whether something was changed, but whether a change would be noticed at all. Storage is not unalterable because nobody alters it, but because an alteration cannot pass unnoticed.
Several technical routes lead to the goal, and none of them requires a large purchase. What matters is the combination of restricted rights, a log that records filing and changes, and a backup that cannot be overwritten from the same account. Leave out one of these three pillars and the requirement is not half met, it is not met at all.
- Separate write rights. The filing role may add but not overwrite or delete. Whoever holds broader administrative rights does not use the same account for day-to-day work.
- Write-once storage. A storage area that cannot be changed after writing solves the question at the root. Alternatively, storage that keeps every version in addition rather than replacing it will do.
- Versions instead of overwriting. If a document is replaced, a new version with its own identifier is created; the previous one stays readable and stays linked to the same posting.
- Keep a log. Who filed, changed, moved or invalidated which document and when. The log is itself subject to retention and must not be alterable after the fact.
- A checksum per file. A hash over the file content, calculated on filing and carried in the index, makes a later change detectable without any special system being required.
Completeness, accuracy and timely recording
Completeness means that every business transaction is recorded individually and none is missing — including the cancelled one, the one amended later, and the one where in the end nothing changed hands. Aggregate postings without a breakdown are the classic pitfall, as is the document that gets lost on its way between a department and accounting. In companies without a fixed handover point, that loss is the most common reason for incomplete storage (field experience).
Timely means that the gap between a transaction and its recording is not long enough to make the allocation uncertain. The guidance is more concrete than many expect: cash receipts and cash payments are to be recorded daily (German Fiscal Code), and for non-cash transactions recording within ten days is regarded as unobjectionable (GoBD). For finalising postings, the principles name the end of the following month as a point of reference (GoBD). Passing documents on in quarterly batches falls outside that frame.
Accuracy, finally, means the faithful representation of the actual transaction, not the impeccability of a system. A transposed digit is not a breach as long as it is corrected visibly. A workflow that makes errors invisible is. The overview below matches each requirement with the point where things typically go wrong and the measure that addresses it.
| Requirement | Where it usually goes wrong | What helps |
|---|---|---|
| Completeness | Documents stay in the department until somebody asks | A fixed handover with a date stamp and a number range without gaps |
| Timely recording | Batch handover at quarter end, cash book written up afterwards | Daily cash records, non-cash transactions captured within ten days |
| Accuracy | Corrections overwrite the same record | The correction as a separate entry referring to the original version |
| Order | Storage by person or project folder, no common scheme | One filing scheme for all document types, written down |
| Immutability | Network drive with write access for nearly everyone | Separate roles, a log, versions and a checksum per file |
| Machine analysability | Documents sit as image files without structured attributes | An index per document and an export that has been tested once |
Machine analysability: the image alone is not enough
Documents subject to retention have to remain machine-analysable for the whole period. That does not mean you can look at them on screen; it means data can be sorted, filtered, joined and exported. A stack of scanned pages in a folder does not meet the requirement, however legible each page may be. The difference becomes tangible the moment somebody asks which incoming invoices from a particular supplier exceeded a certain amount in the third quarter.
From this follows a rule that is often broken in practice: what arrives structured must stay structured. Since 1 January 2025 every domestic company must be able to receive electronic invoices in business-to-business transactions (Value Added Tax Act). With such an invoice, the structured data set is the original. A viewing format generated from it is a rendering, not a substitute; keeping only the rendering and discarding the data set gives up precisely the analysability at stake.
The same applies to conversions inside the company. If a document is transferred into another format, for instance when a system is replaced, the original version must be preserved, the conversion must be logged, and no analysable information may be lost in the process. Turning a spreadsheet with formulas into an image destroys analysability, even though the figures remain visible.
For day-to-day operations that means every document needs a set of structured attributes, regardless of whether it exists as an image, as a data set or as both. This index is what turns a collection of files into analysable storage. It can be represented in almost any system in use in mid-size companies, and in the simplest case can be kept as a separate table alongside.
# Minimum index per document: without these fields you have an image, not analysability
doc_number;doc_date;receipt_date;doc_type;partner;net;tax;gross;journal_entry;file;hash
2026-AP-004182;2026-06-15;2026-06-17;Incoming invoice;Supplier 4711;1240.00;235.60;1475.60;J2026-06-0231;2026/06/AP-004182.pdf;sha256:7d1e...c204
2026-AR-000933;2026-06-16;2026-06-16;Outgoing invoice;Customer 20185;3800.00;722.00;4522.00;J2026-06-0244;2026/06/AR-000933.xml;sha256:b0a9...41ef
2026-CA-000517;2026-06-16;2026-06-16;Cash receipt;Counter sale;42.02;7.98;50.00;J2026-06-0245;2026/06/CA-000517.pdf;sha256:1c67...9a3dThe last fields of each row are the important ones: the link to the journal entry and the checksum over the file. Without the link there is no path from document to posting; without the checksum there is no evidence that the file is unchanged since filing. Where documents converge from several upstream systems and the index has to be populated from different sources, this becomes a matter of data integration rather than something accounting can handle on the side.
From document to posting and back again
Traceability can be tested with one simple question: can a single document be followed through the basic records into the financial statements — and back from the statements to the document again? These two directions, forward and backward, are the yardstick. Anyone who plays them through on three randomly chosen documents will know more about the state of their own storage after half an hour than after any product demonstration.
In practice, backward tracing rarely fails because documents are missing; it fails because links are missing. The document sits in a folder, the posting sits in the accounting system, and what connects the two is one person's memory. If that person is unavailable, the path is broken. Robust storage makes the link part of the data record rather than a habit of individual staff.
A unique document number
A continuous number range per document type, without gaps and without duplicates. The same identifier appears in the index, in the file name and in the journal entry so that it carries everywhere.
Document and receipt date
The date on the document and the date it arrived are two different pieces of information. Both belong in the index, because only the gap between them shows whether recording was timely.
Link to the posting
Every document carries the identifier of the journal entry, every journal entry the document number. This twofold connection is what makes both directions of tracing work.
One common filing scheme
A single structure for all document types, for example by year, month and type, instead of folders grown by person or project. The scheme is written down and applies without exception.
Rights and logging
Separate roles for filing, reading and administering, plus a log of access and changes. The log is itself subject to retention and must not be alterable after the fact.
Search via the index
Research through structured attributes instead of file names. If file names are all you can search, you have storage but not analysability in the sense of the requirements.
Process documentation: what belongs in it
The process documentation is the part most often missing and the one that would cost the least effort. It describes how documents enter the business, what happens to them, where they are kept, who may access them and how long they are retained. Its purpose is to give a knowledgeable third party an overview of the workflow within a reasonable time. If it is missing, or if it describes a different workflow from the one actually lived, that can call the propriety of the accounting into question.
It usually consists of four parts. That structure is not an end in itself: each part answers a different question — what happens in business terms, how it is operated, what it runs on, and how operations are safeguarded.
- General description. Which document types exist, where they come from, what path they take through the business, which systems are involved and who is responsible for what. This part has to be understandable to someone who does not know the company.
- User documentation. How the people involved actually work: who scans, who checks, who approves, who posts, what to do when something goes wrong, what happens during an absence. This description is best written where the work happens, not at an IT desk.
- Technical system documentation. Which programs are in use in which versions, how data flows between them, which interfaces and formats are used, where the data is held and how rights are assigned.
- Operations documentation. How ongoing operation is safeguarded: backups and their testing, granting and revoking access, handling of incidents, changes to the systems, and how analysability is preserved when a system is replaced.
Two points are regularly overlooked. First, the process documentation is itself subject to retention, for as long as the records it refers to (German Fiscal Code). Second, for any given period the version in force at the time applies. It follows that changes are dated, earlier versions are kept, and it is visible from when which rule applied. A file that is silently overwritten does not meet that standard — and repeats precisely the mistake immutability is meant to prevent.
Process documentation that does not match the lived workflow is worse than none at all: it asserts an order that nobody in the business recognises.
The effort stays manageable if you treat the task as recording the current state rather than as writing. For a company with a modest system landscape, a workable first version takes a few days. Its structure follows the same rules as any other process documentation: short, current, dated, and kept somewhere it can be found by a person who is not yet with the company today.
Scanning as a replacement: destroying paper needs a rule
Paper documents may be destroyed after scanning. That option is explicitly provided for, but it depends on three conditions: the scan must correspond visually to the original, the subsequent retention must meet the requirements described above, and the procedure must follow a defined rule. Without such a rule, destruction is not an orderly step but simply the loss of a document.
The scanning rule is part of the process documentation and answers a series of unspectacular questions: who may scan, when it happens, with which devices and settings, how completeness is checked, who reviews the result, what happens with an illegible scan, when the paper is destroyed and how that is recorded. One point deserves particular attention: where the colour of a document carries meaning — coloured approval marks, signatures or markings — the scan has to be in colour, because otherwise information is lost.
Not all paper may go. Records that must be kept in the original stay in the cabinet; these include opening balance sheets and annual financial statements (German Fiscal Code). It is also sensible to keep the original wherever its evidential value in legal matters may play a role, for instance with notarised contracts or deeds. Which records fall into that group in a specific business depends on the individual case and belongs in a conversation with your tax adviser.
Step 1: Prepare and identify
Take out the document, keep multi-page records together, determine the document type and assign the number. The receipt date is recorded at this point, because it cannot be reconstructed later.
Step 2: Scan with defined settings
Resolution, colour or greyscale and file format are defined in advance and not left to the day. Coloured notes, stamps and signatures require a colour scan, because here the colour is part of the content.
Step 3: Visual and completeness check
A second person, or at least a second look, verifies legibility, page count and allocation. An illegible or incomplete scan is repeated before the paper leaves the workplace.
Step 4: File, index and protect
The file moves into unalterable storage, the index is populated, the checksum calculated and the link to the posting established. Only after that does the document count as captured.
Step 5: Destroy and record
The paper is destroyed at a defined point in time, usually collected and with some delay after scanning. What is recorded is which batch was destroyed when and by whom — not every individual document, but the procedure.
Take out the document, keep multi-page records together, determine the document type and assign the number. The receipt date is recorded at this point, because it cannot be reconstructed later.
Resolution, colour or greyscale and file format are defined in advance and not left to the day. Coloured notes, stamps and signatures require a colour scan, because here the colour is part of the content.
A second person, or at least a second look, verifies legibility, page count and allocation. An illegible or incomplete scan is repeated before the paper leaves the workplace.
The file moves into unalterable storage, the index is populated, the checksum calculated and the link to the posting established. Only after that does the document count as captured.
The paper is destroyed at a defined point in time, usually collected and with some delay after scanning. What is recorded is which batch was destroyed when and by whom — not every individual document, but the procedure.
The most expensive mistake is the sequence
Retention, periods and access during an audit
Retention periods differ in length, and since the most recent amendment a shorter period applies to accounting vouchers than before. As matters currently stand, books, inventories, annual financial statements, management reports and the working instructions and organisational records needed to understand them must be kept for ten years, accounting vouchers for eight years, and commercial and business letters received or sent for six years (German Fiscal Code). The period starts at the end of the calendar year in which the last entry was made or the document arose. Because there are transitional provisions and sector-specific particularities, applying this to your own case belongs in a conversation with your tax adviser.
Across those periods the records must not only exist but stay legible and analysable. That is less self-evident than it sounds: media age, formats disappear, programs can no longer be started after two system changes. The IT-Grundschutz framework of the Federal Office for Information Security treats archiving as a building block of its own and requires, among other things, that restoration be tested regularly rather than merely planned (BSI). For document storage that means something very practical: once a year, actually retrieve and open a document from an older year.
During a tax audit the authorities may access the data in three ways: directly on the system with read-only access, indirectly through evaluations the business produces to specification, or by handing over the data on a storage medium (German Fiscal Code). From that follows a very tangible task: the export has to exist, it has to contain documents together with index and links, and it should have been tried out once in calm conditions. An export produced for the first time during a running audit costs time and nerves — and raises questions that could have been avoided.
A system change is the critical moment
Introduction in five steps
Step 1: Record document types and paths (1 to 2 days)
Which document types arise, where they arrive, who handles them, where they are kept today. Experience shows that storage locations turn up which nobody had on the list any more: a shared mailbox, a folder on a desktop machine, a stack in a department (field experience).
Step 2: Define the filing scheme and index (1 day)
One scheme for all document types, a number range per type, an index with mandatory fields, a rule for the link to the posting. These decisions cost little time and determine everything that follows.
Step 3: Set up storage, rights and logging (2 to 5 days)
Separate roles, unalterable or version-keeping storage, logging, checksums, backups with tested restoration. In parallel the export for data access is built and played through in full once.
Step 4: Write the process documentation (2 to 4 days)
The four parts are written alongside the setup, not afterwards. Whoever is defining the workflow right now can also describe it; whoever has to reconstruct it six months later needs a multiple of the time.
Step 5: Brief the team and keep it running (ongoing)
A short briefing for the people involved, then an annual review: is the description still accurate, does restoration work, have new document types appeared. The conversation with your tax adviser belongs in the same rhythm.
Which document types arise, where they arrive, who handles them, where they are kept today. Experience shows that storage locations turn up which nobody had on the list any more: a shared mailbox, a folder on a desktop machine, a stack in a department (field experience).
One scheme for all document types, a number range per type, an index with mandatory fields, a rule for the link to the posting. These decisions cost little time and determine everything that follows.
Separate roles, unalterable or version-keeping storage, logging, checksums, backups with tested restoration. In parallel the export for data access is built and played through in full once.
The four parts are written alongside the setup, not afterwards. Whoever is defining the workflow right now can also describe it; whoever has to reconstruct it six months later needs a multiple of the time.
A short briefing for the people involved, then an annual review: is the description still accurate, does restoration work, have new document types appeared. The conversation with your tax adviser belongs in the same rhythm.
Effort, cost and ongoing operation
For orientation: taking stock of document paths, with an assessment and a sequence, starts at 1,900 euros net. Implementing one workflow — storage, rights, index, logging, export and the accompanying description — starts at 4,900 euros net. The effort depends less on technology than on the number of document types and on the special cases: the document that passes through three stations, the till with its own recording obligation, the upstream system that is hard to export from. Current rates are listed on the pricing page.
The second block is operation, and it is underestimated more often than the build. Storage that nobody watches loses its properties within a few months: rights get added and never withdrawn, new document types appear without a scheme, the description quietly ages. Ongoing support starts at 190 euros net per month and covers monitoring the backup, testing restoration, maintaining rights and keeping the documentation current. If you keep that task in-house, at least give it a name. How the path from paper to an orderly file works in detail is described on the document digitisation page.
Related Articles
Writing process documentation for tax audits
Process documentation under the German GoBD rules: the four required parts, how detailed it must be, how to keep it current and what its absence can mean.
Retention periods digitally: schedules, holds, deletion runs
Retention duties and deletion duties only appear to conflict. How to build a filing concept with periods per record type, legal holds and documented runs.
Migrating Data from a Legacy System Without Surprises
Inventory, field mapping, trial runs and totals checks: how to move data out of a legacy system and how to evidence that the transfer really was complete.