How should I name and file digital receipt and document files?

What does a naming convention have to achieve?

Before choosing a scheme, write down the three jobs it must do, and judge every rule against them:

• Find. Anyone can pull a document by period, by counterparty or by type.
• Sort. A folder listing sorted by name comes out in a useful order without opening a file.
• Separate. No two documents share a name, and a corrected document and the one it replaces are both identifiable.

Which elements go in the name, and in what order?

Harvard Medical School's file-naming guidance says to let the way you will sort and search for files set the order of the elements, and explains why: the computer arranges files by name, character by character, so the first element decides how a folder sorts, and if you will look files up by date, the date goes first. Month-end, year-end and your accountant's questions start from a period, so by Harvard's rule the date leads. Use four elements in this order, joined by underscores, as in YYYY-MM-DD_Counterparty_Type_Ref.pdf:

• Date. Use the date printed on the document, year first.
• Counterparty. Write the vendor, customer, bank or agency exactly as your normalization list spells it.
• Type. Use one word from your fixed list of document types.
• Ref. Use the number printed on the document; if there is none, use 01, or 02 and up for further same-day documents of that type from that counterparty. Write it with letters, digits and hyphens only: drop a leading # or No., and turn slashes, spaces and other separators into hyphens, so INV/2026/0412 becomes INV-2026-0412.

Only two suffixes may follow, always last: a version such as _v02 and a page part such as _p01of02. When both apply, the version comes first: _v02_p01of02. With the date first, every folder lists in date order, and one counterparty's same-day documents sit together, where a clash is easy to spot before you save.

Which date goes in the name, and how is it written?

Use a date printed on the document, never the day you scanned, photographed or downloaded it. Your written key fixes which date each type uses, for example the transaction date on a receipt, the issue date on an invoice or bill, the closing date on a statement and the signing date on a contract. A capture date records when you got around to filing: a March receipt scanned in May files among May's documents, and nothing looks wrong until someone asks for March.

Harvard's file-naming guidance recommends a year-first date because it keeps files in chronological order, and recommends ISO 8601 dates, written YYYYMMDD or YYYY-MM-DD. Compared character by character, month-first names sort wrongly across years: 03-02-2026 lists ahead of 03-14-2025. Write month and day with two digits, as Harvard's guidance advises leading zeros to keep numbered files in sequence; in a listing compared character by character, 2026-3-09 would otherwise list after 2026-10-01.

How do you keep one spelling per vendor and one word per document type?

Keep a normalization list pairing the one spelling you use with every variant you meet: AcmeSupply stands for "Acme Supply Co.", "ACME SUPPLY CO" and "Acme". Run the words together with capitals, drop spaces and punctuation, write "&" as "And", and drop Inc, LLC or Co unless two counterparties need them to differ. Add each new counterparty to the list before naming its first document.

Freehand spelling breaks a sound convention as documents pile up: in counterparty folders, three spellings make three folders, and in a year folder, a search for one spelling can miss files saved under another. The split shows only when a document goes missing, and repairing it means renaming across the archive.

Document types need the same control: Harvard's file-naming guidance says to standardize categories or replace them with 2- or 3-letter codes, and to document the codes. Whole words suit files that anyone else will read. A starting list might look like this:

Free text such as "rcpt", "receipt-scan", or "invoice" for a vendor's bill splits one type into several, so "every bill for 2026" stops being one search.

What should stay out of the name?

Use one test: a detail belongs in the name only if you need it to identify the document once it leaves its folder, for example attached to an email or uploaded to accounting software, and it will not change after filing. Harvard's file-naming guidance makes the first half of that point: a moved or shared file should be identifiable from its name. So the four elements stay even where a folder repeats one of them, and these stay out:

• Anything true of every file in the store. Your business name and words like "Documents" or "Scan" belong on the top folder.
• The format. The extension already says PDF or JPG.
• Anything that changes after filing. Paid, approved, disputed, the expense category and the account belong in your accounting records.
• The filing itself. The capture date and the filer's initials describe your workflow, not the document.

Compare MyBizLLC_Expenses_2026_March_Receipt_HarborHardware_Paid_Scanned_JD.pdf with 2026-03-14_HarborHardware_Receipt_0412.pdf. If your system searches inside documents, check its own documentation before relying on it to sort by document date or show the current version; date-first names and version suffixes do both in a name-sorted listing.

Should folders be organized by period or by counterparty?

Choose one axis for a single level of folders, and let the name carry the rest:

• Period first. Use one folder per year, with date-first names inside. It suits period lookups, such as month-end, year-end or your accountant's questions, many counterparties with few documents each, and high volume, since each year's folder closes.
• Counterparty first. Use one folder per counterparty, with date-first names so each folder lists in date order. It suits a few counterparties, such as a landlord or a main supplier, producing most documents, with lookups that start from the relationship; a period is then found by search.

Do not mix the two, and do not nest year, month, vendor and type: every extra level is another choice two people can make differently, and anything several levels down is invisible from the top. Depth also costs length. Microsoft's Windows file-naming documentation says that before Windows 10 version 1607 the maximum path length is 260 characters (MAX_PATH), and says later versions need a registry key change or the Group Policy tool to remove that limit.

How do you name batches, split scans and corrected documents?

Each case has one rule:

What does the convention look like on real documents?

Applied to four document types in a period-first store, the pattern gives these names:

Without the reference, both receipts would be 2026-03-14_HarborHardware_Receipt.pdf, and the second would collide with the first. In the invoice pair, _v02 is the latest and _v01 stays on file, marked as replaced. Sorted by name, the folder lists in the table's order, which is date order; a counterparty-first store puts the same names in folders such as BrightDental.

Which characters and lengths survive every system the files pass through?

Keep to letters, digits, hyphens and underscores, with one period before the extension and no spaces. Harvard's file-naming guidance says conventions should use only alphanumeric characters, dashes and underscores, and warns that many computer systems cannot handle spaces. The same guidance says file names should be 40-50 characters; keep normalized counterparty names short so the full pattern stays near that length. Two systems' published rules show what else is at stake:

• Windows. Microsoft's Windows file-naming documentation reserves < >: " / \ | ? *, so times carry no colons and dates no slashes. It reserves device names such as CON, PRN, AUX and NUL, also with an extension, applies its file rules to folders, and says not to assume case sensitivity, treating OSCAR, Oscar and oscar as the same name.
• OneDrive and SharePoint. Microsoft's undated OneDrive and SharePoint restrictions page lists " *: < > ? / \ | as characters not allowed in OneDrive file and folder names, warns that they may stop files syncing, and says to rename before uploading. It also bars names such as CON, PRN, AUX and NUL for files or folders, and says file name and path length limits differ between apps and Office versions.

So a CON folder for contracts fails, and capitals can never be what tells two files apart. Other systems, including accounting software's upload screens, have their own rules, so before renaming at volume, put one sample name through every system it will pass, download it back and compare.

What must a renaming or refiling never break?

The IRS's Rev. Proc. 97-22 applies to taxpayers who keep books and records in an electronic storage system, which it defines as a system to prepare, record, transfer, index, store, preserve, retrieve and reproduce them, including by electronically imaging paper documents onto electronic storage media. It does not say whether a folder of scans is, by itself, such a system.

The revenue procedure requires such a system to include reasonable controls to ensure its integrity, accuracy and reliability; reasonable controls to prevent and detect the unauthorized creation of, addition to, alteration of, deletion of, or deterioration of stored records; an inspection and quality-assurance program shown by regular evaluations, including periodic checks of stored records; and a retrieval system that includes an indexing system, a requirement it says is satisfied if the indexing system is functionally comparable to a reasonable hardcopy filing system, with no separate description database needed if comparable results are achieved without one. Records the system reproduces must show a high degree of legibility and readability on screen and in hardcopy. Reasonable controls must also protect the indexing system against the unauthorized creation of, addition to, alteration of, deletion of, or deterioration of any entries. For each electronic storage system used, the taxpayer must maintain, and make available to the IRS on request, complete descriptions of the system, including all procedures relating to its use, and of the indexing system.

Rev. Proc. 97-22 does not mention renaming. If your names and folders are how you identify and retrieve records, a renaming pass changes those entries, so make each change deliberately, log it (step 4) and alter no record's content.

What should you do with the files you already have?

There are two paths, and you can combine them, with a written boundary between them:

• Rename in a bounded pass. Bring a slice you will finish, such as this year and last, fully under the convention. It fixes retrieval there, at a cost in hours that grows with volume.
• Freeze and start fresh. Move the old folders, with their subfolders as they are, into one folder named for the cut-off, such as Archive-to-2026-06-30; count the files before and after, never let a move replace a file, and apply the convention to everything filed after the cut-off. It costs almost nothing now, but the old files stay as hard to find as they are.

On either path, any file that something else points to by name or location, such as a shared link or a path in a spreadsheet, keeps its name and place unless you update that pointer in the same step. Work the pass in this order:

1. Write the convention and both lists first, and update them as the archive turns up new variants.
2. Take the newest month first, so the most-used files are fixed first and you can stop at any month boundary.
3. Count the month's files, then rename and move them one at a time, changing names and locations only, never re-saving, converting, compressing, cropping or re-scanning. The one exception is a batch scan: split it under the batch rule, leave the batch file itself unchanged, and add the new files to the month's count.
4. Log each change with the old name, new name, date and who made it, so every record stays findable by its old name.
5. Never let a rename replace an existing file: a different document takes the next sequence number, a correction follows the version rule, and a second copy is a duplicate to resolve separately.
6. Count again, expecting the first count plus the files created by splits, open a sample of renamed files to confirm they open and read clearly, and mark the month done.
7. When you stop, record in the written convention where the renamed slice ends and the frozen archive begins.

How do you write the convention down so someone else files the same way?

Put it in a short text file at the top of the store, named to list first, such as 0000_README_Filing.txt. Harvard's file-naming guidance says to document naming conventions in a README kept with the files, including a top-level README on how to navigate the structure. Yours should give the pattern with one example per document type, the date each type uses, the type words, the normalization list, the reference, page, version and batch rules, the folder axis, the character rules, the stay-out test and where the renamed slice ends. If your store falls under Rev. Proc. 97-22, this file is also a start on the descriptions of the system, its procedures and its indexing that the revenue procedure requires.

What changes when several people file?

Two people diverge at once without a written rule. Make the file the specification and give one person ownership of the normalization and type lists: nobody else invents a spelling or a type word, and the owner adds each new one before the document is saved.

What changes when an outsider reads the files?

An outside bookkeeper, accountant or reviewer lacks your context, so names must make sense without you: whole words for types, recognizable counterparty names and no private abbreviations. Put a short key at the top of the file explaining every token and suffix in a line each.