Short answer: an incoming fax arrives as an image, not as text. Until it has been run through OCR, named according to a stable convention and filed in a predictable folder, it isn't a document: it's an attachment sleeping in a generic inbox. The chain that actually works has five links — reception on a dedicated address, systematic OCR, standardised naming, a routing rule to the right department, full-text indexing. Each of these links can be set up in less than a day, and it's their sequence, not their sophistication, that makes the difference.
Many organisations have gone halfway. They have dropped the fax machine, moved to an online fax service, and now receive their faxes as PDFs by email. Real progress: no more thermal paper rolls, no more pages lost in an output tray, no more confidential documents left exposed in a corridor. But the stream lands in a shared inbox along the lines of fax@…, checked by three people, and each PDF carries a name like fax_20261009_114237_0033xxxxxxxx.pdf. Six months later, finding "the certificate the firm sent in April" becomes an archaeological dig.
The incoming fax is a special case of document digitisation: image quality is poor by design (204 × 196 dots per inch in fine mode, pure black and white), useful metadata is thin, and the sender follows no convention whatsoever. That is precisely why a method is needed.
What an incoming fax really is in 2026
An image, and not much else
A received fax follows the ITU-T T-series recommendations (T.4 for image coding, T.30 for the session protocol, T.38 for transport over IP). The result is a bitonal image: every dot is black or white, with no shades of grey. There is no text layer at all. A search tool opening that PDF will find nothing to index, even if the human eye reads the content perfectly.
That's the first mistake to fix. A fax that hasn't been through OCR is invisible to any full-text search, and therefore invisible to your organisation the moment it leaves the screen of the person who opened it.
The metadata actually available
A fax session carries little information, but what it does carry is valuable:
| Data | Source | Reliability |
|---|---|---|
| Calling number (CSID / TSI) | Declared by the sending fax machine | Medium — configurable, hence spoofable |
| Reception timestamp | Provider's server | High |
| Page count | T.30 protocol | High |
| Resolution and negotiated speed | T.30 protocol | High |
| Header printed at the top of the page | The image itself | Variable, readable after OCR |
Remember the essentials: the timestamp and the page count are server-side data, solid and usable as evidence. The calling number is a declaration by the sender — useful for routing, insufficient for identification with any certainty. Never build a trust rule on that field alone.
Step 1 — Separate the streams at the point of reception
The most cost-effective reflex costs nothing: stop letting everything arrive in the same place.
If your provider lets you use several inbound numbers, assign one per business stream: one number for administrative documents, one for orders, one for sensitive files, one for partner returns. Each number points to a distinct email address, and each address to a different filing rule. Sorting is then done by the network, upstream, with no human intervention at all.
When only one number is possible, sorting happens downstream, based on the calling number and the OCR'd content. It's more fragile, but perfectly workable — see step 4.
Ground rule: a
fax@inbox checked by everyone is an inbox checked by no one. Appoint one owner per stream, with a named backup.
Our receiving a fax page explains how inbound numbers and fax-to-email work. This article starts where that one stops: once the PDF is in the inbox, what do you do with it?
Step 2 — OCR, the non-negotiable link
Why running OCR on a fax is a particular exercise
A fax arrives at 204 × 196 dpi, well below the 300 dpi recommended for reliable character recognition. Modern OCR engines cope, but on two conditions: that the image isn't skewed, and that the sender hasn't faxed a photocopy of a fax of a scan.
Quality gains are therefore often won on the sending side. When you can influence your regular correspondents, three instructions are enough: original documents rather than photocopies, "fine" mode rather than "standard", and no faxing of a document stapled askew.
On your side, pre-processing before OCR markedly improves results:
- automatic straightening (deskew) — a fax tilted by 3° makes recognition collapse;
- removal of salt-and-pepper noise, typical of noisy analogue lines;
- cropping of the transmission header, which pollutes date extraction;
- detection and rotation of pages scanned upside down.
What result to aim for
The goal isn't typographic perfection. It's search. A recognition rate of 92 to 96% on a clean fax is more than enough to find a name, a case number or an invoice reference. For human reading, the original image is what counts — and that's the one to keep.
Hence the following rule: produce a searchable PDF with a hidden text layer, overlaying the OCR output on the untouched original image. You gain search without ever altering the document. In case of dispute, it's the image that is enforceable, not the transcription.
For organisations that also digitise paper alongside faxes, a duplex document scanner with an automatic feeder cuts processing time substantially, and lets you feed the same OCR chain as incoming faxes. One pipeline, two sources: that's what avoids duplicate procedures.
Step 3 — Name it, or give up finding it
The file name is the most used and most neglected piece of metadata. A stable convention, applied without exception, replaces half a document management system on its own.
The scheme that stands the test of time:
YYYYMMDD_HHMM_STREAM_SENDER_SUBJECT_PAGECOUNT.pdf
20261009_1142_HR_CabinetMartin_EmploymentCertificate_3p.pdf
Four properties make this convention effective:
- the date first, in ISO format: alphabetical sorting becomes chronological sorting, in any system;
- no accents, no spaces, no special characters: a file is never blocked during a transfer, a sync or an upload to a third-party portal;
- the stream before the sender: documents from the same department stay visually grouped;
- the page count: an instant integrity check, on opening as on archiving.
If your organisation still renames files by hand, time the operation for a week. Twenty faxes a day at forty seconds each is over an hour a week — the equivalent of a full working day per quarter spent writing file names.
Step 4 — Route automatically to the right department
Once the text is available, routing becomes trivial. Three families of rules cover most real-world cases.
Routing by calling number
The simplest and most reliable approach for regular correspondents. A "number → department → folder" lookup table effortlessly handles 60 to 80% of the volume in most organisations, precisely because fax is a channel of established relationships: the same partners keep coming back.
Remember to normalise numbers to international format before comparison: +33 1 ..., without the national trunk zero, without spaces, without dots. The same correspondent may present itself in three different spellings depending on its equipment.
Routing by OCR'd keyword
For the rest, expressions in the document decide: "purchase order", "certificate", "requisition", "formal notice", a case number in a recognisable format, a company registration number. A dozen well-chosen rules are usually enough.
Watch out for precision: a rule that's too broad generates false positives, which cost more than manual sorting. Five reliable rules plus a "to be qualified" queue beat twenty approximate rules.
Routing by form
Page count, page format, the presence of a signature box: all usable clues. A one-page fax received at night is very rarely a complete file — often a confirmation, sometimes a misdirected send.
Always provide a last-resort queue, checked at least once a day. No rule covers everything, and an unrouted fax that disappears silently is the worst possible scenario.
Step 5 — Index so you can find things in under ten seconds
The metadata to store systematically
Beyond the file name, five fields deserve to be recorded in an index — spreadsheet, lightweight database or document management tool, it doesn't matter:
- date and time of reception (server source, not document source);
- normalised calling number and identity of the correspondent;
- business stream and the file it belongs to;
- page count and negotiated resolution;
- purge date, calculated at the moment of reception.
That last field is the one most often forgotten, and it's the one that stops you building up a stock of personal data kept indefinitely. The principle of storage limitation appears in Article 5 of the GDPR, and the CNIL regularly reminds us that the retention period must be set according to the purpose, right from collection. A deadline calculated automatically on reception turns a legal obligation into a simple database field.
Backup: the 3-2-1 rule
An indexed fax stream becomes a documentary asset. It deserves the same protection as the rest: three copies, on two different media, one of them off site. For modest volumes, an encrypted external hard drive kept off the premises covers the third copy at very low cost. For sensitive files that have to travel physically, a hardware-encrypted USB key avoids the unprotected copy forgotten in a drawer.
Weekly quality control
Fifteen minutes a week is enough:
| Check | Warning signal |
|---|---|
| Volume received vs. previous week | Sudden drop → inbound number down |
| Size of the "to be qualified" queue | Steady growth → rules need revising |
| Faxes with 0 pages or illegible | More than 2% → line or T.38 problem |
| Naming deviations | Convention not applied → training to redo |
A drop in incoming volume is the most important signal, and the easiest to miss: nothing rings when a number stops receiving. That is exactly what a continuity procedure covers, and it should be monitored like a backup.
The case of the illegible document
It will happen. A black page, a truncated column, a signature that has melted away. The right response comes down to three actions, in this order:
- do not delete the original file, even if unusable: it proves a transmission took place at a given time;
- call the sender back and ask for a resend in fine mode, from the original document, and if possible page by page;
- log the incident in the index, with a reference to the first transmission.
If the problem recurs with the same correspondent, the cause is almost always their sending chain: a badly configured ATA, a photocopy of a photocopy, or a modem speed too high for their line. A good book on enterprise document management helps frame these conversations with recurring partners, setting out in black and white what you expect from them.
Frequently asked questions
Does a fax lose its evidential value once OCR'd?
No, provided you keep the original image. A searchable PDF overlays a recognition layer without modifying a single pixel of the received image. It is that image, together with the server timestamp and the transmission report, that carries the evidential weight. An OCR transcription on its own, by contrast, has no particular force: it's merely a search aid.
Should faxes be converted to PDF/A?
For any document intended to be kept for several years, yes. PDF/A is an ISO-standardised format designed for long-term archiving: it forbids external dependencies and embeds everything needed for rendering. For a fax, conversion is trivial since it's an image with no exotic fonts.
Does OCR work on handwritten signatures?
No, and it isn't supposed to. A signature is an image, not text. OCR will recognise the typed name beneath the stroke, not the stroke itself. To verify that a signature is present, you rely on zone detection, or on a visual check for documents where it matters.
How long should an incoming fax be kept?
That depends entirely on the content, not the channel. An accounting document follows accounting retention periods, a health document follows health periods, a routine commercial exchange can be purged after a few months. The fax channel adds no retention period of its own: it's the nature of the document that governs. Hence the value of indexing the business stream from the moment of reception.
Can everything be automated?
No, and that's good news. Aim for 80% automation and a human qualification queue for the rest. Organisations that attempt 100% end up creating rules so broad that they misfile — and a misfiled document costs more than an unfiled one, because nobody knows it's lost.
Summary
- An incoming fax is an image with no text layer: without OCR, it is invisible to any search.
- Separate the streams at reception, by inbound number whenever possible: free sorting is the best sorting.
- Produce a searchable PDF with a hidden text layer that preserves the original image — search on one side, evidence on the other.
- Apply an ISO naming convention, with no accents or spaces, including the page count.
- Route by normalised calling number first, by keyword second, and always keep a "to be qualified" queue.
- Index five fields, including the purge date, calculated at reception.
- Check the stream once a week: a drop in incoming volume triggers no alarm on its own.
To check the availability of numbers and destinations, see the list of supported countries. Billing questions and volume limits are covered in the FAQ, and you can start a transmission straight from the home page.
Tagsfaxpaperlessplatform
