Going paperless without losing the file: folders, naming and retention

Going paperless fails in a predictable way. The firm buys a scanner, someone spends three weekends digitising the cabinets, and eighteen months later nobody can find anything, so a partner starts printing documents again “just to have a copy”. The paper comes back, and now there are two systems.

The scanning was never the hard part. The hard part is that a filing cabinet enforces a structure whether you think about it or not, and a hard drive does not. Remove the cabinet without replacing what it was quietly doing, and you get a large searchable pile that nobody trusts.

Three decisions do most of the work. Make them before you scan anything.

Decide the structure, and keep it shallow

The most common mistake is recreating the cabinet exactly: a folder per year, then per practice area, then per client, then per matter, then per document type, six levels deep.

Deep hierarchies fail for a structural reason. Every level is a decision at filing time and a guess at retrieval time, and both compound. Six levels means six chances for the filer and the finder to disagree, and they only need to disagree once for the document to be effectively lost.

Go shallow and use attributes instead. One level for the client or matter, one level for a small set of categories inside it, and then let type, date and status do the rest as properties of the document rather than as folders. That way a document can be a tax return and a 2026 document and an item awaiting review without having to be in three places.

Organizing with folders covers building that structure, and document rules and types is the attribute half. Do them together. A shallow structure with no document types is just a shallower pile.

A test worth applying: if you cannot explain your structure to a new hire in ninety seconds, it is too complicated, and it will be filed inconsistently from the day they start.

Decide the naming convention, and write it down

Naming is where paperless projects rot quietly. There is no error message when somebody names a file scan_0047.pdf, and no one notices until the day that file is the one you need.

A convention that survives contact with reality has four properties. It starts with the date in YYYY-MM-DD form, so alphabetical order is chronological order. It uses the document’s own date, not the scan date, because the scan date is meaningless a year later. It names the document in words a person would use, not a code. And it is short enough that people actually type it.

2026-06-14 Insurer letter - denial of claim.pdf is better than anything shorter, and much better than anything cleverer.

Then write it on one page and put it where new staff will see it. An unwritten convention is a convention that has already diverged; the person who set it up knows it, and nobody else has ever been told.

One thing not to do: do not encode information in the filename that the system already holds as a field. Client name, matter number and status belong in the record, where they can be searched, filtered and corrected. A filename that duplicates them is three things to keep in step instead of one. Finding documents shows what search and filters can do once the fields are populated, which is usually more than people expect.

Decide retention before you have a decade of files

Retention is the decision firms defer, and deferring it is itself a decision, one that defaults to keeping everything forever.

Everything forever is not a neutral choice. It grows your exposure with every year, it makes search worse, and it means that any future request or incident reaches back further than it needs to. Storage being cheap is not an argument for it, because storage was never the cost.

You need a written schedule with actual numbers in it, informed by your jurisdiction’s requirements for client files, your malpractice insurer’s expectations, and anything specific to the matter types you handle. Retention periods vary by state and by document type, and this is one of the few places where it is genuinely worth confirming the requirements that apply to your practice rather than adopting somebody else’s numbers.

Then someone has to run it. An unenforced schedule is the same as no schedule, with extra paperwork. Once a year, on a fixed date, is enough for most small firms.

The corresponding capability is that you can actually get things out and actually delete them. Both are worth verifying before you commit; download and export options covers retrieval in bulk, and deleting documents covers the other end, including what deletion does and does not remove.

Scanning: the part everyone worries about

Now the easy bit, briefly.

Do not digitise the entire archive. Digitise the active matters, and digitise closed matters only as they are touched. Most closed files are never opened again, and scanning them is weeks of work for an outcome nobody experiences.

Search inside documents needs the text layer, so use OCR on everything you scan. A PDF that is only an image is a document you can find by its name and by nothing else, which throws away half of what you went paperless for.

And set the day-one rule: from a fixed date, anything arriving on paper is scanned and filed within twenty-four hours, and the paper either goes to the client or goes into a retention box that nobody works from. Firms that keep working from paper “for now” are still doing it three years later.

The check that tells you it worked

There is one test, and it is worth running quarterly. Pick a matter that closed two years ago at random. Ask someone who did not work on it to find a specific document from it, and time them.

Under two minutes, the system works. Between two and ten, the structure is sound and the naming has drifted. Over ten minutes, or if they have to ask a colleague, the thing you built is a pile with folders drawn on it.

That test is more honest than any audit, because it measures what the system is actually for. Everything in this article, the shallow structure, the boring names, the written schedule, exists to keep that number under two minutes in four years’ time, when everyone who set it up has forgotten how it works.

If you want to sanity-check your structure before you commit to it, send us a sketch of it and we will tell you where it will break.