Designing a DAM taxonomy that survives its first year

A taxonomy is not a filing scheme. It is a search interface expressed as vocabulary, and it should be designed backwards from the searches it has to answer.

Start from the questions, not the categories

Before drawing a single branch, write down the twenty requests the library actually receives. Not hypothetical ones — real ones, from the last month.

They will look like this: a portrait of a named person, taken in the last two years, that we own outright. Anything we hold from a particular event. A landscape-orientation exterior of a specific building, cleared for commercial use. Every frame by one photographer.

Now notice what those requests have in common. None of them is a path through a hierarchy. Each is a combination of independent attributes: subject, date, orientation, rights status, creator. A tree cannot answer them, because a tree forces one attribute to be the top level and every other to be subordinate to it.

Facets, not folders

The structure that answers those requests is a small set of independent facets, each with its own controlled list of terms, combined at search time.

A workable starting set for a media archive:

• Asset type — photograph, illustration, graphic, video, document
• Subject — what the asset is of, drawn from a controlled list
• People — named individuals, as a separate facet from subject
• Place — geography, at a consistent level of precision
• Event or job — the assignment the asset came from
• Rights status — owned, licensed with terms, editorial-only, unclear
• Orientation and shape — because it is asked for constantly and costs nothing to record

Any asset can carry terms from every facet. A search combines them. Nothing has to be filed in one place, which means nothing has to be found in one place either.

Keep the hierarchy shallow

Facets can have internal hierarchy — place certainly should — but keep it short. Two levels answers most questions; three is the practical ceiling.

Deep hierarchies fail for a reason that has nothing to do with theory. Each level is a decision somebody has to make at filing time, and each decision is a chance to file something where the next person will not look. A five-level tree does not organise a library; it distributes it.

The symptom to watch for is a level where the choice between siblings feels arbitrary. If the person filing has to think, the person searching will have to guess.

Controlled terms, and one of them

Every facet draws from a fixed list. Free text in a facet field destroys it within months: singular and plural of the same word, two spellings of the same name, an abbreviation and its expansion, a typo that nobody catches.

For each concept pick one preferred term and record the alternatives as synonyms that map to it. Somebody searching for the synonym finds the assets; somebody filing sees only the preferred term.

Scope notes are worth the effort on any term that could be read two ways. A term without a scope note will be applied in two ways, and a facet where the same term means different things to different people is worse than no facet at all.

Rights as a first-class facet

This is the one that gets left out and the one that costs money.

Rights status deserves its own facet with a short, unambiguous list, because the question "may we use this" is asked on every retrieval and cannot be inferred from the subject or the source. "Unclear" must be one of the values. A library where unclear assets are indistinguishable from owned ones is a library that will eventually publish something it should not have.

Governance before growth

Decide, before the library fills, who may add a term, who reviews additions, and how often the vocabulary is audited.

Without that, new terms arrive as needed, by whoever needs them, and within a year the controlled list has become free text with extra steps. With it, additions are deliberate, synonyms are captured, and the audit catches the drift while it is still small.

The audit itself is unglamorous and short: terms used once, terms never used, terms that have drifted from their scope note, and assets sitting in "unclear" that nobody has resolved. An hour a quarter is enough if it actually happens.

The failure we see most

A taxonomy designed as a subject tree, filled for six months, then abandoned because nobody can find anything.

The mechanism is always the same. The tree made subject the primary axis. The real requests arrive on other axes — date, rights, person, orientation — and the tree cannot serve them, so people start searching on filenames instead. Once that happens, the filing discipline collapses, because filing is now work with no payoff.

Conclusion

Design from the requests, use independent facets rather than a single tree, keep each facet shallow and controlled, give rights its own facet with an explicit unknown, and decide who owns the vocabulary before anybody starts adding to it. A taxonomy built that way answers the questions the library is actually asked, which is the only reason it will still be in use next year.