Archival processing when the backlog outpaces you: levels, MPLP, and what AI changes

Every archive holds two collections. The one that is processed — arranged, described, and findable through the catalogue — and the one that is not. Boxes arrive through accession faster than anyone can work through them, and the gap between what a repository holds and what a researcher can actually find widens with every deposit. That gap is the backlog, and archival processing is the work that closes it.

The difficulty is that processing is slow and skilled, and capacity rarely matches intake. A short project contract, a single archivist covering what used to be a team, a volunteer-run local archive with no professional staff at all — against a steady inflow of unprocessed collections, the queue only grows. This is the daily reality the term “archival processing” describes, and it is worth being precise about the work before looking at how to make it go faster.

What archival processing actually means

Processing is the work of turning an accession into a described, navigable archival collection. It has two halves, usually summarised as arrangement and description.

Arrangement is intellectual before it is physical. Following archival principles and archival theory, the processing archivist begins to arrange and describe the records — respecting the principle of original order where it survives, grouping the collection as a whole into series and subseries, and recording how the parts relate. This intellectual arrangement is half of what the profession calls physical and intellectual control: knowing both what the records are and where they sit. Physical processing follows — placing materials into acid-free folders and boxes, one folder and box at a time, with preservation and the condition of the material handled as you go. On a large processing project, a brief collection survey first tells you how much arrangement the collection needs.

Description then captures the contents of the collection at each level: a scope and content note, dates, creators, a conditions governing access note, and subject headings that become the access points a researcher searches on. The output is a finding aid — in UK practice, the catalogue entry — written to a content standard such as ISAD(G) or DACS (Describing Archives: A Content Standard) so that the description of the collection is consistent and exchangeable, whether it sits in a record office, a university library, or a special collections and archives service. To describe the collection well is what makes it usable; a finding aid is only as good as the description inside it.

Crucially, description happens at a chosen level of description. A collection can be described at fonds or collection level in a paragraph, broken down to series, then to file, and — rarely — to individual item. Each step down multiplies the work. Item-level description of a large manuscript collection is, for most repositories, a dream rather than a processing plan.

Why the backlog grows faster than you can process it

The maths is unforgiving. Manual processing of a collection moves at a few boxes a day when it is going well. Intake does not pause to let you catch up, and the richer the description you attempt, the slower you go. So the backlog is not a sign of poor collection management — it is the predictable result of finite capacity meeting open-ended intake.

When time is short, something has to give, and it is almost always descriptive depth. Item-level becomes file-level; file-level becomes series-level; a difficult collection gets a holding entry in the accession records and goes back on the shelf. The National Archives’ surveys have repeatedly found significant proportions of collections unlisted or under-described across the sector — in university archives and special collections, local authority repositories, manuscript collections, and personal papers alike. Whole record groups exist, are preserved, and remain effectively hidden collections.

More Product, Less Process — and its limit

The profession’s pragmatic answer to this is well known. In 2005, Greene and Meissner argued for More Product, Less Process: describe at a higher level, do less item-by-item work, and get more of the backlog accessible sooner. Minimal processing is sound advice. A collection findable at series level today is more use to a researcher than a perfect item-level catalogue that will not exist for a decade.

But minimal processing carries a cost that is easy to under-state. Coarse description is thin description. A series-level note tells a researcher a box of correspondence exists; it does not tell them whose letters, on what subjects, from which years. The collection is listed, but it is barely discoverable. More Product, Less Process trades depth for coverage — and depth is exactly what makes archival material findable.

The discoverability cost: what goes in is what comes out

This is where processing meets the systems archives rely on. A catalogue platform — CALM, ArchivesSpace, AtoM, or any of the systems a repository runs — can only surface what has been described. Search, browse, and online access all work on the archival description, not the records themselves. If the description is sparse, the search results are sparse. What goes in is what comes out.

So the backlog problem and the discoverability problem are the same problem wearing two hats. An unprocessed collection cannot be found because it is not in the catalogue at all. A minimally processed one can be found only at the coarsest collection level, which for most researchers means not found in any useful sense. The records are dark either way.

What AI-assisted archival description changes

For most of the profession’s history the trade-off has been fixed: you could have coverage or depth, not both, because depth meant manual time the repository did not have. AI-assisted archival description is what loosens that constraint.

The change is specific. A model can produce a first-pass description of an archival object — a draft scope and content note, suggested dates, names, and subject headings — in seconds rather than minutes. That does not remove the processing archivist; it moves their time. Instead of composing every description from a blank page, the archivist reviews, corrects, and confirms proposals, directing attention where judgement is actually needed. First-pass description at scale is the part that has always been the bottleneck, and it is the part this approach compresses.

Done responsibly, this is not a shortcut around standards for archival description. The point of our whole approach to AI is that the archivist stays in control: every proposed field carries a confidence indicator, nothing enters the record without human review, and the output exports as standards-ready metadata — ISAD(G), EAD3, Dublin Core — into the systems you already run. We have written separately about keeping AI defensible in the archive, because for an archive the audit trail matters as much as the speed.

What it means in practice is that minimal processing no longer has to mean minimal description. You can take a series-level decision for archival arrangement while still giving the items in a collection a usable description, because the cost of that description has fallen. Better description means better access to the materials, and richer information about the context and the value of the records within a collection. The backlog gets smaller and what comes out of it is genuinely findable.

Processing is about access, not tidiness

It is easy to treat the backlog as a housekeeping problem — a queue to be cleared. It is really an access problem. A collection is only as available as its description allows, and a repository’s real holdings are not what sits on its shelves but what a researcher can find. Archival processing is the work that turns one into the other, and for decades the rate of that work has been capped by how fast a person can write descriptions.

That cap is what is changing. The archivist’s judgement, the standards, and the review still sit at the centre of the work — but the first pass, the slow part, no longer has to be done by hand. For archivists that means more of the backlog becoming findable within a project’s lifetime; for directors and heads of service it means the collection of record actually reflecting what the institution holds. We have written more on the hidden cost of uncatalogued collections if you want the business case in full.

To see AI-assisted, standards-ready archival description working on your own material, join the early-access waitlist — places are limited and every account is personally onboarded.