How to Migrate Decades of Drawings Into a Document System (Without Retyping a Single Title Block)

Every facilities team has the same folder. It's called something like Drawings – OLD or Archive – DO NOT DELETE, it has 20,000 files in it, and everyone agrees it should be in a real document system. It never happens, for one reason: metadata.

To be useful in a system, each of those files needs a building, a floor, a trade, a sheet number, a title. Twenty thousand files at thirty seconds each is about four months of one person's full-time work — and the person who knows the archive well enough to do it is the person who can least be spared. So the folder sits.

Here's how to move a legacy drawing archive into a document system without retyping a single title block, and without a four-month project.

Step 0: Decide what you're not migrating

Not everything in the old folder is a record. Before you touch anything, sort at the folder level (not the file level) into three piles:

• Records — drawings, specs, O&M manuals, closeout packages, permits. These migrate.
• Working files — half-finished markups, "v2 FINAL FINAL" duplicates of things that also exist in a closeout set, exported PDFs of emails. Most of this can stay behind.
• Unknown — folders nobody can identify. Migrate them into a clearly labeled holding area rather than deleting; you can triage later, in the system, with search.

Superseded revisions are records, not working files. Keep them. A document system with version history is exactly where superseded drawings belong. What you're removing is copies, not history.

Step 1: Gather, don't reorganize

The instinct is to clean up the folder tree first — rename things, consolidate, build the perfect hierarchy — and then import. Resist it. Hand-restructuring 20,000 files is the four-month project you're trying to avoid, and a good system will build the folder tree from the metadata anyway.

Instead, collect files by source: one ZIP per old folder, per project, or per building. Keep the batches to a size you can review — a few hundred files is comfortable. The old folder path is itself useful information (it usually encodes the building or project), so note it for each batch.

Step 2: Use the index you already have

Many archives have a spreadsheet somewhere — a drawing log, a transmittal register, a CAD manager's index from 2009. It's incomplete and inconsistent, but every row in it is metadata you don't have to recreate.

A good bulk import lets you upload that spreadsheet as a CSV alongside the files and map your column names onto the system's fields. Your column might say Bldg or Sheet No. — you point it at Building or Sheet Number once, and every row follows. The only column that has to match exactly is the filename, which is how each row finds its file.

If your index covers half the archive, that's half the archive with clean, human-verified metadata on day one.

Step 3: Let AI read the title blocks for the rest

For the files with no index, the metadata is already written down — in the title block, on the sheet itself. Sheet number, sheet title, project name, the architect or engineer, revision, date, often the building and floor. What's missing isn't the information, it's the typing.

AI vision extraction reads each document and fills those fields in. In practice, three things determine how well it works:

Title blocks matter. Drawings with a proper title block extract very well. A photo of a napkin sketch does not. Set expectations by file type.

Combined sets confuse it. AI reads the opening pages of a document. A 200-page combined set will be described from its cover sheet, not sheet by sheet. Split multi-sheet PDFs into individual sheets first, then run extraction on those.

Some formats can't be read at all. Native CAD files (DWG, DXF) and some image formats (TIFF, BMP) are opaque to vision extraction. Those files still import — they just arrive without AI-filled metadata, and the import summary should tell you which ones. In a real construction archive, a list of skipped files is normal, not a failure.

One rule to insist on: AI should fill blanks, not overwrite. If a value came from your CSV or you typed it, it stays. AI only fills what's empty.

Step 4: De-duplicate by content, not by name

The same drawing exists in your archive three times: in the project folder, in the "current set" folder, and in an email attachment someone saved. Usually under three different filenames.

Filename-based duplicate detection catches none of these. Content fingerprinting catches all of them — the system hashes each file's actual bytes and checks against what's already in the organization before uploading. When it finds a match, you should see the list and decide whether to skip it or import anyway.

Run duplicates during import, not as a cleanup afterward. Removing 4,000 duplicates from a live archive is far more painful than never importing them.

Step 5: Import in batches, and verify a sample

Don't run the whole archive in one shot. Import a batch, then spot-check twenty files at random:

• Did they land in the right building and floor?
• Is the sheet number right?
• Is the revision right?
• Did scanned files get OCR'd — can you search for a word that appears on the sheet?

If the sample is clean, run the next batch. If it isn't, you've found a systematic problem (a column mapped wrong, a folder of combined sets that needs splitting) while it's still cheap to fix.

Large batches should process in the background. You confirm the import, close the tab, and the metadata appears on each file as extraction finishes. There's nothing to babysit.

Step 6: Deal with scanned paper honestly

The oldest drawings — original structure, buried utilities — are scans of prints. Vision extraction and OCR both work on them, but quality varies with the scan. Two things help:

• Re-scan the worst ones at 300 DPI or better. The dozen sheets people actually need are worth the hour.
• Accept partial metadata. A 1971 structural sheet may extract a sheet number and a title and nothing else. That's still enough to find it, which is the entire point.

Step 7: Fix the future before you fix the past

The archive got this way because every new project dumped files however it liked. The migration is the moment to change that:

• Adopt a file naming convention and hold new uploads to it.
• Route every project's closeout drawings into the facilities archive, not a project folder.
• Give contractors scoped access to the live system instead of emailing sets that fork immediately.

A realistic timeline

Following the steps above, a single building's drawings — a few hundred to a couple of thousand files — is an afternoon, not a month. A full multi-building campus is a few weeks of batching and spot-checking, done alongside normal work. The four-month retyping project never happens, because nobody retypes anything.

How ArchiveView handles it

ArchiveView's Bulk Import does this end to end, and it's available to every member who can upload — not just administrators. From the Upload page:

• Choose CSV + Files (your index maps onto our fields) or Files Only (one set of metadata for the batch, or let AI fill it in)
• Add documents individually or as a ZIP, which is unpacked into individual files
• Preview everything before a single file is written
• Duplicate detection fingerprints each file's contents against your organization and shows you matches before uploading
• AI Metadata Extraction reads title blocks — sheet number, title, project, firm, trade, building, date — and only fills blanks, never overwriting your values
• Small imports are analyzed on the spot; larger ones run in the background and you can close the page
• Scanned PDFs are OCR'd so their text is searchable, and sheet extraction splits combined sets into individual sheets
• Folders are built automatically from the resulting metadata — entity, site, building, floor, trade — so the tree you'd have spent months constructing by hand appears on its own

The largest archive on ArchiveView today came in this way: roughly 28,600 drawings for a hospital system across 10 sites, most of them decades old, many of them scanned.

Want to try it against a batch of your own? Start a free trial and run one building through Bulk Import this afternoon, or request a demo and we'll walk through it with your files.