Use caseArchives ahead of a migration
A landlord holds eleven years of tenant scans: 412,000 files in date-named folders, no index. The new repository goes live in April.
The share is measured before anything movesEach document gets a type, a lease and a retention ruleThe archivist validates what the AI cannot place
One point needs a person · the archivist confirms or corrects
One file, one tenant file: a lease on pages 1 to 6, its appendix, a move-in inspection on pages 9 to 12, a letter, and one page to review.
source · 0031.tif, 14 pages412,000 files, and the only description any of them carries is a scan date. Each one holds several documents for several tenants. To find one lease, someone opens folders until it shows up. The archive is unreadable as a collection, and if it moves as it is, the new repository inherits the same search.
The archive gets read before it moves. The operations engine measures the share in place, the AI splits each scan, types each document and links it to its lease, and the archivist's rules set its retention. Every document reaches the new repository with a type, an owner and a rule, and a person has validated what the AI could not place.
Taken one by one, these screens are routine. The project lives in their order: the backlog is measured in place, read on a sample, run at scale, and moved in waves that each leave a trace. Ryan and Claire take it a year at a time below, and so do the steps.
Before any document is typed, the team learns what is on the share and agrees where each kind of document belongs.
Ryan, the IT project lead, points the operations engine at the archive share. It lists every folder and file with its format, its page count and its date, and groups exact duplicates. Nothing on the share moves during the scan.
Claire, the archivist, reviews a sample of 300 files with the rental team. They agree on 18 document types, each with its metadata and its retention rule from her records schedule. The target filing plan has one folder per building and one per lease.
Ryan runs the chain on the 2016 folders: 31,000 files. Claire checks 400 documents drawn at random and finds the type right on 391. Most of the nine errors are tenant letters typed as complaints. She adds a few of them as examples, and the rules are replayed on the year.
Eleven years in nightly batches. Each document needs a type, a lease and a date before it can be filed, and some need Claire.
A single TIFF often holds a whole tenant file. The AI splits it document by document and classifies each one on the landlord's 18 types. Pages below every threshold are set aside for a person.
The AI extracts tenant name, address and dates from each document. The operations engine reads the housing management database, which holds every lease since 2004, and associates the document with the one that matches. Documents with two candidate leases, or none, go to review.
Claire's queue shows each document with what the AI proposes and why it stopped. For two leases with the same name in the same building, the date usually settles it. She asks the assistant which lease was open when the letter was written, checks the answer, and confirms.
What is left out is approved, what moves goes in waves, and every wave leaves a trace.
The engine proposes three kinds of documents to leave out: exact duplicates, documents past their retention rule, and blank pages. Claire reviews the lists by type and year. Two leases are under a court case, so a legal hold keeps their documents whatever their age.
Each closed year moves to the new repository as one wave. On the way, each document is converted to PDF/A, filed under its building and lease, and its metadata and retention rule travel with it.
Every Monday Ryan shares progress with the steering committee. Each wave ends on a reconciliation report: what left the share, what arrived, what was set aside and why. The next wave starts only when that report is clean.
When a year closes, each of its documents is filed, waiting for Claire, or approved to stay behind on the share.
The type is above its threshold, one lease matches, the rule comes from the schedule.
What follows: the document moves with its year's wave.A type below threshold, two candidate leases, or none. The queue says which, with the line it rests on.
What follows: the archivist confirms or corrects, and the document joins its wave.An exact duplicate, a document past its rule, a blank page. The engine lists it with the reason.
What follows: the archivist approves the list; a legal hold keeps what it covers.It sets no retention period. It leaves nothing out on its own. It picks no lease between two candidates. Across the nine steps on this page, every document left out and every uncertain link is approved by the archivist, with her name on the list.
The AI splits, types and links. The rules are written by the archivist and tested on a sample, so they hold on eleven years of scans.
Measured on this landlord's backlog. An archive with other formats or other retention rules will land elsewhere.
Where these numbers come from. They are counted on the scenario described on this page, with example data. They are not a customer result.
The project lead, the archivist and the rental team are the ones who live with the result. Here is the difference for each of them.
Nobody can describe the archive you are asked to migrate.
The retention rules are yours, and nobody can apply them to unindexed scans.
One lease, and folders to open until it shows up.
With the inventory. Point the engine at the share and read the report: volumes, formats, duplicates. It changes nothing and sizes everything.
Then one year as a pilot, checked by your archivist on a sample, before the rules run on the rest.
Your move
Give us read access to a sample of your scanned archive. We will inventory it, run a pilot on one year and go through the results with your archivist.