The file server of a manufacturing group with two companies holds more than a million documents in the first and around seven hundred thousand in the second, with forty years of history. Many PDFs are scans of paper, some printed and then scanned again. Folders have very long names and files follow no common naming convention. Importing even five or ten thousand of them would mean labelling them one by one, and the head of innovation put it bluntly: a job nobody would ever have finished. Document archive migration gets moving once the history stops being treated as a single block to be transferred.
Document archive migration: how much work does the history require?
The effort depends on the type of document: native PDFs with a recurring layout are read automatically, while scans and mixed archives require a preliminary analysis.
In a document archive migration this distinction is the first thing to measure. A PDF produced by the ERP, with the data always in the same place on the page, can be read by position: the location of number, date and addressee is known in advance, and the record fills itself in. A scan from the 1990s requires optical character recognition or an artificial intelligence module. An archive with folders already sorted by customer and by type is imported by inheriting that structure, with the addition of data read from the document.
There are also simpler routes, often underrated. The file name can contain a unique code, such as number, date and series, that links it to data already held in the ERP. A watched folder receives files produced outside the system. Documents generated by the ERP come in automatically two or three times a day, classified with master data without anyone opening them.
In a company that kept its documents in folders per customer, alongside SAP, the estimate was around twenty thousand documents a year, including PDFs, photographs and emails dragged out of Outlook. Management ruled out entering them through forms. It asked instead for an artificial intelligence search across the existing archive, and the documents produced by SAP are classified by reading the data directly from the ERP.
The initial count calls for caution
The scripts Aesir runs on the file server count documents and folders, but they do not tell main documents from attachments, and each office often keeps its own copy of the same files. With high volumes the import has to rely on file names or on the folder structure. Across hundreds of thousands of documents, entering the data by hand is not feasible. The project therefore starts with a day of analysis, and only afterwards is the work estimated in detail.
The active archive matters more than the history
In the manufacturing group described at the start, management stopped asking how to import forty years of documents and established which ones daily work actually needed. The answer narrowed the scope to the last four or five years, with artificial intelligence supporting classification. The proposal is to start from new job orders, with documents divided by category and access governed by role. Customers and job orders are read from the ERP, where the master data history had already been imported. The head of innovation remarked that the wall he had faced was gone: a two-hour meeting is enough to define the labels for the new archive.
The risk that worried the team most was double access, with the old file server and the new system open at the same time. A clear scope reduces it, because new job orders are created directly in the document management system. What happens to the rest of the archive still has to be decided, from the analysis stage onwards. The tidiest arrangement leaves the old file server available read-only, with no new saves, until the selected history has been imported or for as long as the excluded documents are still needed for audits and retention obligations.
Insight: in a document archive migration, manual work concentrates on documents with no recognisable rule, such as the oldest scans and folders named by hand. The scope is decided by counting how many of those documents daily work still needs.
Two hundred thousand documents, same criterion
A group with several businesses follows the same logic, with four or five terabytes on the file server and an estimated two hundred thousand historical documents. Its project starts from a single process with low volumes and excludes documents older than five years. Under the producer’s terms the history must be imported within the first year of the licence, a constraint to build into the project plan from the offer stage. For documents containing personal data, bringing into the new system only what is needed is consistent with the storage limitation principle set out in the GDPR.
Automatic document classification: can the system recognise the type on its own?
Yes: a classification module trained once on the company’s document types recognises the document, fills in the record and assigns each field a reliability level.
For a document archive migration one effect matters most: manual labelling, which stalled the project in the case described at the start, is reduced to checking doubtful cases. The module is trained once per document type, with many examples of that type and of other types. There is no need for a model for every layout: the module recognises fields even when they move around the page, and with layouts it has never seen it returns a lower reliability level. For each field it reports where the value sits on the page and how reliable the reading is. Below a threshold set by the company, for example 70%, it leaves the field empty. It can also open a verification task, in which a user checks the values before saving. Ready-made models already exist for invoices, receipts and contracts.
Incoming emails follow the same path: the message is archived and the attachment classified automatically. People working in Word or in another program enter only the data required, and the system files the document in the right place. Saving it elsewhere remains possible. The owner of a professional practice summed up the requirement well: it has to be a deliberate action.
| Starting material | How it enters the system | Work required |
|---|---|---|
| PDFs produced by the ERP | Data read from the ERP or position-based reading | Rules defined once |
| Folders with an orderly hierarchy | Import that inherits the structure | Spot checks on classification |
| Scans and mixed archives | Optical recognition and AI classification | Preliminary analysis and training |
| Emails with attachments | Message archived and attachment classified | Choice of mailboxes to monitor |
Where the model runs and who sees the answers
In every document archive migration the question of confidentiality comes from IT, often in the same words: confidential data must not leave the offices. The artificial intelligence modules run on Microsoft Azure, with data in the European region and a space reserved for each client. Documents are not used to train public models. In the client’s space a consumption cap is also set, with an alert at 80% and a block when the limit is reached.
Permission checks come before every answer. In the files each user sees only the documents they are authorised to consult, and the conversational assistant checks permissions before sending any content to the model. If a user asks about a document they cannot see, the answer states that the necessary information is not available. The administrator can restrict the assistant to certain groups or document types and prepare ready-made questions.
Semantic search returns documents ranked by relevance, even when the person searching does not remember the file name. Every change creates a new version, with a record of who did what, downloads included. The principle is the one described in the article on AI on company data: the result depends on which data the model can reach, and with which permissions.
What to measure before a document archive migration
Before asking for a quote for a document archive migration, it pays to gather a few figures, even rough ones.
- How many documents and folders the file server holds, separating main documents from attachments.
- How many years of history daily work needs, and how many are needed only for audits.
- Which documents originate in the ERP and can be linked to its data.
- What share consists of scans and what share of native PDFs.
- Where duplicate copies kept by individual offices exist.
- Who needs to see what, by role and by document type.
- Who will look after new forms and new workflows after go-live.
Aesir Srl is an ARXivar system integrator: it configures the platform, connects it to the company’s systems and supports its adoption, across Italy and, on specific request, in Switzerland, the United Kingdom and the United States. The link with the ERP relies on programming interfaces and on reading the database directly. Aesir is a Gold Partner of NTS Informatica for Business Experience, and with other ERPs the available connectors are used or new ones are built. The platform can be installed on premises or in the cloud, according to the client’s IT policies.
Questions about the document history import
Can the import of the history be postponed?
A project can start from a single process or from new job orders and bring the history in at a later stage of the document archive migration. The producer’s terms require the import within the first year, so this phase should be planned from the offer stage.
Does the artificial intelligence have to be trained on every format?
Training is done once per document type, not per layout. With a format it has never seen, the module returns a lower reliability level and leaves uncertain fields empty.
Who manages the system after go-live if there is no one in-house?
There are two routes, and they can be combined: a service paid by the day through which Aesir builds new forms and workflows over time, or training for an internal person, including the producer’s own courses.
Let’s talk
In a document archive migration the first useful meeting produces a concrete list: which years and which document types enter the first scope, and what stays on the old archive in read-only mode. From there come the labels, the first process and the way the history will be brought in, on a timetable the company can sustain. For documents that must keep legal value over time, preservation follows the rules described in the article on compliant digital preservation.
If you would like to explore the subject or assess the situation in your own company, you can fill in the form at the bottom of this page or write to support@aesir-tech.it: we will arrange a free consultation and start from your numbers.