Automatic tagging: an end to manual filing

Automatic tagging means that an AI reads every incoming document and classifies it on its own – it fills in customer, project, installation, date and topic directly. Only the document type is presented as a suggestion for a person to approve, because the document type determines who gets to see the document. This replaces the step where orderly filing breaks down in practice: manually sorting documents into folders and naming files according to a scheme. The company sets the framework, the AI fills it in, and a person confirms – quickly, for clear-cut documents.

In brief

• Automatic tagging classifies documents based on their content – nobody has to choose folders, assign file names or memorise filing rules any more.
• From the sender, the content and the company’s known master data, the AI works out who a document comes from and what it belongs to – it enters customer, project, installation, date and topic directly.
• The company specifies the categories; the AI works within this framework – so the order stays consistent rather than creative.
• The document type is always confirmed by a person, because visibility depends on it; beyond that, the review area only asks about what remains open – which is why clear-cut documents go through in a single click.
• Records such as invoices stay under the control of the ERP and accounting – tagging mainly organises knowledge: reports, manuals, correspondence, project documents.

Why does manual filing fail?

Every company has at some point agreed on filing rules: which folder, which naming scheme, which sub-structure. And in almost every company, three years later the folders look as if they just grew rather than being planned. The reason is always the same: manual filing is an extra step with no immediate benefit of its own, and steps like that are the first to be dropped under time pressure. On top of that, everyone classifies things differently – what one person files under “Maintenance” another files under the customer’s name.

The result is not just disorder but mistrust: anyone who does not trust the filing system makes their own copy to be on the safe side or keeps documents in their mailbox. That makes the order fall apart even further. The way out is not stricter rules – they fail for the same reason – but reversing who does the work: the system does the groundwork, and the person only confirms what they need to decide.

Which attributes does the AI recognise in a document?

Automatic tagging reads the content of the document – not just the file name – and works out the attributes that make it findable. It fills in all but one of them itself – the document type remains a suggestion:

Attribute	How the AI recognises it	What it is used for
Document type (suggestion, approved by a person)	Structure and typical terms: a test report looks different from a quote	Filtering by type: “all maintenance reports” – and controlling who may see them
Customer / supplier	Sender, letterhead, names mentioned, matching with master data	Everything about one business partner at a glance
Project / matter	Project names, job numbers, addresses in the text	Assignment to the right site or installation
Date	Document date taken from the content, not just the scan date	Placing it in time: “the report from January 2023”
Topic / content	The summary of the document’s content	Questions in everyday language, even in different words from those in the text

Matching against the company’s master data matters: because the system knows the customers, suppliers and ongoing projects, it assigns a delivery note from the Berger company to the Berger matter – and does not invent a new category “Berger GmbH & Co” just because the letterhead looks different for once.

How reliable is automatic assignment?

The honest answer has two parts. The AI handles clear-cut documents – a quote with a clear letterhead, a report with a job number – very reliably; that is most of what comes in each day. But there are always ambiguous cases: a letter with no clear project reference, a new supplier who is not yet in any master data, a batch scan with three documents in one file.

What matters is how a system deals with this uncertainty. A good system does not guess, it asks: the review area shows exactly what the document did not make clear, and you confirm or correct it in two clicks. One question is always asked: approval of the document type – it determines who gets to see the document later, and that decision stays with a person. For a clear-cut quote, that is one click; for a batch scan without a sender, it will be a few more. The system learns from the corrections for the future. This reverses the relationship – instead of sorting one hundred per cent of documents by hand, you confirm groundwork that has already been done. And even an incorrect assignment is no disaster: unlike a file in the wrong folder, an incorrectly tagged document can always be found via the content search.

Who decides the categories – the AI or the company?

The company. This is not a minor detail; it is what makes the order fit the business. During set-up, you decide which document types exist, which projects and topics are relevant and which terms the company uses – a plumbing and heating installer thinks in terms of installations and maintenance contracts, a tax firm in terms of clients and deadlines. The AI then works within this framework instead of making free associations – the document type it presents for approval also comes from this list.

This framework also does away with the old worry about “tag chaos”, where every employee invents their own terms and in the end “vehicles”, “cars” and “fleet” exist side by side. This kind of inconsistency, which has always been the downfall of manual tagging, disappears when all suggestions come from a single set of rules. Maintenance remains manageable: new projects and new document types are added when they arise – a matter of minutes, not an ongoing task.

What does this mean for invoices, delivery notes and other records?

To be clear about one thing: for records, the system of record is and remains accounting or the ERP. That is where invoices are checked, posted and retained in accordance with statutory periods – a knowledge base does not replace this and is not meant to. The ERP knows what happened: which order, which invoice, which payment status.

The knowledge base adds to this where a record needs context – and above all where no record exists at all: in the how. The setting report for the installation, the correspondence about the complaint, the manual written by an experienced employee, the photo of the type plate. These documents never had a system of their own – they were scattered across folders and mailboxes. For them, automatic tagging is the step that turns scattered files into orderly, findable company knowledge.

How do you introduce automatic tagging?

In practice, tagging is not a separate project but the engine behind document capture as a whole: whatever comes in from email mailboxes, from the scanner, from mobile phones or from network folders automatically goes through text recognition (OCR) and tagging and is then ready for approval. It is therefore set up at the same time as the first sources are connected – the overview Connecting documents and systems shows which collection to start with and the best order for connecting the rest.

In junova, automatic tagging is at the core of the system: the framework is set up together with you, the system sets the field values itself, and you approve the document type in the review area – so you stay in control of who sees which documents. In addition, every AI answer names its source document, and your documents stay on EU servers without being used to train third-party AI models. The implementation process page describes the roll-out step by step – from defining the categories to day-to-day use.