One result, any number of recipients
At the end of the chain is a uniform extraction result that is neutral with respect to destination and procedure: fields, tables, and source references in a form that makes no assumption about the recipient. Where it goes is therefore not a question of processing code, but a configuration question — and that is exactly what creates the following possibilities.
Number of destinations
From a single recipient to several at once — delivery distributes the same result to all registered destinations in the same run.
Type of destination
Document repository, business application via a web interface, or data export — each destination type is an interchangeable building block behind the same handoff layer.
Weight of the destination
Each destination can be marked as mandatory or optional. Only mandatory destinations determine whether a run counts as complete.
Multiple reuse without extra work
A document is read once and reused multiple times: audit-proof storage, operational further processing in the business system, and analysis can all receive the same result in parallel. Until now that meant multiple capture steps — now it is a destination list.
Switch destination systems without touching the process
A system change can be run in parallel operation: first write the new destination optionally, compare results, then reverse the binding and switch off the old destination. Classification, analysis, and test runs remain unaffected.
Different paths per document type
Because the destinations depend on the recognition model and the model depends on the document type, the classification on page 05 indirectly also determines the output. One document type can flow into accounting, another into contract management — without branching in the process.
Connect instead of programming
Address, method, field and table mapping are configured; values are inserted via named placeholders and formatted appropriately for each destination. Deliberately without conditions, loops, and functions — so that a connection remains verifiable from a business perspective and no loophole for code is created in the configuration field.
Attach instead of creating twice
Delivery can first search for the matching target object — customer, case, file — and only then write. This means extracted data ends up on the existing business object instead of creating duplicates.
Partial failure instead of total failure
If a downstream system is unavailable, the entire run does not fail. Optional destinations report their error and do not block anything; a failed mandatory destination leaves the case repeatable rather than closing it half-finished.
The benefit lies in the decision stage: whether an unknown reference is a new record, a review case, or an abort is a business rule — and it is configured per destination, not hardwired into the process.
Classification makes the start of the chain automatable, the neutral handoff layer makes its end interchangeable. Between the two lies processing that does not need to care about either the origin or the fate of a document — and that is exactly why it can grow without redesign.