What IDP is: a complete guide to making intake through registration one flow

IDP (Intelligent Document Processing) is a framework designed to handle intake, form identification, recognition, field extraction, validation, and system integration as one unbroken flow rather than as separate operations.

Why assembling functions does not reduce the workload

Lining up functions does not reduce throughput.

There is a common way of approaching document automation: buy a recognition tool, buy an automation tool, build a review screen separately. Each tool does its job. And the operator's hours do not fall as expected.

The reason sits in the gaps between the tools. Someone downloads the recognition output to check it, reformats it and uploads it again, collects the failures and reprocesses them. Each stage is automatic; joining the stages is manual. IDP is the perspective that removes those gaps. It asks not how well each function performs but whether intake through registration flows in one pass.

What IDP actually is: how it differs from assembling functions

Assembled functions and IDP across three axes

First, the unit of design differs. Assembling functions thinks in tools. IDP thinks in business process. Draw first where a document enters, which states it passes through, and where it is registered, then place functions on that map.

Second, the objective differs. Assembling functions aims to raise the performance of each stage. IDP aims to keep the flow unbroken. Where failures go and who receives exceptions are part of the design.

Third, the metrics differ. Assembled functions are read on per-stage accuracy. IDP is read on the share of received cases completed without human intervention and total handling time per case. High accuracy at every stage does not move those figures if the stages do not connect.

The two approaches are not in opposition. Poor performance in the individual functions means nothing flows however well it is joined. But as a matter of order, drawing the flow first and filling in the functions is the better sequence.

The five stages that form the flow

Five stages of the flow

First, intake. Electronic fax, branch scanning, mobile upload, email — the path changes the condition of the file, so it belongs in the design from here.

Second, identification and splitting. Several documents in one file are divided into units and each form is identified.

Third, recognition and extraction. Values are pulled according to the field definitions that match the form.

Fourth, validation. Format, relationships, and agreement between documents are checked by rule, and anything caught is separated as an exception.

Fifth, integration. Validated data is registered into the document management system and the business system, and the processing history is recorded.

How IDP is applied in practice

Draw the flow diagram first

The first deliverable is not a technical specification but a processing flow diagram. Draw where documents enter, the input and output of each stage, where failures go, and the points at which a person intervenes.

With that picture, the bottleneck becomes visible. In real cases the bottleneck has often been not recognition but form identification and splitting. Choosing tools before drawing the flow misses it.

Design the exception path as part of the flow

Thinking about exceptions afterwards means cases pile up on an operator's desk once production begins. Include in the diagram which conditions create an exception, which screen it surfaces on, and how it returns after confirmation.

With the exception path designed, the exception rate itself becomes an operational metric. A rising figure points at new forms entering the flow or falling document quality, and the cause can be traced.

IDP in the Korean environment

Fax remains part of the intake path in Korean finance and the public sector, and Korean word processor files, scanned PDFs, and image files coexist within one workflow. Format-specific preprocessing has to be separated out from the moment the flow is drawn.

On top of this, network separation requires the whole flow to run inside the internal network. A configuration performing recognition internally while placing the review screen or the integration platform outside drops out at this point.

Audit requirements come alongside. Who confirmed and approved what, at which stage, has to be recorded, so noting those points on the flow diagram is worth doing.

Frequently asked questions