Which fields to define before document data capture
Define the fields before document data capture: the name of each field, where it sits on a sample, who confirms it, and where a blank or an unreadable value goes. Leave the documents unsent until someone has agreed the list. An amount field stays off the downstream path until a person has confirmed it.
On this page
Start with a field list
Each document type gets a field list before any extraction. Every field names five things: the field name, where it sits on a sample, what counts as an acceptable value, who confirms it, and which queue a blank or an unreadable value enters. Leave any of the five blank and leave that field out of extraction.
How extraction, review, and export meet the daily work is covered under Document processing and data capture. How to write a step that must wait for a person is covered in Which human gates to write before workflow automation.
The records standard names metadata
ISO 15489-1:2016, Information and documentation — Records management — Part 1: Concepts and principles, is edition 2. The catalogue page gives the publication date as April 2016. The life-cycle table marks 7 April 2016 as the date the international standard was published, and 9 November 2021 as the confirmation. The page says this edition was reviewed and confirmed in 2021, so this version remains current.
The abstract says the standard defines the concepts and principles from which approaches to the creation, capture, and management of records are developed. It lists records, metadata for records, and records systems; policies, assigned responsibilities, monitoring, and training; and processes for creating, capturing, and managing records. It applies regardless of structure or form. The public page does not list metadata element names. This note does not copy elements the page does not list, and it does not quote a price. The field list below is the one you write from your own samples.
Agree the list against samples
The agreement is a short record: the document type, the date the samples were read, the field list, and the person who agreed it. Keep the samples where that record can point, including copies that are clear and copies that are hard to read. This note does not set a minimum number of samples.
When the same document type has two layouts, write the second layout’s field positions on their own lines, or treat it as another document type. Using one layout’s positions on the other leaves blanks and wrong fields to be found after extraction. Until the person who agrees is named, the list is still a draft.
Five facts on every field
Use the field name the downstream system already uses, such as date, supplier, identifier, or amount. Write the position so a person can find it on the sample, such as the upper right of the page or the total row of the table. Write an acceptable value as a format, such as a date with year, month, and day, or an amount with a currency. If you have no format rule of your own, leave a generic example off the list.
The confirmer is the person who will change a value, return the document, or confirm it in the review queue. The destination for a blank or an unreadable value names the queue and the owner of that queue. The owner and the confirmer may be the same person. The list still shows the two roles apart.
A person confirms every amount
A field marked as an amount stays out of the downstream system, and is not treated as posted, until the named person confirms it. Confirmation checks that the figure on the sample and the figure in the field are the same, and that the currency is an acceptable value.
If an extraction tool emits a score, the cutoff for using that score has to be written by you, with a date. This note does not fill in a score. While the cutoff is unwritten, the amount field still waits for the confirmer. A tax identifier, a signature, and any field that would change a payment follow the same line: name the confirmer before the value moves downstream.
Send a mismatch to the review queue
Write the reasons a row enters the queue: unreadable, blank, inconsistent with the master record, or two samples of the same document type placing the same field in different positions. Each row carries the document identifier, the field name, the reason, and the time.
The queue has an owner. That person may change the value, return the original, or confirm it after looking at the sample. A queue with no owner is a pile. Until confirmation, the field stays unwritten downstream. When the value does not match the master record, check the master record named on the field list.
Write only confirmed fields
Only a confirmed field is written to the system you already use. Before the write, check that the destination field name matches the list. When that system has no such field, stop the write and name the missing field on the pending list. The same stop is described in How to connect equipment to existing business systems.
The export format is the one that system already receives. A new spreadsheet kept as the official record leaves the downstream record empty. This note does not name an extraction product.
When to leave capture off
When the field list has nobody who agreed it, an amount field has no confirmer, or the review queue has no owner, leave the documents unsent. Write those three first.
This note does not state a recognition rate or hours saved. The ISO catalogue date is April 2016, the confirmation date is 9 November 2021, and the page does not list metadata element names. Until a field list and a review record exist, check capture by whether every field has a name, a position, a confirmer, and a destination for failure, and whether an amount waits for confirmation before it is written.