Technology and architecture
What Is Cross-Document Verification? Matching Multiple Documents as the Final Step of Automating Review, Explained Completely
Korea Deep LearningPublished
Cross-document verification is the process of pulling the value of the same item from several documents that make up one work case, comparing them to check whether they match, and passing on to a person only the items that do not agree.
Why a Single Document Is Not Enough to Finish the Work
Screening ends with comparison.
If you watch a financial review at work, the thing the officer spends the most time on is not reading. It is laying out the merchant enrollment application, the business registration certificate, the bankbook copy and the certified corporate registry side by side and checking that the representative's name is the same, the account number is correct, and the business registration number matches.
Measurements show that document comparison and verification alone take a considerable amount of time per case. In an environment where one case brings in several kinds of documents, this work takes up a large share of the total time. However accurately each document is read, automation stops halfway as long as this step remains.
The Precise Definition of Cross-Document Verification: How It Differs from Item Verification
Three Axes Along Which Item Verification and Cross-Document Verification Differ
First, the target differs. Item verification targets values within a single document. It checks whether the digit count is right and whether the date format is right. Cross-document verification targets the same item scattered across several documents.
Second, the way it judges differs. Item verification compares against a defined format. Cross-document verification compares documents with each other, so there is no separate correct answer. It only confirms that a mismatch exists, and a person decides which side is right.
Third, the problems it exposes differ. Item verification catches recognition errors and format errors. Cross-document verification also exposes inconsistencies in the documents themselves and the possibility of forgery or alteration. In one actual case, comparing the item values across several documents automatically isolated a single inconsistency.
The two verifications are not alternatives to each other. Values that match in format can still disagree between documents, so both are needed.
Five Conditions for Making Cross-Document Verification Work
Five Conditions for Operation
First, documents must be grouped by work case. If they are handled one sheet at a time, there is nothing to compare them with. The step that splits the received files and determines which case each belongs to must come before this.
Second, the correspondence between items must be defined. A person must decide in advance that the representative's name on the application and the representative director on the registry are the same item.
Third, notation differences must be absorbed. The same entity is often written in full formal name in one document and in an abbreviation in another. There was an actual case where bank name standardization was needed.
Fourth, time order must be handled. For items where sequence carries meaning, such as transaction history and reported date and time, the check is not only whether the values are equal but whether they are in the right order.
Fifth, the nature of the mismatch must be distinguished. A mismatch caused by a recognition error and a mismatch inherent in the documents call for different responses. Presenting the location in the original lets the reviewer separate the two quickly.
How to Apply Cross-Document Verification in Practice
Draw the Comparison Table First
The first output of the design is a comparison table. It records which items are compared across which documents, which document is used as the reference, and how a mismatch is handled.
With this table in place, the layout of the review screen follows naturally. It becomes a form that shows each document's value for each item side by side and marks the cells that do not match.
Set Mismatch Handling Criteria Together with the Business
Not every mismatch is a reason for rejection. A difference in how a floor is written in an address can be let through, but a difference in an account number must be checked without exception.
If you agree in advance with the business owner how much difference is tolerated for each item, far fewer cases fall through to exceptions.
Cross-Document Verification in the Domestic Setting
In domestic financial work, one case often brings in around six kinds of documents, and many of them are faxes or photographs. The same value comes out at different quality in each document, so absorbing notation differences is especially important.
In public work, forms are revised often, so the name of the same item can differ from one document to the next. Managing the item correspondence as data and keeping its versions means you do not have to rework everything at each revision.
Choose the Items to Compare by Business Criteria
You do not need to compare every document on every item. Selecting only the items the business actually uses for judgment reduces the processing load and makes the meaning of a mismatch clear.
In financial screening, the items commonly chosen are the representative's name, business registration number, account number, amount and date and time. In one bank's damage-relief automation, the name, account number, bank name, contact information, claimed amount and date of the incident were compared across three kinds of documents, and out of 46 items the process automatically found a single mismatch in the bank name, where a formal name and an abbreviation had been mixed.
What this case shows is that how to handle notation differences must be decided in advance. Notation differences that should be treated as the same value are standardized, and only differences that should be treated as different are raised as mismatches, so reviewers do not accumulate unnecessary checks.
Frequently asked questions
How many documents can be compared?
There is no limit. In actual work, an average of around six kinds arrive per case, and as long as the item correspondence is defined, the number of documents is not a constraint.
How are values with different notation handled?
They are absorbed by standardization rules. For items such as bank names or company names where the abbreviated and full forms are both used, a dictionary is kept so that they are recognized as the same value.
Does it also decide which value is correct?
No. It presents the fact that a mismatch exists, the value in each document, and the location in the original. The judgment is made by a person.
Can it also be used to detect forgery?
It provides a clue. Points where values disagree between documents become items to check, which narrows down where the officer should focus.
Which items should be compared?
Select only the items that the business uses for judgment. In financial screening, the representative's name, business registration number, account number, amount and date are often used.
If the notation differs, is it always a mismatch?
No. Differences that the business treats as the same value, such as a formal name versus an abbreviation, are better standardized, and only the differences that should be treated as different should be raised as mismatches.
Send us the documents that come in for one work case, and we will send back the comparison results as they are.
Contact us



