Back to the newsroom

Op-ed

The substance of financial AI lies not in storing data (DX) but in how complete its cognitive automation (AX) is

Korea Deep LearningPublished

The substance of financial AI lies not in storing data (DX) but in how complete its cognitive automation (AX) is
In a May 2026 column for AI Times, Korea Deep Learning argued that the essence of financial AI lies in how complete the AX that turns unstructured data into business intelligence is.

Kim Dong-hyun, head of Korea Deep Learning's Future AI Development Team, contributed a column to AI Times titled "The substance of financial AI lies not in storing data (DX) but in how complete its cognitive automation (AX) is."

The illusion left by digitization

The column first points to an illusion in the financial sector's digital transformation (DX). Digitization, which turned paper documents into PDFs and images, worked for preserving records, but it also piled up large volumes of unstructured data that could not be used right away in daily work. The diagnosis is that data that only grows in storage, without intelligence, actually increases operational inefficiency.

"Just because data has been converted into a digital format does not mean the information in it can be used in work right away."

The author defines the AX that companies should aim for as the "internalization of intelligence," in which AI grasps and acts on the business logic in data on its own. If DX was the transition that moved analog into digital, AX is about having an engine that interprets data to create value, the author explains.

The limits of technology that reads characters, and document intelligence

Financial documents are where this difference shows most clearly. In documents such as insurance claims and financial statements, where complex forms and logical relationships between figures are intertwined, OCR that only reads characters inevitably reaches its limits. The author believes that VLMs, which process visual and linguistic information together, make possible the "document intelligence" that understands a document's structure and meaning.

As evidence, the author cited the OCRBench v2 results of KDL Frontier, the company's own document-specialized VLM. It scored 85.4 in visual text understanding, first among 8 categories, and 92.2 in relation extraction, and also ranked ahead of global big tech models in the overall rankings. The author reads this as evidence that an architecture optimized for a specific domain can carry out work more precisely than a general-purpose giant model, and stresses that the depth of domain training matters more than the volume of data. This ability to understand structure also carries over to unstructured Korean documents that mix handwriting and complex forms, the author added, and it has been verified in the real environments of public institutions and major financial companies.

Financial competitiveness lies in the speed and accuracy of AX

The conclusion comes down to where a financial company's competitiveness comes from. In a conservative financial market, the value of AI is decided by accuracy, and companies should not stop at buying solutions but should set their own AX standards for turning unstructured data into assets, the author recommends.

"The true competitiveness of a financial company depends not on the volume of data it holds, but on the speed and accuracy of the AX that turns it into real business intelligence."

The full column was published in AI Times.

About Korea Deep Learning

Korea Deep Learning (KDL), founded in 2019, is an enterprise AI agent company. Built on KDL Frontier, its own vision-language model (VLM) specialized for documents, KDL makes AI agents that take on and complete the document work of businesses and public institutions. Its main products are DEEP OCR, DEEP Parser, DEEP Agent and DEEP Ops.

Source