DEEP Agent

Agents that finish the work, inside your walls.

Agents finish the work, from reading documents and checking them against your team's standards to entering them into core systems, and from answering questions based on internal regulations to analyzing data. Your staff only need to review the exceptions.

80+

Organizations served

We supply document automation agents to more than 80 organizations, including financial institutions and public agencies.

0.2s

Per document

Reading and verifying one document takes 0.2 seconds. Based on our own benchmark.

99%

Accuracy

On the same in-house benchmark, documents are read with 99% accuracy. Any case that falls short of the standard goes to a person.

2 weeks

Average deployment target

We aim to deploy in an average of 2 weeks without retraining. The timeline depends on the document types and the scope of integration.

Three kinds of agents, one workspace

Agents that process work, answer from internal knowledge, and analyze data are built and operated in one shared workspace.

Credit review agentRunning

09:12Received 12 applications, sorted by product
09:13Cross-checked income and employment information
09:14Calculated limits under internal standards
09:15Passed 2 exceptions to the person in chargeHuman check
09:16Entered 10 cases into the core system
09:17Drafted 10 review opinions
09:18Sent 3 requests for missing documents
09:19Recorded the audit log with its source documents
09:21Shared the results in the team channel
09:22Returned 1 application with its reason
09:24Queued the next batch of 12

Run the process

From intake to entry into core systems, agents process work in your team's order and hand only the exceptions to the person in charge.

# Credit Review Team

Park Ji-won10:24

@DEEP Agent Do personal business loans follow the same limit criteria?

DEEP Agent10:24

No. Sole proprietors have their own criteria: annual sales and years in business.

Source · Credit regulation 5.1

Park Ji-won10:26

@DEEP Agent What is the limit when there is no collateral?

DEEP Agent10:26

Without collateral only the annual sales rule applies, with a ceiling per industry.

Source · Credit regulation 5.3

Answer your people

Staff ask questions in the messenger they already use, and the agent answers based on internal regulations.

Analysis

Show last week's processed cases by team

Household lending2,906
Card review1,537
Corporate lending1,284
FX review948
Post-loan management612
Guarantee review431

The team with the highest share of human checks is corporate lending.

See across the team

Ask about scattered processing records in plain language, and the agent aggregates and compares them for you.

From the inbox to the core system

Reading is only the start. Each document moves through four automated steps, and the work is finished only when it is entered into the system.

Read

DEEP OCR

Reads documents received by fax, email, or photo without any form registration. Blurry scans are restored in image quality first.

Structure

DEEP Parser

Converts items, tables, and figures into key-value data. Mixed-in documents are grouped by business unit.

Check and act

DEEP Agent

Verifies authenticity, format, and rules, then makes a decision. Only the cases that fall short of the standard go to the person in charge.

Connect

Your systems

Enters verified data into account, review, and ERP systems. Existing legacy systems are connected through query-based integration.

After deployment, DEEP Ops keeps measuring processing accuracy and exception rates to maintain the same quality.

See DEEP Ops

Everything the work needs, in one workspace

Build agents in a workspace for each business area, decide where people step in, and keep operating them alongside your existing systems.

Domain workspaces

Separate workspaces for each business area, such as finance, insurance, public sector, healthcare, and manufacturing.

Ready-made work

Templates for each task, such as credit review, customer identification, and post-loan management, are already prepared.

Workflow blocks

Connect the steps your team follows as blocks, and rearrange the flow as often as needed.

People in the loop

Set the steps a person should check, and only the cases that fall below the standard go to the person in charge.

Answers with sources

Answers are based on internal regulations and approval history, and each answer links to the location of its source text.

Wired to your systems

Enter data directly into core, ERP, and CRM systems, and send results to existing systems through an API.

Versions before release

When a standard changes, the new version is compared with the previous one under the same standard before it is released.

Every change on record

Who changed what, when, and who approved it are recorded together with the basis.

Already running in the back office

We are already at work in back offices, where the most documents pile up. The original case studies are published on the KDL blog.

20,000

account opening files a month

Securities firm A

Each month, 20,000 ID cards and investor information confirmation forms come in. The agent verifies the items and their authenticity, enters them into the account system, and passes only the cases with incorrect formats to the person in charge.

Read the case

onboarding document types

Large capital company

Automatically classifies the 46 types of documents received during customer onboarding and extracts the information needed. Staff review mainly the exception cases.

Read the case

86%

less re-entry across 14 field sites

DEEP Agent Mobile

Applied first at 14 sites before launch, reducing the four steps of photographing, forwarding, and re-entering to two: photographing and web confirmation.

Read the case

Security

Runs inside your closed network.

Runs just as well in on-premise and closed-network environments fully cut off from external networks. Documents and models never leave your organization.