China Merchants Bank (CIHHF): AI Credit Underwriting Cuts Service Time to 2.72 Hours

On 2026-08-31, China Merchants Bank Co., Ltd. (CIHHF) said on its earnings call that AI credit underwriting is now attached to all three stages of its corporate lending process - before a loan is granted, during the loan, and after disbursement - and that average service time has fallen to 2.72 hours [1]. The bank put no cost or credit-loss figure against the system.

China Merchants Bank is a nationwide joint-stock commercial bank headquartered in Shenzhen. Its revenue comes mainly from the spread between deposits and loans, plus fees from wealth management, bank cards and settlement. Corporate and small-business lending is one of its main channels for deploying assets.

Where the large model sits in the credit process

The model is attached at three points in credit work. Before a loan is granted, it drafts the main content of the due diligence report. At the approval stage, it carries microloan submission and approval and extracts key information to support the credit decision. After disbursement, it watches booked corporate exposures and pushes risk alerts to relationship managers. The users are credit approvers and relationship managers, and the objects are loan applications, due diligence reports and outstanding corporate credit. That places the application on the main line of credit work.

What changed since the March disclosure

The last time this workflow was discussed was the 2026-03-29 earnings call, and the measure then was coverage: about 82% of microloan submission and credit approval was done by the large model, microloan approval was 44% faster than the year before, and risk early warning came 42 days earlier than in the past. Management also explained that regulators require a bank's AI to be used alongside human staff [2].

The 2026-08-31 call switches yardsticks to process time and early-warning lead time, and the covered scope extends from microloan approval outward in both directions, to pre-loan due diligence drafting and in-loan information extraction. Up to 90% of the content in small-business due diligence reports can be drafted by AI, though the bank did not say how many reports are actually produced that way [1]. The two disclosures share no common metric. What can be compared is the widening scope of the application and the pace of the work, not a change in a reading on the same basis.

What the numbers measure, and what they do not

The 2.72 hours appears in the same sentence as 36 hours, as a before-and-after comparison of average service time in corporate credit [1]. The bank did not say which customer segment or which stage of the process it covers, and gave no measurement period. It is a one-time before-and-after reading, not a continuous series on approval turnaround.

On the post-loan side, risk alerts are triggered 45 days earlier than under the traditional mode. The comparison basis is explicit, but the bank never said how long the old alerts took, so the disclosure supports the size of the improvement and not the absolute timeliness of the warning itself [1]. Taken together, the two ends mean the same extension of credit absorbs fewer manual steps, and relationship managers can move on a deteriorating exposure earlier.

Why it has not shown up in the accounts

Saved staff hours point toward the cost-to-income ratio: AI drafts the main content of due diligence reports, the model executes most microloan submission and approval, and the same lending volume carries fewer manual steps, which followed far enough would push operating and administrative expenses down. That is a direction, not a result. China Merchants Bank does not disclose expenses for the credit business separately; the cost-to-income ratio and credit cost that appear on the call are bank-wide figures covering far more than this workflow, and management has never attributed them to AI.

Evidence runs the other way as well. Over the same period, non-performing and overdue indicators on retail small and micro loans were rising, and management attributed the improvement in the corporate non-performing loan ratio to property risk disposal and provisioning. Regulators require humans to work alongside AI, so an execution share such as 82% measures how much of the work the machine carries rather than showing that staff have been replaced - and that ratio combines the submission step and the approval step in one number [2].

What can be confirmed today is that the large model stands on the main line of China Merchants Bank's credit work, and that the gains in operating efficiency have specific readings behind them. What cannot be confirmed is where it sits in the accounts. The bank has not yet tied any change in credit-business headcount or in operating and administrative expenses to this workflow, or said how much of such a change came from AI. Until that disclosure arrives, the value of the application counts only at the level of operating efficiency, not as cost savings already realized.

Application assessment

• AI Credit Underwriting and Loan Monitoring | Business position: core operations | Deployment stage: limited production | Deployment scope: single business unit | Value type: cost reduction

Sources

[1] Drillr - China Merchants Bank Co., Ltd. (CIHHF) - 2026-08-31 - earnings call

"So the average service time is reduced from 36 hours to 2.72 hours."

[2] Drillr - China Merchants Bank Co., Ltd. (CIHHF) - 2026-03-29 - earnings call

"and also for corporate credit loan business and also the AI model is also helping them before loan granting and during the loan lending and also after that especially for microloans around 82% of the microloan loan submission and also credit approval is done by AI enlarged model"