SMBC sets up agentic AI venture in Singapore

Account opening may drop to two days and loan processing to five.

Sumitomo Mitsui Banking Corp. (SMBC) has established an agentic artificial intelligence (AI) startup in Singapore to speed up development of enterprise AI tools and attract talent that traditional banking structures often struggle to secure.

The venture, first announced in June 2025, is designed to operate with greater independence so it can experiment more quickly, test automation solutions in controlled settings, and sidestep the lengthy internal processes that usually accompany innovation inside big financial institutions.

“We are trying to bring the best of both worlds: having an independent entity incubated and invested by SMBC to attract talent, do experiments faster, and have the nimbleness and agility,” Mayoran Rajendra, managing director and head of the AI transformation office at SMBC, told Asian Banking & Finance.

Unlike generative AI that produces content from prompts, agentic AI can make decisions, take action, and work through multi-step objectives with limited supervision. It can pursue complex goals autonomously whilst minimising the risks common to broad, open-ended language models.

Rajendra said this approach helps the bank reduce uncertainty and variability in processes by narrowing workloads into smaller, defined steps. “You are not overly relying on a large language model (LLM), but the small, small judgements that are needed in a repetitive workflow,” he said via Zoom.

SMBC is focusing initial development on onboarding of corporate clients and know-your-customer (KYC) reviews. These areas involve heavy documentation and significant labour hours, especially for multinational clients using several SMBC services, including loans, trade finance, remittances, and foreign exchange.

In many cases, compliance teams must sift through 400 to 500 pages of documents for a single corporate customer to obtain the required data points for each jurisdiction.

Rajendra said agentic AI could take over much of this work by extracting information, checking accuracy, and validating data with multiple authorised sources.

With automation, corporate account opening may drop to two days from five, he said. Loan processing, which can take as long as about seven months depending on the borrower and country-specific regulatory requirements, could fall to five days.

Rajendra said automation would not remove human oversight. SMBC designs its agentic workflows so staff can always view the exact text or source page from which a data point was extracted.

“You automate a lot of the repetitive work, but we always design it in a way that is auditable, traceable, observable,” he said. “There is always a possibility that it cannot be 100% [accurate]."

To address data-security concerns, SMBC keeps all information processed by AI strictly within its own infrastructure. Sensitive data is not fed into external models, nor is it used to train open LLMs. “First and foremost, we make sure any information is not going to an LLM to train the LLM. It sits within SMBC premises,” Rajendra said.

Beyond AI development, SMBC is rolling out other tools for corporate clients, including free CFO (chief financial officer) dashboards that provide real-time visibility into operations through analytics and data visualisation. The bank also invests in startups across India and Southeast Asia through the SMBC Asia Rising Fund.

Rajendra said the goal is straightforward: “We are bringing these agentic AI applications or capabilities to support our customers’ operation automation.”