MYbank opens Bailing 2.0, a small-business finance Agent, to 42 million merchants

At the 2026 Bund Conference, MYbank disclosed the first public details of its AI banking rollout. On the back end, AI has entered core production workflows including risk control, manual review, marketing, product and engineering; on the front end, Bailing 2.0 — described as the world’s first inclusive-finance Agent for small and micro businesses — is now open to 42 million merchants.

A case reported by QbitAI shows how that entry point works. A restaurant owner who has run one store for two years spots a 65-square-metre street-front space and wants a second location. He types a single line — “I’d like my credit line raised” — with no other context. Bailing reads his prior business and credit records and asks about the new store’s location, size, opening date, use of funds and how much he can contribute himself, turning a vague request into a concrete goal: funding a second restaurant.

As the conversation continues, the Agent works out that the new store needs about 600,000 yuan — roughly 450,000 for branding, fit-out and equipment, and 150,000 for rent, initial inventory and working capital. With 200,000 yuan of his own money, the gap is around 400,000 yuan. Combining the first store’s cash flow, industry norms, the local market and the opening plan, it analyses expected revenue and repayment pressure, then raises his available credit line from 150,000 to 400,000 yuan, drawable in stages as the store opens, with interest-only payments for the first six periods. The whole process takes ten minutes.

The chat window is only the interface; the more consequential changes sit on the back end. MYbank has built eight AI workbenches across its operations and highlighted four at the conference. Qianliyan, or “Clairvoyant Eye”, pushes risk analysis from customer segments down to individual customers. A photography chain in Suzhou carried two risk labels — weak rating and high debt — which under traditional logic would trigger an immediate rejection. Tracing the decision chain, the system found the owner had opened nine stores across six cities in two years with steady monthly revenue above 700,000 yuan. After specialists verified the evidence item by item, the rejection became a 300,000-yuan facility matched to his business cycle.

On the approval side, Dinghaizhen, or “Sea-Anchoring Needle”, encodes human reviewers’ experience into the model. In an internal test, a merchant selling storage devices online applied for a higher limit; the Agent produced a preliminary “no increase” within seconds, citing a guarantor’s overdue payment, weak peer support and several recent loan applications. Human reviewers then established three missed facts: the guarantor was repaying on schedule under a restructuring plan, e-commerce is asset-light so weak peer support says little about the business, and the loan applications followed storage-price rises that required buying inventory early. Re-reasoning on the new facts, the system approved a 1.8 million yuan credit line.

Baoliandeng, or “Lotus Lantern”, addresses demand that spikes suddenly. When a typhoon passed through Guangxi in July, a mandarin grower in Wuming, Nanning, typed “I’ve been hit by the disaster” into Bailing. The system cross-checked the conversation, satellite imagery, business cash flow and regional events, concluded that growers and breeders — plus agricultural-input dealers upstream and cold-chain logistics firms downstream — were affected too, and generated tailored repayment deferrals and interest-free plans from past disaster playbooks. From his first message to tens of thousands of customers in that chain receiving a plan took one hour.

On the engineering side, Jindouyun, or “Somersault Cloud”, produced a counter-intuitive finding: writing code faster is not enough. MYbank says AI coding has doubled as a share of its development work and that 15% of business change requests are now completed by AI, but it stresses this is not “vibe coding” — code still goes through the same compliance reviews, testing and audits. By having AI generate interactive, high-fidelity prototypes, convert requirements into precise system analyses and run sub-agents in parallel, a weather-alert feature called Nongxiaobao reached a customer-verifiable version in two days, a process that used to take at least a month.

Banking is an industry that cannot afford mistakes: one bad credit decision becomes a non-performing loan, one compliance slip invites regulatory attention. MYbank’s answer is “Agent First” — when designing any piece of work, ask first what the Agent can do. For 42 million small merchants, that means bespoke financial service once reserved for large enterprises is moving into their chat windows. The question to watch is how far the Agent travels beyond conversation into credit, bills and tax workflows, and where banks draw a repeatable line between efficiency and compliance.