AI brings inclusive, professional banking to China's small businesses

AI is making banking more accessible and professional

At a recent financial inclusion forum in Beijing, bank leaders and academics converged on a clear message: AI is moving from pilots to production. For managers in lending, operations, and service, the shift is practical-faster analysis, tighter controls, and simpler workflows for small and micro enterprise clients.

The throughline is disciplined use of large models for reasoning, tool use, and data processing. Success depends on compute, proprietary data, and domain-specific modeling-not hype.

Case study: MYbank's push to act as "CFO for every small business"

MYbank, which serves small and micro enterprises with loans under 2 million yuan, rolled out an AI banking strategy focused on three areas: risk in lending, customer operations and services, and wealth management. The goal is to give every banker an intelligent assistant and every client a faster, more consistent decision.

AI assistants now support credit approval by automating industry research, due diligence, and loan assessment. Tasks that took experts one to two months-data collection, building analytical frameworks, isolating key risk factors-are compressed into minutes. Video and photo checks improve remote due diligence where branches are absent.

Results are converging: AI-generated credit limits now match expert assessments in roughly 90 percent of cases. The team highlights the need for strong computing resources, clean proprietary data, and specialized models tailored to banking workflows.

Case study: Bank of Jiangsu's practical wins in strategy, risk, and ops

Since early 2023, Bank of Jiangsu has built and deployed large-model capabilities, accelerating after an open-source reasoning model from DeepSeek became available in January. The bank applies AI in business strategy, risk control, and operations.

In strategy, the bank profiles low-risk, high-value small business clients by learning from years of customer data, then extends targeting to adjacent segments. In risk prevention, AI flags potential fraud patterns by learning from past cases. In operations, contract review time dropped from minutes to seconds.

For teams exploring similar paths, the open-source option is worth a look: DeepSeek's reasoning model.

Expert view: Keep people first, stay clear on use cases

Industry experts note a measured stance compared with the earlier Internet Plus wave. Institutions are dissecting which AI features solve concrete issues of efficiency or cost, rather than chasing novelty.

The principle is simple: the bank defines the problem, AI assists with the work. Technology reduces manual tasks, but accountability, judgment, and customer trust remain human-led. Overpromising gets you stuck at the "last mile"-where process change, data quality, and human adoption decide outcomes.

What this means for management and operations

• Pick high-friction workflows with measurable outcomes: credit memo drafting, KYC/AML checks with video verification, contract review.
• Pair domain experts with data scientists to reach parity (and then consistency) between expert and model outputs.
• Stand up model risk controls: bias tests, drift monitoring, human overrides, clear audit trails, and periodic back-testing.
• Use proprietary data where you have advantage; keep a clean data pipeline with lineage and quality checks.
• Instrument everything: approval turnaround, loss rates, fraud hits, customer effort score, and cost per decision.
• Decide your stack early: open-source plus in-house guardrails vs. vetted vendors with banking-grade compliance.

90-day execution plan

• Weeks 1-2: Map the credit and ops journey. Select 2-3 pilot use cases with clear KPIs and a compliance review path.
• Weeks 3-6: Assemble datasets and labels. Compare an open-source reasoning model with a managed service on accuracy, latency, cost, and governance.
• Weeks 7-10: Build prompts, tools, and policies. Run side-by-side tests against expert outputs; capture disagreements and refine.
• Weeks 11-12: Deploy to a limited cohort with human-in-the-loop. Track metrics; decide go/no-go and scaling plan.

Governance, trust, and customer protection

Adopt clear principles for fairness, accountability, and transparency in AI. Industry frameworks such as MAS's FEAT principles offer practical guidance for banks rolling out AI in production.

Reference: MAS FEAT principles. Build explainability into decisions, maintain escalation paths, and keep customers informed when AI supports outcomes.

Tools and upskilling

If you're scoping solutions, a curated list of finance-focused AI tools can shorten vendor evaluation. See: AI tools for finance.

For team enablement by role, explore practical learning paths: courses by job. Focus on credit, fraud, and operations workflows your teams own today.

Bottom line

The signal is strong: AI can lift access and professionalism in banking, especially for small and micro enterprises. The advantage goes to teams that define narrow problems, measure relentlessly, and keep people in charge of decisions.

Start small, ship fast, and let results speak.