#236 2015 · MYbank (网商银行) · Banking / microfinance
A bank stopped asking small merchants for paperwork they could never produce and read their QR code trail instead
the problem
Street-stall and small-shop merchants are exactly the borrowers a bank most needs collateral, audited financial statements and credit history from to underwrite a loan safely — and exactly the borrowers who structurally cannot produce any of it, locking an entire tier of otherwise creditworthy small business out of formal lending
background
Traditional bank underwriting requires documentation — audited books, collateral, a credit history — that verifies a borrower's ability to repay without the bank having to watch the business operate directly. Street vendors, small shop owners and micro-merchants running cash-and-QR-code businesses have none of this: no formal accounting, nothing to pledge as collateral, and often no prior borrowing history at all, which is precisely why traditional banks treat them as unbankable regardless of how healthy their actual cash flow is. A human loan officer manually underwriting a small loan this size also costs the bank roughly the same fixed cost as underwriting a much larger loan, at roughly ¥2,000 in labor per loan — making tiny loans structurally unprofitable to process by hand no matter how creditworthy the applicant.
MYbank, an online bank founded within Alibaba's Ant Financial ecosystem, built its '310' lending model around a different evidence source entirely: the transaction data already flowing through a merchant's payment QR code every time a customer paid them. Instead of asking for documents these merchants could never produce, MYbank read their actual cash-flow history directly from the payments infrastructure they already used daily.
what everyone would do
Help merchants produce the paperwork banks require — financial literacy coaching, simplified applications, government-backed guarantees to substitute for missing collateral. It fails because these businesses are structurally cash-based and informal; no amount of coaching turns a street vendor into someone who can produce audited financial statements, and manual underwriting still costs a bank roughly the same fixed labor per loan regardless of size, making even a successfully documented small loan unprofitable to process by hand.
what they saw
MYbank saw that the documentation these merchants couldn't produce was never the actual thing that mattered, it was just the proxy banks had always used to infer creditworthiness. The real signal — genuine, continuous cash-flow history — was already being generated automatically every time a customer paid through the merchant's QR code, sitting unused inside payments infrastructure nobody had thought to underwrite loans against.
the move
The '310' model — 3 minutes to apply, 1 second to approve and disburse, 0 human loan officers in the loop — runs loan decisions through an automated system built on more than 100,000 data indicators, over 100 predictive models and more than 3,000 risk-control strategies, replacing manual document review entirely with algorithmic analysis of transaction-data patterns the bank could already observe.
why it works
Every QR-code payment a merchant receives is already logged by the payments platform, producing a direct, continuous, hard-to-fake record of actual business activity — arguably more reliable than a self-reported financial statement, and available for a population that has no financial statements to report at all. Feeding that transaction stream through automated models lets the bank assess creditworthiness without requiring merchants to produce anything beyond what their normal daily business already generates. Because the entire process runs without a human loan officer, the fixed labor cost that made small loans structurally unprofitable disappears, dropping per-loan operating cost roughly a thousandfold — converting an entire population from unbankable to cheap-to-serve at scale, with default rates comparable to conventional lending because the transaction data turns out to be a genuinely good predictor of repayment, not merely a substitute for one.
the payoff
The automated model cut average per-loan operating cost from roughly ¥2,000 under manual underwriting to approximately ¥2.30, while serving more than 40 million small merchants cumulatively, with an average loan size of ¥36,000, average loan duration under 120 days, and over 80% of borrowers receiving their first-ever formal bank business loan; the bank has maintained a 30-day non-performing loan rate around 1–1.5%, comparable to or better than conventional small-business lending despite serving borrowers traditional banks had excluded entirely.
where it breaks
The model only works for merchants who generate a substantial, observable digital transaction trail — a cash-only business with no QR-code or digital-payment adoption produces no data for the system to read, leaving that population excluded through a different door. It also depends on the lender having direct access to that transaction data, which is why the approach works specifically because MYbank sits inside Alibaba's own payments ecosystem; a bank without a comparable data relationship can't replicate the advantage. And the model's accuracy depends on historical transaction patterns still predicting repayment ability going forward — a sudden shock or shift in a merchant's business that hasn't yet shown up in its transaction history could degrade the algorithm's judgment in ways a human loan officer, drawing on broader context, might catch differently.
what came after
MYbank's '310' model is widely studied in Chinese fintech and inclusive-finance literature as a template for algorithmic microlending built on transactional data rather than traditional collateral and documentation, and the bank has since licensed the underlying risk model to partner financial institutions under its 'Fanxing' (凡星) program specifically to extend the same underwriting approach to small merchants outside its own direct customer base.
references
- [1]界面新闻 (Jiemian) — 网商银行发布'凡星计划'：开放'310'能力，服务3千万小商家Jiemian News, 2018jiemian.com
- [2]中证网 (China Securities Journal) — 网商银行：科技让金融更好服务小微企业China Securities Journal, 2019cs.com.cn