How BigTech Turns Data Into Credit—and Leverage
This chapter traces how BigTech platforms use payment and nonfinancial data to make credit decisions, and why predicting repayment is different from enforcing it. The distinction matters because the same platform that can identify a borrower with a thin credit file may also control the account, marketplace, or payment route through which that borrower earns and spends. By following the borrower, the lender, the bank partner, and the payment flow together, we can see when a credit-access benefit becomes a new form of dependence.
From a digital footprint to a borrower profile
A lender has two basic problems. First, it must estimate whether a borrower is likely to repay. Second, it must have a credible way to recover value if the borrower does not. Traditional collateral helps with both. A standardized asset can make a borrower easier to assess, and the lender can seize it after a default. But many households and small businesses have little collateral, or hold assets that are intangible, illiquid, or difficult for a conventional lender to evaluate. A lack of collateral or a thin credit history can therefore exclude a borrower who might otherwise be viable.
BigTech changes the information available at the first stage. A platform may see payment patterns, e-commerce activity, and social or other nonfinancial signals that a bank does not ordinarily collect in one place. Machine-learning systems can use those signals to build a profile of behavior and estimate creditworthiness. The appeal is particularly strong for users who have no conventional credit history. A small merchant’s sales on an e-commerce platform, for example, may give a lender a continuing signal of business activity even when the merchant cannot pledge valuable property.
This is sometimes described as turning “soft” information into something usable for a financial decision. The phrase should not be misunderstood. The information is not automatically true, complete, or fair because an algorithm can process it. Payment data may indicate regularity or cash flow, but it cannot by itself provide a full view of a person’s financial health. Social or commercial behavior may be informative in one setting and misleading in another. A more detailed profile can improve a decision, but it can also make the decision more opaque and give the platform a stronger position over the person being assessed.
The platform’s role also changes the shape of the loan. The IMF identifies three broad ways in which BigTech distributes credit. A BigTech firm or its licensed subsidiary may originate and market the loan. It may share risk with an incumbent bank through a joint arrangement. Or a commercial bank may originate the loan while the BigTech platform markets and administers it. These are not cosmetic differences. They determine who supplies funds, whose balance sheet absorbs losses, who makes or approves the decision, who controls the customer interface, and who receives the data generated by repayment.
Consider the third arrangement. The IMF gives the example of Amazon partnering with ING to offer loans to small and medium-sized merchants in Belgium and Germany. Amazon uses information from its e-commerce platform to identify eligible merchants, then redirects them to the bank’s website to submit a credit application. The bank remains central to the formal application and lending relationship, but the platform has influenced who sees the offer and why that offer is possible. The arrangement can extend a bank’s reach without making the platform and the bank interchangeable.
The same division appears in joint lending. A bank may carry most of the risk while the platform supplies distribution, data, and customer access. The imbalance can be commercially useful: the bank brings a license and funding, while the platform brings a large user base. It can also create an incentive problem. If the platform earns a fee for referring or administering loans, it may benefit from sending more customers through the process. The bank, or another entity, may bear much of the loss if the referrals are poorly screened or if customers are encouraged to borrow beyond what they can afford.
Direct lending has a different constraint. BigTech firms often do not hold banking licenses or take deposits. Without the low-cost funding base associated with deposits, they may rely more heavily on market financing. The IMF therefore notes that BigTech lending can carry higher funding costs and higher rates than bank loans, and may focus on higher-risk borrowers, including people who were excluded or underserved by banks. The use of more data does not remove the price of funds or the risk of the population being served. It can make a loan possible without making it cheap.
The resulting credit relationship has at least three layers. The first is an information layer: what the platform knows and how it turns observations into a decision. The second is a financial layer: who provides the funds and bears the credit loss. The third is a control layer: who can influence where the borrower receives income, how repayment is collected, or whether the borrower remains inside the platform. Much of the debate about BigTech credit focuses on the first layer. The third is what makes platform credit distinct.
Screening a borrower is not the same as making repayment happen
A good prediction can reduce uncertainty about default, but it cannot force a borrower who is able to repay to do so. The Bank for International Settlements frames this as the difference between screening and enforcement. Its FinTech vision imagines behavioral information and machine learning improving screening, while a shared programmable ledger records contracts and automatically settles payments. In an idealized setting, a borrower’s future sales could be routed toward repayment before the borrower has an opportunity to divert the money.
The obstacle is side trading. If the borrower can sell output for cash or through an anonymous alternative, the sale may never appear on the ledger. The borrower can default on the recorded obligation and still participate in the wider economy. The ledger has recorded the promise, but it has not controlled the place where the value is created. This is why a shared record is not self-enforcing. Its effectiveness depends on adoption, identity, the availability of other payment routes, and the incentives of the parties that operate it.
The BIS model’s BigTech solution is to bundle trading and payment. A platform can require or strongly steer users toward its internal ledger and tokens, while restricting the usefulness of cash inside its marketplace. If future sales pass through that ledger, the platform can automatically divert a portion toward repayment. The future revenue then behaves like “digital collateral.” It is not collateral in the traditional sense: the borrower has not pledged a building, vehicle, or security. The platform has instead made a future cash flow more observable and more controllable.
The model does not require every transaction in the economy to occur on the platform. Agents choose their holdings before knowing exactly where they will trade. If the platform controls enough trade, holding cash for a side transaction becomes less attractive because cash cannot be used easily for purchases within the platform. A sufficiently large platform can therefore discourage the private liquidity that would make side trading possible, even without complete market coverage.
This is the causal chain in simplified form:
The branches in the diagram matter. A platform may supply the loan, share the risk with a bank, or merely distribute a bank’s product. The repayment path may still run through a platform account, but the legal and financial consequences differ. Automatic collection can make repayment more reliable, yet it may also remove money from an account needed for other essentials. Exclusion can protect a lender’s future transactions, yet it can harm a borrower if the platform provides a critical service and there is no practical substitute.
The BIS mechanism is useful precisely because it exposes the condition on which enforcement rests: control over the venue of exchange. It should not be read as a claim that every real platform can legally or technically force all relevant sales through a closed system. Christine Parlour’s discussion in the same volume notes that real economies have credit registries, delays between contracts and settlement, durable goods and relationships, and factoring markets in which future sales already support borrowing. Those institutions change the enforcement problem. They may let lenders use future cash flows without granting one platform complete control over the payment environment.
They also change the policy options. If transaction information can be made portable, a borrower may not need to remain inside one platform to preserve a financial history. If factoring contracts are standardized and enforceable, a business may borrow against future invoices through a market rather than through the platform that generated the sale. If consumers have genuine switching options, a threat of exclusion becomes less coercive. These alternatives do not make platform enforcement irrelevant, but they limit the claim that a walled garden is the only route from data to repayment.
Why more data does not guarantee safer or cheaper credit
BigTech credit can widen access because the platform sees signals that traditional lenders lack. It can also increase exposure to errors that are difficult for a borrower to challenge. An algorithm trained on a large database may detect patterns, but the borrower may not know which data mattered or how to correct a mistaken inference. If the platform combines financial and nonfinancial activity, a decision about a loan can become entangled with behavior that the user did not understand as financial information.
The incentive to expand credit can be strong. Payments often generate thin margins, while lending may produce more revenue. A platform can use payments to acquire information, then use the information to offer credit to a population that other lenders have not served. But the commercial opportunity is not the same as a duty to lend responsibly. A platform may be tempted to use its distribution advantage to increase volume, while a partner bank carries most of the credit risk. This is the incentive asymmetry that makes the identity of the risk-bearing institution important.
Higher rates can be part of the same story. A BigTech lender without deposit funding may pay more to finance its balance sheet. It may then charge more and target borrowers whose risk is also higher. The loan can still be valuable if it is the only formal credit available, but “available” and “affordable” are different judgments. A borrower who receives credit for the first time may gain a productive opportunity, or may be exposed to a repayment burden that becomes more severe when income falls. The sources support the possibility of both outcomes, not a universal conclusion.
Platform control can also turn a credit decision into an access decision. If a user defaults, the consequence may not stop at a lower credit score or collection process. The platform may deduct funds automatically, suspend an account, or restrict access to a marketplace or another service. The IMF warns that this can harm consumers where exclusion blocks a basic or critical service and direct substitutes are scarce. The benefit of a credible repayment threat therefore depends on the borrower’s ability to leave and continue participating in ordinary economic life.
The enforcement mechanism also raises a privacy question. A platform may need enough information to recognize repayment capacity or detect a default, but broader sharing can expose users to profiling and reduce their control over personal data. If competing lenders exchange default information, credit markets may function more effectively, yet the same information can make it difficult for a borrower to recover from an error or to move to another provider. Open access to data and strong control over data are not automatically the same policy.
A responsible assessment of BigTech credit therefore asks more than whether the model improves prediction. It asks who chose the borrower, who supplied the funds, who can change the terms, who collects repayment, what happens after default, and whether another provider can offer a meaningful alternative. The answer may vary across the three lending arrangements even when the user sees a similar offer on the screen.
That is why the next boundary to examine is the checkout itself. Buy now, pay later can present borrowing as a smooth payment choice at the moment of purchase. The same questions remain underneath: what information shaped the offer, who bears the risk, how is repayment enforced, and which protections apply when a commerce feature functions as credit?