Why the Same Loan Helps One Borrower and Pressures Another
The CFPB evaluation of a standalone credit-builder loan asks a more important question than whether the product can raise a score on average: what happens when people with different starting credit positions receive the same offer? This chapter uses that randomized evaluation to separate the effect of being offered a loan from the effect among people who open one, then explains why existing debt changes the product-fit calculation. The result is a more careful judgment about credit-builder loans than a universal endorsement or rejection.
A randomized offer, not a simple before-and-after story
A before-and-after comparison would observe people who opened a credit-builder loan, note what happened to their scores, and attribute the change to the product. That approach would be difficult to interpret. People who choose a product may already be more motivated, more organized, or better able to make the payments. Their scores might change even without the loan. The CFPB evaluation was designed to reduce that problem.
The study enrolled 1,531 members of a Midwestern credit union who expressed interest in building credit. Researchers randomly placed participants in one of two groups. One group had immediate access to the credit-builder loan. The other was told that they needed to complete an extra step before opening one. This did not force anyone to borrow and did not permanently deny the comparison group access. Participants could still decide whether to follow through.
That design creates two related comparisons. The first asks what happened when a person was given easier access to the product, whether or not they ultimately opened it. This is the offer-level, or intent-to-treat, comparison. It resembles a real market in which an institution makes a product available but consumers choose whether to use it. The second asks what happened among those who actually opened the loan, adjusting for the fact that the easier-access group had a higher take-up rate. That estimate can help describe the product’s effect for takers, but it depends more heavily on assumptions about who took up the offer.
Take-up was not complete. Thirty percent of participants in the immediate-access group opened a credit-builder loan, compared with 12% in the extra-step group. That difference is not a nuisance detail. It explains why a product can have a large effect among borrowers who use it and a much smaller effect when evaluated as an offer to a whole group. In ordinary life, an institution cannot assume that every person offered a product will enroll, complete every payment, or remain current through the term.
The researchers tracked more than the CBL account itself. They examined whether participants had a credit score, how scores changed among those who already had one, whether participants made late payments on the CBL or other loans, and whether savings balances changed. That combination matters because a credit-building product can improve one outcome while straining another. A higher score is not the only relevant result if the borrower becomes more likely to miss a different payment.
The evaluation therefore does not compare a product group with a clean group of people who never wanted credit. It compares randomized access among people who expressed interest, while preserving their choice to use the product. That is a strong design for estimating the consequences of an offer in a real-world setting. It also means the findings must be read in relation to the people who enrolled, the terms of this particular loan, and the difference between access and use.
Random assignment helps balance unobserved differences between the two groups at the point of assignment. It does not make the groups identical in what they do afterward. The extra-step requirement changed the ease of access, and some people crossed from the comparison condition into the product. That crossover is one reason the offer-level estimate is valuable: it preserves the practical fact that real programs have incomplete take-up and imperfect follow-through. It is also why a result among borrowers who opened the account should be labeled as a result among borrowers who opened it, not silently presented as the effect of an invitation sent to everyone.
The outcome design matters for another reason. “Having a credit score” and “changing an existing score” are different endpoints. A person with no score has room to become scoreable; a person with a score may need to improve the score, avoid a decline, or keep other accounts current. Treating those outcomes as one measure would make the product look weaker for people who already had a score and stronger for people whose first success was simply appearing in the scoring system. The study kept those questions separate.
The average result hides two starting positions
The study’s central finding is that existing debt divided the outcomes. Participants without existing debt were more likely to gain a credit score. The offer increased their likelihood of being scoreable by about 24%. Among participants who already had debt, the effect on scoreability was minimal because nearly all of them already entered the study with a score. There was simply less room for the loan to create the first credit signal.
Score changes followed the same pattern. In the intent-to-treat comparison, participants without existing debt saw an 8.9-point increase, while participants with existing debt experienced a small decline of about 3.1 points. The difference between those groups was statistically and economically meaningful in the study’s setting. When the analysis adjusted for the different take-up rates and focused on people who opened the loan, the estimated effect for people without existing debt rose as high as 60 points compared with the existing-debt group.
The 60-point figure is easy to overread. It is not the amount that every person without existing debt should expect, and it is not the same estimate as the randomized offer effect. It is a treatment-on-the-treated estimate that depends on who chose to open the product and how the researchers adjust for that choice. The more conservative offer-level estimates are smaller because they include people who were offered access but did not take up the loan. Both estimates can be informative, but they answer different questions.
The different results make sense when the loan is viewed as an additional recurring obligation. For a person with no existing debt, a scheduled payment can create a first reported installment history without competing against other loan payments. For a person already managing installment debt, the same payment enters an existing queue of due dates. The product may still report positive activity, but the borrower has to protect that payment without allowing another account to become late. The issue is not that existing borrowers lack motivation or that a CBL can never help them. It is that the product changes the payment system they are already trying to manage.
Late-payment results make that risk concrete. Thirty-nine percent of participants who opened a CBL made at least one late payment on it. The evaluation also found an increase in late payments on non-CBL loans, particularly among participants who entered with existing installment obligations. This is an important transmission path: the harm need not come only from failing to pay the new product. Adding a new due date can make it harder to keep an older account current.
The savings findings are more uncertain. The study found an average increase in savings balances, but the authors treated that result as less conclusive than the credit findings. The change may partly reflect money moving between accounts rather than a durable increase in net savings. This qualification matters because the product is often described as building credit and savings at the same time. The credit outcome has a clearer connection to the reported repayment history; the savings outcome depends on what borrowers do with the funds released to them.
The study thus rejects two tempting summaries. “Credit-builder loans do nothing” is wrong for the participants without existing debt, for whom the evaluation found meaningful improvement in scoreability and scores. “Credit-builder loans work” is also too broad, because the same evaluation found late payments, different results for existing borrowers, and uncertain savings effects. The relevant conclusion is conditional: the borrower’s starting obligations alter both the room for improvement and the risk of adding another payment.
There is also a difference between a product outcome and a market outcome. The evaluation can show that an offered loan changed measured credit outcomes for some participants under these conditions. It cannot show that making the product more visible would produce the same result for the entire population identified as credit insecure. People who never express interest, cannot reach the lender, or are not able to take on the payment are not represented by the study’s enrollment pathway. The causal result is strong within its setting, but it is not a national estimate of how many people should use a CBL.
What the study changes—and what it cannot tell us
The evaluation changes the way the market’s reach gap should be interpreted. If providers target everyone who lacks a prime score, they will combine people who need a first credit record with people who already have several active obligations. A single enrollment number will conceal those differences. A more meaningful approach asks whether the product addresses the consumer’s starting problem and whether the payment can be carried alongside existing commitments.
That implication is not a universal eligibility rule. Existing debt is a useful signal of product fit, but the study does not establish a debt threshold that can sort every applicant into “safe” and “unsafe.” A borrower’s income timing, payment history, product terms, access to counseling, and other financial obligations may change the outcome. A program that pairs the loan with coaching may also produce a different experience from the standalone product evaluated here.
The study has other boundaries. It examined one product offered by one Midwestern credit union. Its participants were people who expressed interest in building credit, and the comparison condition involved an extra step rather than a prohibition. The design supports causal conclusions about the evaluated offer in that setting, but it cannot establish the effect of every credit-builder loan, secured card, or wraparound program. Nor does it show that a short-run score change automatically becomes lower borrowing costs, stable housing, transportation access, or wealth accumulation.
Those limits do not weaken the most useful lesson. They show how evidence should guide product judgment. Start with the person’s credit position. Ask whether the product creates a missing opportunity or adds another obligation. Separate the effect of an offer from the effect among people who take it up. Track late payments on other accounts, not only performance on the new product. Treat savings and long-term progress as outcomes that need their own evidence.
Once those questions are in view, the market’s problem becomes more specific. A suitable product may exist, but the consumer may not encounter it. A consumer may encounter it, but the payment schedule may not fit. A payment may fit in one month and fail when income or expenses change. Product use is therefore a chain of conditions, and the next judgment is to connect the product-fit evidence to the path by which people actually reach a product in the first place.