Here's How Affirm's AI Upgrade Could Strengthen Credit Underwriting

Affirm launched a transformer-based real-time underwriting model that increased completed purchases by 3.4% versus the prior-system control group by approving additional applicants, including consumers with thin or no FICO credit histories. Early loans from the expanded approval pool performed better than a comparable expansion under the old model, suggesting potential upside to originations, merchant activity and revenue without a commensurate deterioration in credit quality. The benefit remains contingent on scaling the model while sustaining loan performance.

Analysis

The relevant earnings lever is not simply higher checkout conversion: better ranking of marginal applicants can improve AFRM’s unit economics twice—incremental merchant network revenue and lower provision/funding drag per dollar originated. The market will require evidence that this survives broader deployment, because underwriting models often look strongest before exposure migrates from thin-file but prime-adjacent borrowers toward truly marginal cohorts. The next two earnings reports should reveal whether GMV growth can accelerate without an offsetting rise in delinquency, loss reserves, or funding costs.

AFRM’s data advantage is most defensible in its own merchant and repayment loop, whereas UPST’s bank-partner model is more exposed to lender appetite and model-validation constraints. PYPL has distribution scale but may face a less favorable trade-off between approval expansion and loss tolerance given its broader payments franchise; AFRM could gain merchant share if it can demonstrate superior incremental conversion with stable merchant economics. Conversely, the feature is unlikely to create a durable valuation rerating until management quantifies cohort-level loss performance and contribution-margin uplift.

Near-term sentiment may over-credit a company-authored test result after a substantial share-price run. The contrarian read is that greater approval rates can mask adverse selection for 6-12 months, particularly if consumer stress rises; a worsening charge-off vintage would compress both earnings estimates and the growth multiple. Falsify the constructive view if 30+/60+ day delinquency or provision expense rises faster than originations over the next two quarterly prints, or if merchant-funded take rates decline as conversion gains are shared with partners.

Key Decisions for Investors

• Maintain a watch-list long bias on AFRM, but do not chase the announcement. Initiate only after the next earnings release if management reports accelerating GMV/originations alongside stable or improving credit-loss and transaction-cost ratios; target a 10-15% upside over 1-3 months, with a 7-8% stop or exit on deteriorating delinquency commentary.
• Express relative confidence through long AFRM / short UPST in equal dollar amounts over 3-6 months, contingent on AFRM disclosing cohort-level credit improvement. AFRM’s closed-loop commerce data should monetize more directly, while UPST remains more sensitive to partner-bank credit-box tightening; cover if UPST loan volumes reaccelerate without higher funding costs or AFRM’s loss metrics worsen.
• Use AFRM earnings as the catalyst rather than buying implied AI optionality now: consider a defined-risk call spread only if implied volatility is below its prior four-quarter pre-earnings range. The key upside trigger is quantified contribution-margin expansion; absent that disclosure, treat this as operational validation rather than an earnings-estimate revision.
• Monitor PYPL’s Pay Later merchant adoption and disclosed credit losses over the next two quarters. A material PYPL conversion initiative or aggressive merchant pricing would weaken the AFRM share-gain thesis; until then, PYPL is a watch item rather than a direct short.