Account aggregator volumes grew fast in the framework's early years, largely on the back of lending use cases. But adoption curves across several institutions we work with have flattened well short of where transaction-data-driven underwriting should plateau — and the reasons are more operational than architectural.
The consent screen is a drop-off point, not a formality
Every additional bank a customer has to select, every extra consent-artefact screen, costs completion rate. Institutions that treat the AA consent journey as a single well-designed flow — pre-filling known information, minimising the number of screens between intent and completion — see materially better completion than those that bolt the AA SDK on as an afterthought.
Latency changes what the data is useful for
A data-fetch that takes 30 seconds is fine for an overnight underwriting batch and unworkable for a real-time credit decision at checkout. Several BNPL and credit-line integrations we've reviewed were designed assuming near-instant data-fetch, then had to be redesigned around asynchronous decisioning once real-world FIP response times became clear.
The fix isn't waiting for FIPs to get faster — it's designing the credit decision flow to work with the latency that exists today, with a fallback path when data doesn't arrive inside the decisioning window.