A lot of the AI conversation right now is centered on better models. The next frontier release, the next benchmark, the next capability jump. That progress matters. But in finance, past the basics, better models are not the main constraint.
The real constraint is whether the surrounding system is good enough to make the model useful in a repeatable, defensible way. Infrastructure, architecture, and evaluation harnesses matter more than another incremental model upgrade once you are dealing with live investment workflows.
A stronger model gives you a better first answer. It does not give you a process you can trust.
The distinction matters in practice. A team running live credit analysis needs outputs they can rely on across analysts, across deal types, across market conditions. A better model sitting on top of an inconsistent retrieval layer, unclear context structure, or no evaluation rubric produces better individual answers with the same systemic unreliability. The variance problem does not go away with a model upgrade.
The system also needs to be auditable. In credit, a wrong answer that cannot be traced back to its source is as much a process problem as it is an accuracy problem. Model quality is one input. The surrounding architecture determines whether that quality is reproducible and reviewable.
One of my clearest takeaways from building EigenStrategy is that AI ROI in finance comes from building the right system around the model, so that domain judgment, data, workflow, and reliability reinforce each other. That is not a model problem. It is a systems problem.
The teams that get this right are not waiting for a better model. They are building the infrastructure that will make any sufficiently capable model useful. That work is transferable. A stronger model, when it arrives, slots into a system that is ready for it.