Opinions expressed by Digital Journal contributors are their own.
Bad data in finance does not announce itself. It accumulates across processes and systems until a negative event forces it into view. When a location closes with no clear record of which invoices or meters are associated with that site, or a missed invoice triggers lates fees and a utility shutdown, those failures surface as exceptions employees and finance teams were never designed to resolve. A finance team drowning in spreadsheets and static reports is not failing at analysis; they are failing at building a strong foundation.
Martin Resch, President and Chief Executive Officer (CEO) of Cass Information Systems, has spent his career building the infrastructure that fixes that foundation, and his view on where AI belongs in that work is precise. “LLMs can identify and highlight issues,” Resch states. “But they should never be independently making the decision to move money from one place to another.”
Bad data makes every other problem worse
A retailer opening hundreds of stores a year, many in strip malls, acquiring three electric meters per location can end up with three different addresses for the same site, each sitting in a different system with no common key linking them. Cost accounting breaks, general ledger (GL) allocations become unreliable, and environmental, social, and governance (ESG) reporting at the geographic level is impossible. When a store closes, finance cannot identify which invoices to eliminate because the records were never unified to begin with.
Cass uses AI-native tools to reconcile those misalignments at scale, normalizing meter assignments, correcting address records, and feeding clean data back to clients so both parties operate from a single source of truth. The impact flows immediately into operations. Site-level reporting becomes reliable, store onboarding and offboarding accelerate, and the data quality required for regulatory and ESG reporting is no longer a quarterly firefight. Data quality is not a technical housekeeping task. It is the prerequisite for subsequent financial decisions.
AI should execute deterministic logic, not interpret ambiguous instructions
When AI moves money, the accountable party is the human whose credentials initiated the transaction, not the AI. That accountability requires that the decision logic itself be deterministic, not interpreted. Resch draws the distinction with precision: Cass uses AI to support application development and build platforms, but the decision-making rule sets are deterministic conditions, not prompts handed to a language model for interpretation.
The risk of ambiguity in financial decisions is not theoretical. An instruction that can be legitimately read two different ways is dangerous when $90 billion flows through the system annually. Large language models (LLMs) at Cass exist to identify exceptions, escalate them to humans and provide guidance to the underlying root cause. They do not decide whether money moves. That decision is encoded in advance, agreed by all parties, and triggered only when deterministic conditions are met. That is not a limitation on AI capability; it is the correct architectural choice for any organization that takes financial accountability seriously.
Redeploy the team, do not reduce it
The real value AI creates for finance teams is not in eliminating headcount; it is in redirecting skilled people from report production toward fixing the root-cause issues those reports expose. A report that previously required a team of people working for two weeks can now be generated in one to two days. The chief financial officer (CFO) who responds to that by reducing the team is taking advantage of the wrong opportunity. The CFO who redirects those people to address delinquent vendors, improve days outstanding, and resolve the exceptions AI is surfacing is building a finance function that gets materially better over time.
AI should tell finance teams where to focus their attention, not replace the judgment required to act on it. When expected funding fails to arrive, AI can cross-reference signals across platforms to assess whether the situation is a low-risk processing delay or a high-risk indicator of financial distress. That analysis elevates the finance function from report production to strategic decision support, where the real competitive advantage in financial operations is built.