Customers Bank and Palantir partner on AI-assisted commercial deposit onboarding

Customers Bank announced a collaboration with Palantir Technologies on October 15, 2026, to deploy Palantir’s Foundry and Artificial Intelligence Platform for commercial deposit-account onboarding. The companies say the project will organize document collection, entity verification and risk-signal aggregation into a more responsive digital application experience.

Commercial account opening is a data-integration problem disguised as a form. A bank may need incorporation documents, ownership information, tax details, financial statements and transaction expectations, then compare those records with sanctions, fraud and credit signals. Putting the workflow on an operational data platform can help employees see missing information and exceptions sooner, but it also makes data lineage and access controls critical.

Palantir engineers are working alongside Customers Bank’s internal AI and technology teams, according to the announcement. That co-development model suggests the bank is treating the project as an operating workflow rather than a generic chatbot. The useful output would be a clearer queue of verified facts, unresolved questions and required approvals; the system should not turn a probabilistic model into an unreviewed account-opening decision.

The partnership does not disclose financial terms or a specific production launch date. Customers Bank remains responsible for know-your-customer, anti-money-laundering, fair-lending, privacy and recordkeeping obligations. Any AI system used in onboarding will also need monitoring for false positives, inconsistent treatment of entities and errors that cause legitimate businesses to be delayed or rejected.

The broader significance is that banks are applying enterprise AI to the back office before putting it directly in front of customers. If the project reduces manual document handling while preserving auditability and human accountability, it could improve both speed and control. If the data is incomplete or the model’s reasoning is opaque, the same automation could simply move friction into a harder-to-diagnose system.