Limited published evidence · Commercial banking
AI commercial loan process: Customers Bank example
Customers Bank reports a pilot of an AI-assisted commercial loan process, built with OpenAI, in which selected loans closed in seven days, followed by an expansion announced in October 2026. Which planned tasks are deployed is not established, and sample size, error rates and approval controls are not disclosed.
Expand an AI-assisted commercial loan process
Customers Bank
Reported selected-loan pilot followed by expansion; the 22 October 2026 update does not establish which planned tasks are deployed.
A published third-party example, not a Nuvepro customer case or evidence of a partnership.
The workflow at a glance
Commercial-loan pilot and later expansion
Illustrative summary of the published account
Reported work components. Their arrangement does not imply a fixed execution order or simultaneous work.
- Bank account
Selected commercial-loan pilot
Customers Bank reports selected pilot loans and later expanded functionality and use.
- Planned work
Document and memo preparation
An earlier announcement proposed collection, credit/legal-document preparation and collateral monitoring.
- People
Bankers and deal structure
The announcements place bankers’ focus on relationships and deal structure.
What the source reports
Customers Bank reported selected commercial loans closing in seven days during a pilot, compared with its stated typical 30–60 days. Its October update reports expanded functionality and use of the loan process. An April announcement proposed document collection, credit memo and legal-document preparation, and collateral monitoring, with bankers focused on relationships and deal structure. Those planned tasks are not a verified list of shipped capabilities.
Reported technology: Customers Bank with OpenAI; task-level deployment details incomplete. Scope and outcomes are source-reported.
What the enterprise environment would need
The following is a proposed project design for an enterprise doing similar work. Published details are marked; the remaining systems, data and scenarios would be agreed for your project.
Proposed practice environment
Prepare a traceable loan-file draft
Illustrative project design to scope with your enterprise.
- 1 · Set up the environment
Fictional applications and bank rules
Include financial statements, collateral records, required documents and contradictory evidence.
- 2 · Preparation and agent contributions
Agent assembles evidence and gaps
Draft the file with source-linked amounts and unresolved issues.
- 3 · Independent checks
Qualified reviewers reconcile records
Check every material statement; missing documents or policy exceptions must prevent approval status.
- 4 · People decide acceptance
Banker and legal reviewers decide
Retain credit judgments and legal-document approval with authorized people.
A reviewable evidence package
A draft loan file showing sources, exceptions and outstanding human decisions.
If a check fails → withhold acceptance.
Collateral values conflict across records. Preserve both values and request review rather than choosing a favorable figure.
Failure or owner rejection returns to review
- Hold
- → 2 · Revise work
- → 3 · Rerun independent checks
- → 4 · Owner review
Operating boundary. No automated lending decision or claim that planned bank functions are already deployed.
- Loan records
- Proposed: fictional applications, financial statements, collateral records and bank-supplied document requirements. Include missing and contradictory evidence.
- Preparation and approval
- Proposed: AI assembles an evidence-linked draft and flags gaps. Authorized bankers review credit judgments; legal reviewers approve documents. No automated lending decision is implied.
- Exceptions and handoffs
- Proposed: route missing records, conflicting amounts and policy exceptions to named reviewers. Record the decision and supporting evidence before advancing the file.
A proposed simulation
An illustrative practice design derived from this work, not an executed assessment.
Proposed: prepare a traceable loan-file draft from fictional records and identify what a banker still needs to decide.
Review gate: A qualified reviewer can trace the draft to source records, resolve exceptions and identify the outstanding credit and legal decisions.
Challenge the result: Two supplied records disagree about collateral value. The draft must retain the conflict and request review rather than choose a favorable value.
What needs checking
- Proposed: reconcile every amount and material statement in a draft memo with its supplied record; flag unsupported conclusions.
- Proposed: verify that missing documents and policy exceptions prevent a file from being represented as approved.
- Proposed: measure preparation, review and rework separately on agreed cases; do not reuse the reported pilot duration as an enterprise target.
Review environment fidelity, human capability and agent performance separately. The enterprise’s accountable owners decide acceptance of the shared workflow and any operating use.
Questions about this example
- What does Customers Bank use AI for in commercial lending?
- The bank reports an AI-assisted commercial loan process, with a selected-loan pilot followed by expansion. An April announcement proposed document collection, credit memo and legal-document preparation, and collateral monitoring, but those planned tasks are not a verified list of shipped capabilities.
- Where do people stay involved in Customers Bank's loan process?
- The April announcement described bankers focusing on relationships and deal structure. Approval controls are not disclosed. As a proposed design, not the bank's, authorized bankers review credit judgments, legal reviewers approve documents, and no automated lending decision is implied.
- How strong is the evidence for Customers Bank's loan results?
- Evidence is early-stage. The bank reports selected pilot loans closing in seven days against its stated typical 30 to 60 days, but that applies to selected loans only. Sample size, error rates and approval controls are not disclosed, and hours-saved estimates are not attributed to this workflow.
- What would an enterprise need to build something similar?
- This is Nuvepro's proposed design, not a reconstruction of the bank's system. It would use fictional applications, statements and collateral records including contradictions, route gaps and exceptions to named reviewers, and reconcile every amount in a draft memo with its supplied record.
Shape this around your enterprise
Bring a similar workflow and the people responsible for it. Together, define the records and systems to use, the result your team will build or configure, and how its owners will check it. Nuvepro scopes the simulation, environment and preparation. Agree any existing-agent evaluation, implementation assistance and production responsibilities separately.
Sources and limitations
Curation review: . This is the date this source example was reviewed, separate from source publication dates.
- The bank reports the pilot and later expansion; sample size, error rates and approval controls are not disclosed in these releases.
- The seven-day result applies to selected pilot loans, not every borrower or loan product.
- Company-wide estimates of hours and staffing capacity are not attributed to this workflow here.
- The environment and checks on this page are proposed; they do not reconstruct the bank's implemented system.
- Customers Bancorp third-quarter 2026 resultsSource publication: 2026-10-22
- Customers Bancorp second-quarter 2026 resultsSource publication: 2026-07-23
- Customers Bank announces commercial banking collaborationSource publication: 2026-04-27
More banking and insurance examples
- AI auto-insurance claim recommendation: Jerry Liu exampleJerry Liu and LlamaIndex instructional example
- AI fraud risk review: Inscribe exampleInscribe
- AI insurance claims intake: DXC Technology exampleDXC Technology
- AI after-hours insurance intake: Ken Reiners exampleKen Reiners insurance-intake recipe (agency hypothetical)