← Guides

AI for Banking: How Banks Use AI Agents for Loan Files, KYC Refreshes, Disputes and Exam Prep (2026)

AI for banks has moved past the chatbot. This guide covers the banking workflows AI agents now run end to end, the controls examiners expect, what banks pay for AI in November 2026, and a 90-day rollout plan for a community, regional or digital bank.

Flat risograph illustration of AI for banking: a friendly robot hands a checked loan folder to a smiling banker in front of a columned bank building with document trays and a calendar

AI for banking means using AI to run the document-heavy work inside a bank: assembling loan files, refreshing KYC and beneficial ownership records, preparing BSA/AML case files, answering customer and branch questions, working card disputes and building exam and board packs, with bankers approving every decision. SPIRITT gives banks AI agents that do this work across the core, the loan origination system, email and spreadsheets, deployable in SPIRITT cloud, the bank's own cloud, on its own servers or fully air-gapped, with per-action approvals, a full audit trail and budgets with hard stops. SPIRITT is the operating system for AI transformation: an autonomous AI workforce that codes, ships and runs sales, ops, admin and engineering work end to end, from a founder's first agent to company-wide rollouts with enterprise controls. In a bank that means agents that complete whole workflows, and that also write and test the small tools a process is missing, such as a document tracker, a reconciliation script or an internal status page.

How do banks use AI today?

Most banks now use AI in three layers:

  1. Models inside core products. Fraud scoring on card and payment rails, credit decision models, transaction monitoring and anomaly detection. These are usually bought from core, card and fraud vendors and governed under model risk management.
  2. Assistants for staff. Chat assistants such as ChatGPT Business or Microsoft 365 Copilot help employees summarize, draft and research.
  3. AI agents for operations. Agents that finish multi-step work across systems: collect the documents, check them against the checklist, update the system of record, draft the letter and hand a banker one decision. Most back-office hours sit in this layer, and it is where SPIRITT works.

The rest of this guide focuses on the third layer, because that is where a bank moves from "employees use AI" to "work gets done by AI under supervision."

Banking workflows AI agents run in 2026

WorkflowBy hand todayWith an AI agentWho decides
Commercial and small business loan filesAnalysts chase documents by email and rekey financialsRequests missing items, spreads statements, computes ratios, drafts the credit memo summaryLoan officer and credit committee
KYC and beneficial ownership refreshesPeriodic reviews pile up in a queuePulls the customer file, checks public registries, lists changes, drafts outreachKYC analyst
BSA/AML alert and case prepInvestigators assemble evidence manuallyGathers transactions, prior alerts and customer profile into a case file, drafts the narrativeBSA officer files or closes
Card and ACH disputesAgents read statements and type lettersClassifies the dispute, gathers evidence, tracks network deadlines, drafts customer lettersDisputes team
Customer and branch questionsPhone queues and shared inboxesAnswers from approved policy on chat, phone, SMS and email; hands off with a summaryStaff on anything outside policy
Exam, audit and board packsWeeks of collecting evidenceBuilds the request list, collects evidence, tracks owners, drafts the binder indexCompliance and internal audit
Treasury and reconciliationSpreadsheet matching every morningMatches GL to statements, flags breaks, drafts the exception reportFinance

Five banking workflows in detail

1. Loan files that arrive complete

SPIRITT reads each new application, compares it with your document checklist and writes a precise request for what is missing. When tax returns and statements arrive, the agent spreads them into your template, computes debt service coverage, leverage and global cash flow, and drafts the credit memo summary in your format. The loan officer reviews one clean file instead of a folder of attachments. For deal teams on the capital markets side, see AI for investment banking.

2. KYC refreshes on schedule

Periodic reviews are easy to fall behind on. The agent works the refresh queue: it pulls the customer record, checks state business registries and public sources, compares beneficial owners and addresses with what is on file, and drafts the outreach for anything that changed. Analysts approve the update; the agent records every source it checked.

3. BSA/AML case files ready for review

For each alert, SPIRITT assembles the transactions, counterparties, prior alerts and customer profile into one case file, writes a factual timeline and drafts a narrative in your standard structure. The BSA officer decides whether to file or close. The agent tracks filing deadlines so cases never age silently.

4. Disputes worked inside network deadlines

The agent classifies each dispute, gathers the statement lines, merchant details and customer statements, tracks Regulation E and card network timelines, and drafts the provisional credit and resolution letters. Where the dispute system has no API, SPIRITT uses it on its own cloud computer the way an employee would.

5. Exam and board prep without the scramble

Ahead of an exam, the agent turns the request letter into a tracked list, collects evidence from shared drives and systems, chases owners and builds the binder index. For board packs it collects the figures, drafts commentary and checks every number against its source. Community institutions can see the member-service version in AI for credit unions.

Controls a bank needs before AI touches real work

Examiners expect AI to sit inside existing frameworks: model risk management guidance (SR 11-7), the 2023 interagency guidance on third-party risk management, and your own information security program. In practice that means:

  • Deployment you control. SPIRITT runs in SPIRITT cloud, in your own cloud account, on your own servers or fully air-gapped, so customer data can stay inside your perimeter.
  • Isolation. Every workspace runs on its own isolated cloud computer. Keys are injected at the network boundary, so agents never see credentials.
  • Permissions per action. Read, Write and Destructive rules per integration (allow, ask or never), with single-use approvals for anything that changes a record or sends a message.
  • Audit trail. Every action, source and approval is logged and exportable for internal audit and examiners.
  • Identity. SSO through SAML or OIDC, scoped roles and SCIM provisioning.
  • Cost control. Budgets per organization, team, project, agent and task, with alerts, hard stops and itemized receipts for every model, tool and data call.

SPIRITT composes models within each task, so a bank is never tied to a single lab's model, and it can connect the bank's existing ChatGPT, Claude or Grok plans. Forward-deployed engineers and certified implementation partners work with your team on your systems until a workflow runs in production.

What banks pay for AI (checked November 2026)

OptionWhat you getPublished price
SPIRITTAI agents that complete banking workflows across 1,500+ apps and legacy systems, with approvals, audit and budgetsFree to start; Starter $32, Pro $256, Max $1,024 a month; Enterprise custom
ChatGPT BusinessChat assistant workspace for teams$25 per user a month, or $20 billed annually
Microsoft 365 CopilotAssistant inside Office apps$30 per user a month, paid yearly
Microsoft Copilot StudioLow-code agent builder$200 per 25,000 Copilot Credits a month

Core, loan origination and fraud vendors such as nCino, Jack Henry, Fiserv and FIS sell AI features inside their platforms on negotiated contracts. SPIRITT works alongside them: it uses those systems the way your staff do and connects the steps between them.

A 90-day rollout plan for a bank

  1. Weeks 1 to 2: pick one workflow. Choose a high-volume, document-heavy process with a clear checklist, such as small business loan document collection or KYC refreshes.
  2. Weeks 3 to 4: run on test data. Let the agent work fictional files end to end. Review every artifact, tune the checklist and templates.
  3. Weeks 5 to 8: shadow mode. The agent prepares real files with every external action set to "ask." Measure time per file and rework.
  4. Weeks 9 to 12: supervised production. Approve actions in one click, keep destructive actions on "never," and add the second workflow.

Track hours saved per file, days from application to decision and the share of files that arrive complete.

Where to start

Start with the workflow your team complains about most. Mortgage teams can compare notes with AI for mortgage brokers, and finance leaders running the whole function can read AI agents for finance. Security reviewers can begin with SPIRITT security and deployment options.

FAQ

What is AI for banking?

AI for banking is the use of AI models and AI agents in banks to detect fraud, support credit decisions, serve customers and run back-office operations such as loan file assembly, KYC refreshes, dispute handling and regulatory reporting, with bankers supervising every decision.

How is AI used in banks today?

Banks use AI for fraud scoring, transaction monitoring and credit models inside core products, chat assistants for staff, and increasingly AI agents that complete operational workflows end to end, such as collecting loan documents, preparing BSA case files and building exam binders.

What are the best AI agents for banking?

The right choice depends on the job. Core and fraud vendors embed AI in their own platforms; assistants like ChatGPT Business help staff draft and summarize; SPIRITT runs complete operational workflows across all of a bank's systems with approvals, audit trails and on-premises or air-gapped deployment.

Is AI safe for banks to use with customer data?

It can be, when the deployment fits your risk program: data stays in your environment, every action needs the right permission, every step is logged and the vendor is reviewed under third-party risk management. SPIRITT supports your own cloud, your own servers and air-gapped deployment, SSO and per-action approvals.

Will AI replace bank employees?

AI agents take over the repetitive assembly work: chasing documents, rekeying data, drafting letters and building packs. Bankers keep the decisions on credit, compliance and customers, and spend more of their day on them.

How much does AI for a bank cost?

Assistants cost about $20 to $30 per user a month. SPIRITT is free to start, then Starter $32, Pro $256 or Max $1,024 a month, with Enterprise pricing for bank-wide rollouts, private deployment and forward-deployed engineers.

Can AI agents work with legacy core banking systems?

Yes. SPIRITT connects through APIs where they exist and, where a system has no API, operates the application on its own cloud computer the way an employee does, inside the permissions you set.

How we checked

We read the current Google top results for "AI for banking" and "AI for banks" in November 2026, the published pricing pages for ChatGPT Business, Microsoft 365 Copilot and Copilot Studio, and SPIRITT's own product and pricing documentation. Regulatory references are to public guidance (Federal Reserve SR 11-7 and the 2023 interagency third-party risk guidance). Prices are US list prices on November 21, 2026 and change often; confirm on each vendor's site.

Pilot an AI agent on one banking workflow

Buy from builders who use what they sellBuilt usingSPIRITT