Verified intelligence · Expert-reviewed from public records, platform data, primary sources, and private research beyond AI search.

AGENTSFinancial servicesUSA

AI agents for financial-institution service and operations

Financial-institution service agents should be approved by obligation and authority: identity, payments, complaints, disputes, KYC, fraud, credit, insurance and system-of-record actions each need audit proof.

The colophon

Landscape reviewed: 2026-06-01
Latest revision: 2026-06-17
Kinetics: Financial-institution service AI is moving from FAQ containment into authenticated routes that can affect access to money, disputes, KYC, fraud, credit, investments, insurance claims, collections and customer trust. Near-term EU transparency, high-risk classification, model-risk and insurance-AI signals can change approval evidence faster than annual contact-centre renewals.
To watch: 23 June 2026: European Commission high-risk AI consultation closes; credit, insurance and profiling-service routes may need different classification evidence ⁷³.
2 August 2026: AI Act transparency obligations apply; EU financial-service agents need disclosure evidence that may split from US service-agent approvals ⁶⁹.
Known unknowns: Independent end-to-end financial-institution agent benchmarks linking authentication state, entitlement and service-account scope, institution type, approved source, disputes, complaints, KYC, fraud, payment boundaries, credit and insurance decision support, supervised communications, human baselines, repeat contact and audit export.

Contents

Executive summary
Financial-institution coverage map
Institution authority register
Provider lanes
Accepted-output economics
Financial workflow routing
Identity, entitlement and data boundary
Regulated communication controls
Systems-of-record integrations
Voice, outbound and collections
Evidence and case studies
Private evals and auditability
Deployment and contract risk
Jurisdiction controls
Implementation plan
Known unknowns
Vendor directory
Source ledger
Glossary

▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒Executive summary

The agent is safe only where the service route is safe. One financial-service channel can hide many regulated jobs, and each job needs its own owner. A balance answer, card dispute, payment correction, KYC prompt, fraud explanation, credit-denial response, insurance claim update and adviser message can all arrive through the same chat, voice or app surface. They do not share one customer duty, source set, system-of-record action or repair owner.

The CFPB's chatbot work supplies the public warning label: automated financial service can leave people with wrong answers, unresolved disputes and no human path ¹¹. Inaccurate service information can also become a consumer-finance compliance problem ².

Treat vendor case metrics as pattern evidence. They show real deployments and useful workflows ¹⁵²², but they rarely expose false resolution, repeat contact, complaint recognition, authentication failure, reviewer minutes or auditability across comparable test sets.

Service routes now span banks, credit unions, BaaS programmes, payment firms, lenders, broker-dealers, advisers, insurers and financial-crime teams. KYC/CDD need FFIEC evidence ⁴⁸⁵⁰; securities and insurance routes need supervision and AIS governance ⁶¹⁷².

Service authority is the first approval decision. A branch-hours answer is not the same route as a dispute, payment, fraud summary, credit-denial explanation, adviser draft, claim interaction, borrower call, or system writeback.

Regulatory anchors belong behind that authority split: US consumer-finance and BSA/AML sources ²⁴⁸, bank and credit-union safeguards ⁴³⁴⁴, EU AI Act and DORA signals ⁴⁶⁶⁹, UK Consumer Duty evidence ⁶⁸, and securities and insurance anchors ⁶¹⁶³.

The buying decision

Decision	Default route	Proof to require
Answer basic service questions.	Approve with approved-content source, versioning and escalation.	Current policy source, answer test set, refusal rule and human fallback.
Use authenticated account context.	Approve only by channel, identity state and data class.	FFIEC-style access assessment ⁸, MFA state, data masking, session logs and role permissions.
Handle disputes, complaints and errors.	Gate tightly; treat as regulated intake, not containment.	Recognition tests for Reg E ⁴, Reg Z ⁵, complaint language, timing triggers, handoff and records.
Route KYC, fraud, AML or sanctions signals.	Assist and preserve facts; do not clear, suppress or discuss financial-crime review.	CIP ⁴⁹ and CDD source ⁵⁰, risk signal, case ID, owner, escalation reason and approved customer-facing language.
Support remittance or mortgage-servicing interactions.	Classify error, payment, hardship and loss-mitigation signals as regulated intake.	Notice content, transfer or loan identifiers, timing rule, servicer/provider owner, borrower/sender communication and handoff record.
Support credit, wealth, brokerage or insurance service.	Draft, retrieve and route; block autonomous decisioning or regulated communications until separately approved.	Approved source, licensed-data entitlement, supervised communication record, adverse-action or claim-control owner, and archive export.
Take payment, transfer, collection or outbound action.	Human gate unless the route is narrow and reversible.	Consent, payment boundary, script approval, action log, rollback and exception queue.
Buy specialist versus platform route.	Prefer banking specialist where core/digital banking controls are hard; prefer platform where workflow spans many systems.	Connector depth, action permissions, audit export, contract remedies and same-stack references.

The executive decision rule is simple: approve narrow routes with named evidence, gate consequential routes with human owners, and block any route that cannot export source, identity, action and outcome proof.

Related report

AI for customer-service workflows

Use the broader CX report for market structure and companion coverage across helpdesk, CCaaS, vertical support, voice infrastructure and QA/control-layer routes.

Related report

AI customer-service pricing and ticket economics

Use the pricing report when outcome pricing, billing-event audit, retained work and cost per accepted resolution drive the decision.

The institution type decides which route can be approved. Financial institutions share channels. They do not share one approval path. A customer may call the same number to ask about a balance, dispute a transfer, update business ownership, report fraud, ask about a credit denial, discuss an investment account, or submit an insurance claim. The agent channel looks the same. The regulated work behind it is different.

The useful coverage map therefore starts with institution type and workflow consequence. Treasury's financial-services AI report names uses across underwriting, fraud, AML/CFT, customer service, risk management, cybersecurity and operations ⁵⁷. The Bank of England and FCA survey adds adoption and third-party-dependency evidence across financial firms, and Treasury's 2026 AI Innovation Series points to AI moving into core workflows ⁵⁹⁷¹.

The approval route should split when the obligation changes. KYC, CDD, beneficial ownership, sanctions and SAR-sensitive fraud review need financial-crime evidence ⁴⁸⁵². Remittance and mortgage-servicing errors need product-specific intake controls ⁶⁴⁶⁵. Broker-dealer, UK conduct and insurance AI governance cannot be collapsed into generic support containment ⁶¹⁶⁸.

Member and customer service remains the surface; the approval map has to reach the regulated work that often sits behind the request.

The boundary matters most in outsourced and embedded finance. A fintech or programme manager may operate the interface, but sponsor banks, processors, issuers, servicers, advisers or insurers still need evidence showing who owns the regulated decision and who can repair the failure ⁹⁵⁶.

Institution lane decides the evidence

Institution lane	Service surface	Hard boundary	Evidence to require
Retail banks and credit unions.	Account service, card status, branch/contact-centre support, disputes and complaints.	Reg E/Z ⁴⁵, complaints, UDAAP ²³, authentication, GLBA/safeguards ⁷, NCUA or banking information-security controls ⁴³⁴⁴.	Approved answers, identity state, error/dispute intake, human queue, access logs and service-provider evidence.
Digital banks, neobanks and BaaS programmes.	Deposit, card, payment, direct-deposit, account-access and programme support.	Sponsor-bank accountability and third-party dependency ⁹⁵⁶, programme-manager role and complaint/dispute handoff.	Responsibility matrix, ledger owner, case owner, third-party monitoring, termination path and customer-facing remediation process.
Card issuers, payments firms and MSBs.	Authorisation status, payment exceptions, chargebacks, remittance help, merchant/cardholder support and fraud alerts.	Money movement, error rights ⁴⁶⁴, sanctions screening, AML/CFT monitoring ⁴⁸⁵³, fraud queues and payment-data handling.	Processor status, payee/merchant data, consent, exception queue, screening result, fraud case and reversal or chargeback path.
Consumer lenders, auto finance and mortgage servicers.	Application status, document collection, servicing, payment help, hardship and denial-adjacent questions.	Credit decisioning, adverse-action reasons and fair-lending controls ⁵⁴⁶⁷, mortgage-servicing errors ⁶⁵ and loss-mitigation handoff ⁶⁶.	Approved status language, document checklist, adverse-action owner, servicing notice/error case, loss-mitigation owner, deadline and human approval.
Wealth, brokerage and advisory firms.	Client communications, account service, portfolio Q&A, advisor assist and operational follow-up.	Recommendation, suitability, fair-and-balanced communications, supervision and books-and-records evidence ⁶¹, and customer-information safeguards ⁶².	Human-send rule, principal/supervisory review where required, archive export, source entitlement and Reg S-P safeguard record.
Asset managers and capital-markets firms.	Research assist, client reporting, surveillance support, operational exceptions and data-heavy workflow help.	Licensed-source rights, market-facing communications, trade-execution boundary and recordkeeping ⁶¹⁶².	Source rights, entitlement filter, reviewer approval, no-trade-execution boundary, archive record and surveillance case trail.
Insurers, brokers and MGAs.	Policy service, first notice of loss, claim status, renewal help, coverage questions and fraud intake.	Underwriting, pricing, claim denial/payment, policyholder communications and third-party AI governance ⁶³⁷².	AIS governance record, coverage-source hierarchy, adjuster review, claim note, fairness test and third-party AI inventory.
Financial-crime and risk operations.	KYC refresh, identity mismatch, fraud report, sanctions hit, AML alert and case summarization.	SAR-sensitive evaluation, tipping-off risk, sanctions override, false-negative risk and investigator accountability.	Risk signal, source links, case ID, investigator note, escalation reason, customer-facing script and no-clearance control.

Lane evidence identifies the obligation. Autonomy depth decides whether the approved path may answer, assist, stage work or stay blocked.

Autonomy depends on decision authority

Depth	Where it fits	Buyer move
Full route approval.	High-volume, customer-facing workflows with clear source, identity, handoff and record requirements.	Approve by institution lane, not by vendor brand or channel.
Assist-only or staged action.	Credit, fraud, AML, claims, brokerage, advisory and market-facing work where the agent can prepare evidence but not decide.	Require human-send, human-clearance or human-approval gates with exportable records.
Blocked until separately governed.	Credit decisioning, adverse-action generation, claim denial/payment, trade execution, AML/fraud clearance and unsupervised investment recommendations.	Create a separate regulated route with legal, compliance, model-risk and audit sign-off before any autonomous action.
Companion-report territory.	Model selection, financial-file parsing, capital-markets research, finance operations and broad CX platform economics.	Use this frame for service authority; use companion reports for model, file, market-data and cost details.

Approve the action, not the assistant. An institution-grade agent needs a route register that says what the agent may know, say, change and escalate. The register is the signature proof because it turns a vendor demo into an approval artefact. It separates information, authentication, regulated intake, money movement, KYC, fraud, credit, investments, insurance, collections and outbound voice before the shortlist starts.

Most support automation dashboards reward containment. Financial service needs a different first page: which service route is approved, what authority it carries, what evidence it must retain, and when it must stop.

The same vendor may be approved for public-product questions, conditional for authenticated balance support, assist-only for KYC refresh, and blocked for dispute closure, SAR-sensitive clearance, claim denial, credit decisioning or investment recommendations. The route, not the logo, carries the risk.

A service-authority register should be owned jointly by the customer channel, product owner, compliance, financial-crime, risk, security, legal, vendor management and the system-of-record owner. If no owner can sign one row, that row is not production-ready.

Authority decides what can launch

Route	Examples	Default gate	Record to keep
Public answer	Hours, locations, product FAQs, fee schedule explanations.	Approved-answer content and versioning.	Source article, answer version, channel, transcript and escalation flag.
Authenticated read	Balance, transaction status, direct-deposit status, card state.	Identity state, session scope and data-class policy.	Authentication event, data source, fields shown, masking and user session.
Guided service action	Password reset, card lock, branch appointment, document upload.	User confirmation and narrow tool allowlist.	Intent, confirmation, tool call, result, reversal path and handoff.
Dispute, complaint or error intake	Unauthorised EFT, credit-card billing error, fee complaint, fraud claim.	Regulated-intake recognition and human queue.	Customer words, product, date, amount, regulatory trigger, deadline and owner.
KYC, CDD or beneficial-ownership update	New business owner, changed address, expected-activity change, document gap, identity mismatch.	Financial-crime/KYC owner review and event-driven risk-profile handling.	Customer information, document source, risk signal, ownership/control data, reviewer and update rationale.
Fraud, AML or sanctions signal	Account takeover, suspicious transaction, sanctions hit, unusual activity, payment scam.	Assist-only investigation route; no autonomous clearance, suppression or SAR-sensitive discussion.	Risk signal, source links, case ID, investigator queue, approved language and escalation reason.
Credit or lending support	Application status, document collection, hardship, credit-line question, adverse-action-adjacent question.	Approved credit-policy language and human owner for eligibility, denial and exception decisions.	Product, application/loan ID, document checklist, source reason, adverse-action owner and handoff.
Mortgage servicing or loss mitigation	Payment application question, escrow issue, servicing error notice, hardship request, loss-mitigation status.	Servicer owner review, error-intake recognition and approved borrower communication.	Borrower words, loan ID, notice/error category, timing trigger, loss-mitigation owner and communication log.
Wealth, brokerage or advisory service	Advisor assist, portfolio Q&A, account service, client message draft, transfer instruction.	Human-send, supervision, source entitlement and books-and-records archive.	Source data, adviser/reviewer, communication version, archive ID, account authority and restriction.
Insurance service or claims intake	Policy question, first notice of loss, claim status, coverage question, renewal support.	Approved coverage source, adjuster/underwriter review and no autonomous denial or payment.	Policy/claim ID, source provision, customer statement, adjuster owner, decision boundary and audit note.
Payment or transfer	Loan payment, card payment, ACH transfer, remittance transfer, payment plan.	Explicit customer confirmation and high-risk human gate.	Payee, amount, account, consent, confirmation, processor status and exception.
Collections and hardship	Payment reminder, hardship conversation, promise to pay, cease request.	Approved script, channel limits and collector review.	Consent basis, script version, contact time, consumer request, promise and escalation.
Outbound AI voice	Payment reminder, fraud alert, service callback, appointment reminder.	TCPA/consent, disclosure and suppression-list approval.	Consent source, called number, disclosure script, recording notice, opt-out and outcome.

Route authority says what can launch. Layered authority names the owner, permission and system evidence that keep the approved path from creeping after launch.

Layered authority blocks route creep

Authority layer	System or owner	Agent may do	Gate or evidence	Stop condition
Customer or member identity and authentication	IAM, digital banking and contact-centre authentication.	Match the interaction to the authenticated customer or member, channel, device and caller state.	MFA/session state, voice or digital step-up, device/caller risk and failed-authentication log.	Identity is partial, failed, stale or inconsistent with the requested action.
Entitlements, RBAC and service account	IAM, cybersecurity, operations role owners and API owners.	Use only the customer authority, employee role and AI service-account scope assigned to the approved route.	RBAC matrix, tool allowlist, maker/checker review, privileged-action exception and revocation test.	Shared super-user access, missing revocation or a reviewer approving their own route change.
Core banking, ledger or policy system	Core banking, payments, servicing, claims or IT platform owner.	Read allowed fields or stage reversible changes under the customer's identity and product state.	Field map, source record, tool payload, idempotency check, approval and rollback path.	Ledger-impacting writeback, fee reversal or account-status change lacks approved route authority.
CRM and case	CRM, complaint operations and customer-service case owner.	Create or update the approved case type, assign the owner and attach the transcript.	Case payload, routing rule, queue, deadline, owner and closure restriction.	The agent closes, denies or reclassifies a complaint, dispute, error or AI incident.
Contact centre and voice	Contact-centre operations, telephony owner and voice compliance.	Triage, authenticate, transfer, summarize and use approved scripts for narrow voice routes.	Disclosure, consent, recording notice, suppression list, transfer QA and call outcome.	Disclosure, consent, authentication or transfer proof is missing.
Payments and cards	Payments, cards, processor and digital-banking payments owner.	Explain status, lock or stage a card/payment action where the approved route allows it.	Payee, amount, account, card state, processor status, customer confirmation and exception queue.	Money moves, payee changes or card/account state changes without route approval.
Disputes, remittances and servicing errors	Dispute operations, complaint owner and product compliance.	Recognise intake, preserve the customer's words, classify product route and start handoff.	Product, amount, date, deadline, communication history, owner and customer-facing template.	Autonomous denial, closure, discouragement, remedy selection or deadline handling.
Collections	Collections operations, servicing compliance and contact-strategy owner.	Record requests, use approved language, route hardship and schedule human review.	Script version, contact time, consent basis, restriction status, promise and supervisor review.	Cease request is missed, contact repeats after restriction or negotiation runs without review.
Fraud, AML and KYC routing	Fraud operations, financial-crime/KYC owner and security.	Route suspicious identity, transaction or account-access signals to the owner and preserve context.	Risk signal, transcript, case ID, queue owner and approved customer-facing language.	The agent clears, suppresses or discusses financial-crime review instead of routing it.
Lending, mortgage servicing and deposit servicing	Loan servicing, mortgage servicing, deposit product owner and customer operations.	Explain approved status, request documents and stage servicing tasks.	Product policy, account/loan state, approved language, case owner and human handoff.	Autonomous credit decision, eligibility promise, fee refusal, loss-mitigation outcome or hardship outcome.
Audit, model risk and compliance	Compliance, model risk, internal audit and operational risk.	Produce route evidence, eval results, exception queues and incident traces.	Route approval, test result, model/policy version, reviewer, override and audit export.	No trace can reconstruct source, identity state, action, reviewer and final outcome.
Vendor contract authority	Legal, procurement, vendor risk and ICT third-party owner.	Use vendor functions only inside approved product, data, region, subprocessor and support scope.	Order form, DPA, security artefacts, subprocessor path, audit right and exit evidence.	Workflow, model route, subprocessor or audit-export need sits outside contract authority.

Once the authorities and records are named, the buyer can make the harder decision: where the agent may answer, where it may prepare work and where autonomy should stop.

Autonomy limits by workload

Workload	Answer	Read	Stage action	Execute	Rule
Routine service.	Allow approved content.	Allow masked data by identity state.	Gate with customer confirmation.	Limit to reversible actions.	Approve only where source and handoff are controlled.
Disputes and complaints.	Allow rights explanation from approved source.	Gate product or account pull.	Gate case opening and deadlines.	Block autonomous denial or closure.	Recognition and routing matter more than containment.
Payments and transfers.	Allow option explanation.	Gate account and status verification.	Gate prepared action.	Block unapproved movement.	Money movement needs confirmation, evidence and rollback.
Collections and outbound.	Gate to approved script.	Gate account and contact restrictions.	Gate promise or plan proposal.	Block unsupervised pressure.	Channel, timing, consent and tone are part of compliance.

Provider lanes come before economics because the denominator changes with ownership. A banking specialist, agent platform, voice layer, fraud case layer or archive layer leaves a different mix of accepted output, retained human work and audit export with the institution.

Terms used in this report

Term	Plain meaning
Route	The approved path from work to accepted output: source set, data boundary, model or vendor channel, tools, reviewer, evidence record and reapproval trigger.
API	Application programming interface: a governed software connection used by an app or workflow to call a model or service.
LLM	Large language model: a model trained to read and generate text-like tokens.
Token	A chunk of text used for billing, context windows and rate limits. It is usually smaller than a word, not a page count.

Definitions are reader aids, not separate evidence claims.

MCP PREVIEW INVENTORY

What the locked report contains

Schema endpoints

Sample object: /api/v1/reports/rpt_financial_services_member_service_ai_agents_2026_05/sample
Inventory object: /api/v1/reports/rpt_financial_services_member_service_ai_agents_2026_05/inventory
Schema contract: /api/v1/reports/rpt_financial_services_member_service_ai_agents_2026_05/schema

Coverage fields

Category: AGENTS
Sector: Financial services
Region: US/EU/UK
Topics: banking AI agents, financial services member service, credit union AI, consumer finance chatbots, bank contact centre AI, digital banking assistants, voice agents, complaint handling, Regulation E disputes, collections AI, GLBA, bank information security, credit union safeguards, DORA, ICT third-party risk, EBA ICT risk, third-party risk, core banking integrations, approved answers, AI audit logs, financial institution AI, KYC AI, AML AI, fraud investigation AI, broker-dealer generative AI, insurance AI governance, BaaS sponsor bank oversight, adverse action AI, Regulation S-P, remittance transfer AI, mortgage servicing AI, Consumer Duty AI

Schema contract

Contract: colophon.report_document v2026-05-28
Payload schemas: document, summary, sample, inventory, full_report
Required top-level fields: title, subtitle, abstract, category, sector, region, topics, sources, evidence, claims, colophon, body
Integrity rules: evidence.sourceId must reference sources[].id claim.evidenceIds[] must reference evidence[].id evidence.claimIds[] must reference claims[].id section and block evidence references must resolve colophon.kinetics.speed must match the momentum band for the business impact x momentum model

Data-point inventory

vendor_or_actor_ledger: entity names; product categories; source references; confidence notes
buyer_decision_fields: pricing signals; deployment constraints; risk flags; monitoring fields
evidence_index: source class; access date; claim linkage; verification status

Section inventory

1. Executive summaryPreview
decision-ready conclusions; major findings; confidence caveats
2. Financial-institution coverage mapPreview
structured findings; supporting evidence; buyer-use fields
3. Institution authority registerPreview
structured findings; supporting evidence; buyer-use fields
4. Provider lanesLocked
structured findings; supporting evidence; buyer-use fields
5. Accepted-output economicsLocked
structured findings; supporting evidence; buyer-use fields
6. Financial workflow routingLocked
structured findings; supporting evidence; buyer-use fields
7. Identity, entitlement and data boundaryLocked
structured findings; supporting evidence; buyer-use fields
8. Regulated communication controlsLocked
structured findings; supporting evidence; buyer-use fields
9. Systems-of-record integrationsLocked
structured findings; supporting evidence; buyer-use fields
10. Voice, outbound and collectionsLocked
structured findings; supporting evidence; buyer-use fields
11. Evidence and case studiesLocked
claim support; known gaps; verification notes; 77 evidence objects
12. Private evals and auditabilityLocked
structured findings; supporting evidence; buyer-use fields
13. Deployment and contract riskLocked
structured findings; supporting evidence; buyer-use fields
14. Jurisdiction controlsLocked
structured findings; supporting evidence; buyer-use fields
15. Implementation planLocked
structured findings; supporting evidence; buyer-use fields
16. Known unknownsLocked
structured findings; supporting evidence; buyer-use fields
17. Vendor directoryLocked
normalized entity rows; source ledger; evidence-to-claim mapping
18. Source ledgerLocked
normalized entity rows; source ledger; evidence-to-claim mapping
19. GlossaryPreview
structured findings; supporting evidence; buyer-use fields

Relevance signals

Confidence score: 0.87
Source classes: regulator, regulation, standard, vendor_docs, case_study, privacy_terms, vendor_press, api_docs, pricing_page
Known unknowns: Independent end-to-end financial-institution agent benchmarks linking authentication state, entitlement and service-account scope, institution type, approved source, disputes, complaints, KYC, fraud, payment boundaries, credit and insurance decision support, supervised communications, human baselines, repeat contact and audit export.
Watch signals: 23 June 2026: European Commission high-risk AI consultation closes; credit, insurance and profiling-service routes may need different classification evidence [[E77]]. 2 August 2026: AI Act transparency obligations apply; EU financial-service agents need disclosure evidence that may split from US service-agent approvals [[E73]].

Locked continuation

Purchase or subscribe to unlock the full findings, evidence ledger, and buyer-ready recommendations.