The model and its regulatory perimeter
is a business arrangement, not a single charter or a new exemption. A bank may originate loans, issue cards, hold deposits or provide payment access while a fintech supplies distribution, software, servicing or a customer interface. Each activity needs its own legal and economic map. A deposit program, a loan-origination program and a card network sponsorship should not be treated as interchangeable.
The agencies’ July 2024 statement on third-party deposit arrangements explains that outsourcing does not diminish the bank’s responsibility for applicable law. It identifies fragmented operations, missing records, compliance execution, rapid growth and concentrated funding as potential concerns. The statement reiterates existing guidance rather than creating a new licensing regime. [1]
Current guidance and the September 2026 proposal
The agencies announced proposed replacement third-party risk guidance on September 11, 2026; the Federal Register published it September 15, with comments due November 16. The release says existing guidance would be rescinded and replaced when the new guidance is finalized. That future step should not be treated as completed. The proposal emphasizes tailoring and is nonbinding , not a new sponsor-banking license or an exemption from consumer-protection requirements. [4][5]
For program planning, retain two documents: the controls supported by currently applicable law and existing guidance, and a separate assessment of proposed changes. A proposal can justify reviewing an inefficient diligence process; it does not justify abandoning ledger access, customer servicing or escalation. The recommendations below are operating judgments, not a claim that every listed metric is a regulatory mandate.
Accountability through the full relationship
The 2023 interagency guidance addresses the life cycle of third-party relationships and calls for practices suited to the institution and arrangement. It expressly says it does not impose new requirements. The useful principle is proportionality: understand what the partner actually does and scale diligence and oversight to the risk. [2]
Recommended division of responsibilities starts before launch. The bank should identify who approves underwriting changes, owns customer notices, investigates disputes, reconciles balances and can suspend activity. Contractual responsibility, operational capability and access to evidence must align. A contract that assigns the bank final approval is weak protection if the production system permits unlogged partner overrides.
For lending, review the economics and legal terms of origination, sale, retained exposure, servicing, repurchase obligations and any guarantee. A partner promise to absorb losses is itself counterparty exposure. It does not eliminate the bank’s need to assess the program or the enforceability and collectability of that promise.
The ledger is a customer-protection control
The deposit-arrangements statement warns that inadequate access to records can impair a bank’s ability to identify its obligations and delay customer access. It also distinguishes deposit insurance against bank failure from loss or disruption caused by a nonbank’s failure. Pass-through coverage depends on its requirements being met. [1]
Recommended test: choose a customer and reconstruct the opening balance, every transaction, pending items, fees and closing balance using records the bank can retrieve independently. Reconcile the aggregate customer ledger to the relevant bank accounts and settlement records. An omnibus balance that reconciles in total does not establish that every customer balance is correct.
Require a defined process for exceptions, including their age, owner and disposition. Differentiate timing items from unexplained differences. The strongest evidence is repeated successful reconstruction and correction; the weakest is an assurance that the middleware provider is handling everything.
Map the movement of funds, data and decisions
A useful program map follows one customer transaction through authorization, settlement, posting, servicing and dispute resolution. Identify the legal entity and system responsible at each step. Record whether money is at the bank, moving through a payment rail, or still held by a nonbank. A customer-facing balance can include pending items that are not settled deposits. Treat that distinction as an operational fact to explain, not as a reason to make a broader insurance promise.
The table is an illustrative control design. Assign a named bank owner, an evidence source and a backup operator to each row. A contractual right to inspect data is weaker than a tested ability to retrieve it promptly in a usable format. Bank oversight should reach material subcontractors even when the commercial contract runs through the program manager.
Scroll horizontally to see all columns.
| Control point | Evidence to obtain | Failure response |
|---|---|---|
| Customer and account mapping | Stable customer-to-account identifiers; ownership and authorization records | Stop ambiguous onboarding and resolve identity or mapping exceptions |
| Daily reconciliation | Customer ledger, bank account and settlement totals; transaction-level exceptions | Assign breaks by cause, dollar amount and age; investigate unexplained differences |
| Credit-policy changes | Approved policy version, production deployment log and override record | Suspend unapproved settings and quantify affected accounts |
| Disputes and complaints | Receipt time, legal deadline, case owner and resolution evidence | Escalate before deadlines; preserve a direct customer contact route |
| Partner exit | Readable exports, substitute servicing rights and a rehearsed transfer | Limit new exposure when continuity cannot be demonstrated |
Worked example: fee margin versus concentration
Illustrative economics: a program generates $5 million of annual fees, costs $2 million to operate and oversee, and requires $1 million of expected fraud, credit and remediation cost. The apparent contribution is $2 million before capital, tax and corporate overhead. If one stressed event creates $3 million of incremental loss, more than a year of contribution disappears.
A separate funding example: a single fintech supplies $200 million, or 40%, of a hypothetical $500 million deposit base. Its customers are numerous, but the bank still has a common distribution and operational dependency. Model a coordinated outflow after an outage or partner migration. Customer count alone does not establish diversified funding.
Do not combine all these numbers into an industry forecast. Their purpose is to reveal which assumptions drive a specific program’s economics: loss allocation, reserve collectability, deposit stability and the cost of independently maintaining controls.
Growth gates and exit readiness
A useful launch sequence is limited-volume operation, observed reconciliation, sampled compliance outcomes and only then wider distribution. Define measurable gates for new products, merchants, geographies and credit-policy changes. A volume milestone should not override a growing queue of unresolved customer errors.
The community-bank third-party guide provides a practical framework for planning, diligence, contracts, monitoring and termination. Use it to organize evidence, not as proof that an arrangement is safe merely because each box is checked. [3]
Recommended exit exercise: assume the partner is unavailable tomorrow. Can the bank identify customers, receive payments, respond to disputes, service loans, communicate accurate balances and transfer records? Identify which subcontractors must cooperate and whether the bank has direct rights to their data. Price the transition and identify the staff who would actually execute it.
What to watch in credit and fraud
Track underwriting exceptions, approval changes, first-payment defaults, losses, dispute timeliness, ledger breaks and customer complaints by partner and product. Review concentration in both assets and funding. Investigate whether a partner’s incentives reward originations while leaving losses or remediation with the bank.
My assessment favors sponsor programs where independent control, loss-adjusted economics and credible exit capability grow with the business. The case weakens when fee revenue depends on continued rapid expansion, bank staff cannot reproduce the ledger, or contract reserves are small relative to plausible exposure. The question is not whether partnerships are inherently good or bad; it is whether the bank can demonstrate control over the risks it accepts.
Two stress tests that a profitable program can still fail
Hypothetical reconciliation test: an omnibus account contains $100 million and the customer subledger also totals $100 million. Customer A is understated by $50,000 and customer B is overstated by $50,000. Aggregate reconciliation passes while both customers’ records are wrong. A stronger test reconciles transaction identifiers and customer balances, tests duplicate and missing postings, and follows corrections through the bank and partner systems. Zero net difference is not the same as zero unresolved errors.
Hypothetical test: the earlier $200 million partner-sourced deposit channel loses 30% of its balances, a $60 million outflow. Assume the bank has $20 million of available cash plus $50 million of unencumbered securities that can raise $45 million after an assumed 10% haircut. Gross resources of $65 million leave only $5 million before other obligations. If the securities are already pledged, the apparent second source disappears. Haircuts, encumbrance and access timing should be separate inputs; this is not a regulatory liquidity-ratio calculation.
Partner indemnities require a comparable stress. Assume a $1 million funded reserve supports a $3 million incident. The remaining $2 million is an unsecured claim unless another enforceable funding source exists. If the incident also causes the partner’s failure, its promise is least useful when needed most. Review reserve control, replenishment triggers, collateral rights and claim priority with counsel; do not net an uncertain recovery against immediate customer obligations.
Management should compare growth requests with these stress results. Useful release conditions include reproducible customer records, timely resolution of aged exceptions, available contingency funding and successful servicing transfer tests. There is no universal acceptable break count or reserve percentage: the board needs limits tied to plausible loss, customer impact and operational capacity. A program with lower fee revenue can have better economics if its controls and exit costs are demonstrably manageable.
Research update triggers
Revisit the analysis when a program changes its legal entities, product set, ledger provider, servicing model or loss allocation. Also monitor authoritative supervisory changes and public enforcement findings. This is an operating-model deep dive; it does not infer a confidential rating or allege a problem at a named .
For a bank considering entry, the next decision should be a scoped pilot supported by a full cost model and a tested customer-service path. For a bank already operating at scale, independent record access and exit testing should precede another growth commitment.
A wind-down plan starts before termination
Recommended planning identifies which activities must continue even if new business stops: customer access, settlement, disputes, record retention and required communications. Ending a commercial agreement does not end every customer obligation. Assign an owner and a funded operating path for each continuing function.
The July 2024 deposit-arrangements statement highlights record-access and fragmentation risks. The following framework turns those concerns into a hypothetical operating exercise; it is not an agency-prescribed timetable. Separate a bank failure from a nonbank partner outage or insolvency, because their legal consequences and available remedies differ. [1]
Scroll horizontally to see all columns.
| Milestone | Demonstration | Unresolved risk if it fails |
|---|---|---|
| Containment | Stop new unsupported activity while preserving necessary servicing | Exposure continues growing after control loss |
| Reconstruction | Produce customer-level positions from bank-accessible evidence | Aggregate cash cannot be reliably allocated |
| Continuity | Process essential payments and disputes through a tested path | Customer harm persists despite a reconciled ledger |
| Transfer or closure | Reconcile migrated records and residual obligations | Lost history or duplicate balances |
| Residual claims | Fund disputes, recoveries and correction work | Commercial exit leaves an unfunded servicing tail |
Separate an indemnity from available cash
Hypothetical: a partner promises to reimburse $2 million of losses but fails while only $300,000 of usable collateral is available. Counting the full promise as immediate protection overstates accessible support by $1.7 million before uncertain recoveries. Legal priority, restrictions and enforcement affect what can actually be collected.
Recommended contracts and operating plans distinguish who initially funds corrections, who ultimately bears loss, and who handles customer communications. A disagreement over indemnification should not be mistaken for evidence that no correction obligation exists. Assess whether reserves, guarantees and insurance are independent or depend on the same distressed counterparty.
Prove reconstruction without the partner’s help
Choose a recent date and disable reliance on the partner’s staff for the exercise. Recover the bank-accessible snapshot, apply subsequent verified transactions and reconcile customer-level positions with actual money movements. Document uncertain ownership, pending transactions and unresolved exceptions rather than forcing a false zero difference.
Set recovery objectives from customer promises, settlement constraints and applicable obligations; this article prescribes no universal recovery time. Test who can authorize access and whether critical credentials, data formats or reconciliation logic remain controlled by the failed provider. A data-export clause offers limited comfort if nobody can interpret the export.
Finally, connect the result to a business decision: remediate the missing capability, limit growth or restrict the affected activity. Repeat after material architecture changes. The point is not to predict that a partner will fail, but to show that the bank can meet its responsibilities when a consequential dependency stops working.
Sources
- Banking agencies, Joint Statement on Banks’ Arrangements with Third Parties to Deliver Bank Deposit Products and Services; July 25, 2024Official sourceBack to text: ↑1↑2↑3↑4
- Federal Reserve SR 23-4, Interagency Guidance on Third-Party Relationships: Risk Management; June 7, 2023Official sourceBack to text: ↑
- Federal Reserve SR 24-2, Third-Party Risk Management: A Guide for Community Banks; May 7, 2024Official sourceBack to text: ↑
- Banking agencies and NCUA, proposed third-party risk management guidance announcement; September 11, 2026, page updated September 15, 2026Official releaseBack to text: ↑
- Federal Register, Proposed Third-Party Risk Management Guidance, 91 FR 58536; September 15, 2026; comments due November 16, 2026Official sourceBack to text: ↑