Automate consumer loan servicing and collections
TLDR
Match authorized payments to settlement and ledger records. Release payoff, borrower and collections actions only when the evidence for that decision is confirmed.
Funded loans
Payment feeds
Lender rules
Case records
Board accounts
Prepare cases
Track due dates
Post payments
Route outcomes
Outcomes
Servicing ledger
Case worklists
What the servicing operation must deliver
The Head of Servicing owns a portfolio whose balances and account states reconcile to accepted evidence. The operation starts with a funded loan or an authorized transfer into servicing. It delivers current account records, resolved borrower cases and accepted handoffs through payoff, closure, transfer or permitted recovery. The scope is U.S. direct nonbank unsecured closed-end personal installment lending. Bank-owned, third-party collector, optional product and contract-specific branches apply only where their roles and evidence are established.
In the synthetic example, a settled $240 payment supports a $205 principal and $35 interest split when the allocation rules and processor trace agree. A pending debit leaves payoff on hold. A hardship request on the same account goes to its authorized owner with the current balance, terms and history. Keep the case on hold while the decision is pending.
How the lifecycle connects
- Start with funded-loan boarding. Accept the opening balance, terms, identities, servicing authority and restrictions before using the account in later processes.
- Keep contractual balance and due state current. Verify payment authority independently of the due date before executing a debit.
- Send accepted processor events to payment posting and carry linked returns to payment exceptions. Compare cash, ledger entries and loan controls in servicing reconciliation.
- Treat hardship, permitted collection activity, furnishing disputes and protections as conditional branches. A verified bankruptcy or scope restriction can suspend a downstream action independently of the account balance.
- Confirm actual qualifying funds and prerequisites before closure. Reconcile refunds and residual money independently. Keep accepted records available for borrower service, reporting, custody and required retention.
Independent controls run alongside account work. Access and authority controls govern release. Servicing remediation owns the affected population after a systemic error. Backup readiness, incident recovery and successor servicing require their own evidence before work resumes or transfers.
Process directory
The directory covers 58 distinct processes and 236 anchored tasks. Follow the conditional branches when their triggers occur. Use the first task link to open the detailed source comparison and validation instructions.
- Account setup, requests and payments
- Account closure, residual money and optional products
- Borrower protections, disputes and furnishing
- Collections, relief and recovery
- Loan owners, contractual reporting and servicing transfers
- Servicing controls, records and operational continuity
Account setup, requests and payments
| Process and type | Accepted output and owner | Critical failure and release check |
|---|---|---|
| Accept a loan or book into servicing · Core servicing | Accepted serviceable account/book with reconciled opening state and owned held/rejected exceptions. Owner: Servicing boarding lead. | Compare the funded principal in the transfer manifest with the imported opening ledger in the same currency and cutoff. A one-cent difference keeps the account out of active servicing. Human risk: Unfunded/duplicate input, conflicting ownership or missing documents, opening position or restrictions. AI risk: An agent can replay an action when an earlier response timed out. Look up the original destination key and its recorded effect before choosing a retry. Hold: Accept only usable reconciled accounts. Hold missing authority, balances, documents or restrictions. Return duplicates/unfunded cancellations to originations/transferor. Task checks |
| Maintain borrower, representative and account-access instructions · Core servicing | Accepted effective-dated party, access, contact and delivery-permission record. Owner: Servicing account-maintenance lead. | Use the effective permission revision for the exact representative and account. A replaced authorization cannot unlock access. Human risk: Mismatch, uncertain identity, impersonation or unsupported capability. AI risk: A model can match similar records by name or description while linking the wrong identity. Require the source key and its corroborating record. Hold: Hold unverified instructions. Refuse unsupported authority. Revoke expired/deceased-principal POA as applicable. Keep a repair/callback owner. Task checks |
| Maintain contractual balances, accruals, fees and due state · Core servicing | Auditable contractual balance, allocation basis, schedule and account due/status record. Owner: Loan-account administration lead. | Apply the version of the terms effective for the calculation period. A newer effective date can change accruals and due amounts on older events. Human risk: Unverified rules, stale transaction, unauthorized charge, missing release, conflicting or provisional account effect. AI risk: A model can choose the latest-looking document or timestamp while missing the version effective for this event. Verify the governing revision and time from the source. Hold: Reject unsupported charges or replacement terms. Retain old state/history. Hold disputed calculation until contract/configuration source corrected. Task checks |
| Deliver borrower account statements and event notices · Core servicing | Accurate dated borrower account, event or result communication in the actual source-permitted mode, with applicable provision/dispatch/delivery evidence or a controlled failed or unsatisfied disposition. Owner: Servicing communications lead. | Verify the permitted recipient for this notice and account. A matching email subject or shared address does not prove the recipient is correct. Human risk: Wrong recipient, missing content basis, unverified trigger or stale/revoked consent. AI risk: A model can match similar records by name or description while linking the wrong identity. Require the source key and its corroborating record. Hold: Accepted facts, content, recipient, actual permitted mode and clock must support the communication. Prepared or queued is incomplete. Provision, dispatch, delivery, receipt or an attempted/failed disposition satisfies only the source standard that permits it. An unspecified report/provide mode is not a universal written mandate. Mode and outcome remain evidenced configuration, without invented acknowledgment or recording technology. Task checks |
| Maintain payment authority and scheduled instructions · Core servicing | Versioned valid payment authority and executable, canceled or suppressed future instruction with consumer-copy evidence. Owner: Payments authorization lead. | Match the instruction to the signed authorization effective at the time of submission. A superseded mandate holds the instruction. Human risk: Incomplete authentication, wrong account, invalid signed timing, misclassified card, unapproved postreturn authority. AI risk: A model can match similar records by name or description while linking the wrong identity. Require the source key and its corroborating record. Hold: Block unsupported/revoked/stale authority. Keep pending instructions explicit. Bank stop-payment action remains external. Unchanged obligation remains payable. For CheckConnect, record the attempt separately from effective cancellation, retain pending/failed instruction status, and pass actual withdrawal or discrepancy evidence when relevant. Employer payroll transfer changes are outside OneMain’s payment instruction control. Task checks |
| Execute and receive permitted borrower payments · Core servicing | Identified accepted/rejected/canceled payment attempt or receipt with effective/settlement status and external transaction reference. Owner: Payment operations lead. | The proposed debit date must match the effective payment instruction. A revised contractual due date supplies no debit authority by itself. Human risk: Unidentified/wrong borrower, stale permission, unauthorized method, unavailable facility or blocked property. AI risk: A model can match similar records by name or description while linking the wrong identity. Require the source key and its corroborating record. Hold: Do not treat submission as settled funds. Reject unsupported or blocked tender. Return ambiguous identification for repair and retain trace. Task checks |
| Apply, reverse and adjust loan payment entries · Core servicing | Accepted loan-level principal, interest and fee posting/reversal with revised customer state and audit trail. Owner: Loan posting lead. | Link an entry or reversal to the original processor transaction. Equal amounts can still belong to different payments. Human risk: Unknown loan/funds/status, unauthorized allocation, provisional processing, unsupported credit. AI risk: A model can return plausible amounts or a balanced total after dropping a line. Recompute exact amounts and reconcile individual records as well as the total. Hold: Keep unmatched cash explicitly unapplied with owner. No silent deletion/duplicate credit. Downstream state recomputed after reacceptance. Task checks |
| Resolve payment returns, errors and retry restrictions · Core servicing | Corrected payment disposition with bounded permitted retry, new-authority request, alternative payment or explicit stop and required notices/remedy. Owner: Payment exception lead. | Read the return reason from the original processor record. A mismatch blocks the decision about whether a retry is permitted. Human risk: Wrong or absent original, mismatched timing/code, duplicate complaint or uncertain withdrawal. AI risk: A model can match similar records by name or description while linking the wrong identity. Require the source key and its corroborating record. Hold: Do not retry an unauthorized return or bypass account-level stop by rail switch. Ambiguous codes/timing remain held. Consumer bank produces its Regulation E error decision. Task checks |
| Reconcile servicing cash and loan control totals · Core servicing | Accepted cash-to-channel-to-subledger reconciliation with resolved or aged, owned reconciling items. Owner: Servicing cash reconciliation lead. | Compare settlement, ledger and bank totals at the same cutoff. An extra hour can create a false break or hide an unmatched receipt. Human risk: Late feed, wrong cutoff, scope or unavailable cash count. AI risk: A model can return plausible amounts or a balanced total after dropping a line. Recompute exact amounts and reconcile individual records as well as the total. Hold: Do not lock unexplained differences. Preserve suspense and missing/late results. Escalate stale breaks and recheck after adjustment. Task checks |
| Complete ordinary servicing inquiries and requests · Core servicing | Usable account-specific answer or completed ordinary request with accepted specialist referral when needed. Owner: Borrower service lead. | Match the authenticated request to the exact servicing account before returning records or updating a request. Human risk: Unverified caller, wrong account, inaccessible route or unverified service scope. AI risk: A model can match similar records by name or description while linking the wrong identity. Require the source key and its corroborating record. Hold: Retain case ownership until answer or receiving specialist acknowledgment. Escalate error, dispute, hardship and protected status to their producer. Task checks |
| Decide routine contractual due-date changes · Triggered when applicable | Approved/declined routine date-change disposition and accepted effective schedule instruction. Owner: Servicing schedule-change authority. | Apply the accepted schedule change to the correct effective date. Keep any payment instruction on its separately authorized schedule. Human risk: Unavailable feature, disallowed date, conflicting pending withdrawals or delinquency requiring assistance. AI risk: An agent can interpret a submitted or queued event as a completed effect. Use the destination’s accepted status and evidence for the exact item. Hold: Hold inconsistent dates/pending withdrawals. Decline ineligible request with explanation or assistance referral. Verify actual generated schedule before release. Task checks |
Account closure, residual money and optional products
| Process and type | Accepted output and owner | Critical failure and release check |
|---|---|---|
| Produce an account-specific payoff quote · Core servicing | Dated payoff quote with validity/assumptions, amount and permitted payment instructions. Owner: Payoff servicing lead. | Verify the quote’s good-through date against the approved calculation. A shorter or expired quote needs review before it is used. Human risk: Missing requester authority, uncertain balance or cutoff. AI risk: A model can return plausible amounts or a balanced total after dropping a line. Recompute exact amounts and reconcile individual records as well as the total. Hold: Hold missing facts. Recalculate stale/changed quote and failed assumed payments. Quote does not satisfy or close the obligation. Task checks |
| Confirm satisfaction and close servicing account state · Core servicing | Accepted legally correct satisfied/settled/discharged/closed state with required confirmation or satisfaction instrument and named residual owners. Owner: Servicing closure authority. | Close the account only after the exact qualifying receipt is confirmed. A submitted or pending payment leaves closure on hold. Human risk: Reversed/missing funds, nonfinal legal action, unauthorized actor or inconsistent balance/rights. AI risk: A model can return plausible amounts or a balanced total after dropping a line. Recompute exact amounts and reconcile individual records as well as the total. Hold: Do not equate chargeoff, quote or pending payment to forgiveness. Prevent future improper debit/collection. Retained records/refunds/tax/disputes survive closure. Task checks |
| Determine and complete eligible postfunding loan cancellation · Triggered when applicable | Accepted cancellation/unwind or reasoned ineligible/failed outcome, with financial and linked-product/reinstatement reconciliation. Owner: Servicing cancellation lead. | Use the cancellation window from the applicable contract or program and preserve its clock source. Similar dates can change whether the request qualifies. Human risk: Wrong program/date, unfunded request, or no applicable postfunding right. AI risk: A model can fill a missing fact from a nearby record or a note. Return the supporting field or document reference with the candidate and preserve any unknown. Hold: Hold incomplete/unreturned funds. Where no postfunding right exists route to payoff. Return of purchased goods alone does not cancel the loan. Task checks |
| Return borrower credit balances and remedy payments · Core servicing | Cleared cash/deposit borrower refund or documented source-required return/tracing disposition, with any separate other-law property referral accepted. Owner: Borrower refund lead. | Match the approved recipient to the borrower credit or remedy record before releasing a refund. Human risk: Installment-only excess, missing written request or actual withdrawal, wrong payee, unsupported approval or unverified delinquency outcome. AI risk: A model can read a request or prior approval as authority for the current action. Match the actual permission, its scope and its effective revision. Hold: Do not close on check issuance alone. Reissue/trace returns. Route lawful dormant remainder and preserve liability. Task checks |
| Resolve and remit dormant borrower property · Handoff | Accepted reunited property or state notice/report/remittance and reconciled remaining liability/claim record. Owner: Lender holder unclaimed-property authority. | Confirm the jurisdiction derived from the verified holder and owner-address evidence. A different state changes the reporting path. Human risk: Uncertain ownership, amount, jurisdiction, source history or early handoff. AI risk: A model can return plausible amounts or a balanced total after dropping a line. Recompute exact amounts and reconcile individual records as well as the total. Hold: Hold disputed ownership/amount. Preserve separate notice and remit stages. Supplemental/nil rules and owner claim response are state-specific. Task checks |
| Complete optional-product cancellation and attributable refunds · Triggered when applicable | Accepted optional-product cancellation/refund entitlement or disposition, with evidence of required actual premium, fee and attributable-interest effects and source-required confirmation. Owner: Optional-product cancellation lead. | Identify the exact optional product and applicable cancellation terms before computing an attributable refund. Human risk: Wrong product, unavailable designated channel or lost in-person request. AI risk: A model can fill a missing fact from a nearby record or a note. Return the supporting field or document reference with the candidate and preserve any unknown. Hold: Submission alone incomplete. Track provider rejects/no-refund basis. Coordinate cash-versus-loan-credit route. Do not omit required associated-interest recalculation. Task checks |
| Coordinate protection claims and accept benefit application · Triggered when applicable | Accepted protection-program decision, claim/proceeds or benefit consequence within the actual established authority, with evidence of separately produced payment/account treatment and required explanation. Owner: Protection-claim servicing lead. | Match the provider’s decision and benefit payment to the exact claim before applying a credit to the loan. Human risk: Wrong policy/actor, missing claimant evidence or no actual claim. AI risk: A model can fill a missing fact from a nearby record or a note. Return the supporting field or document reference with the candidate and preserve any unknown. Hold: Keep pending/denied/appeal state and continuing payment duties explicit. Servicing coordinates and accepts actual program outputs. Claim approval, cancellation entitlement and payment may have different authorities. No generic insurer adjudicator or promised benefit is assumed. Task checks |
| Deliver required postpurchase optional-product disclosures · Triggered when applicable | Accepted source-qualified purchase-disclosure case with actual the “Deliver borrower account statements and event notices” process initial letter/email evidence, actual the “Deliver required postpurchase optional-product disclosures” process recorded live-agent call OR the “Deliver required postpurchase optional-product disclosures” process documented attempts plus actual the “Deliver borrower account statements and event notices” process second-copy evidence, and evidence of applicable the “Deliver borrower account statements and event notices” process continuing written-message payload. Owner: Optional-product servicing disclosure lead. | Use the effective optional-product plan associated with the request and triggering event. An older disclosure may belong to a different version. Human risk: Missing purchase/Plan date, wrong covered product or language/deadline content. AI risk: A model can fill a missing fact from a nearby record or a note. Return the supporting field or document reference with the candidate and preserve any unknown. Hold: When the Enhanced Disclosures requirement applies, require the initial letter or email sent within 72 hours of purchase, PLUS EITHER (a) the recorded live-agent/Customer call more than 48 hours after purchase and before the 60-day period expires, OR (b) source-deemed call satisfaction from three documented phone attempts on three different days during that 60-day period AND actual provision of the second copy by letter or email. The fallback substitutes only for the recorded call, not the initial written send. Attempts alone are incomplete. No separate 72-hour second-copy deadline is stated. Apply the actual Plan date and used-product exception. Continuing paragraph 79 payload travels in actual specified written messages, excludes SMS and creates no extra standalone send. Missing actual evidence or clock remains an owned issue. Task checks |
Borrower protections, disputes and furnishing
| Process and type | Accepted output and owner | Critical failure and release check |
|---|---|---|
| Furnish and reconcile accurate credit-account information · Triggered when applicable | Accepted furnisher CRA submission/result report/correction and applicable reporting-item modification, deletion or permanent block, with controlled rejection, dispute flags and source-to-known-published reconciliation. Owner: Furnishing operations lead. | Furnish the date supported by the original delinquency evidence where applicable. A later reporting cycle does not restart it. Human risk: Unknown actual furnisher, conflicting balance/date, unsupported conversion or unresolved relevant event. AI risk: A model can return plausible amounts or a balanced total after dropping a line. Recompute exact amounts and reconcile individual records as well as the total. Hold: Generated file alone is incomplete. Prevent duplicate/re-aged or blocked re-furnishing. Missing source data/rejected fields remain owned. Task checks |
| Investigate credit-reporting disputes and complete corrections · Triggered when applicable | Reasoned reporting-dispute investigation or supported direct-dispute exception decision, with integrated case acceptance of the required consumer response and CRA reporting/correction evidence. Owner: Furnisher dispute investigator. | Investigate the exact version in dispute. A correction to a neighboring version does not resolve the reported record. Human risk: Wrong route, misdirected address, unknown prior information or unsupported account match. AI risk: A model can fill a missing fact from a nearby record or a note. Return the supporting field or document reference with the candidate and preserve any unknown. Hold: Investigate/review all relevant source-route information. Forwarding is incomplete. The “Investigate credit-reporting disputes and complete corrections” process accepts required actual the “Deliver borrower account statements and event notices” process consumer response and the “Furnish and reconcile accurate credit-account information” process CRA report/correction/effect evidence, or retains the supported unresolved/exception case. A direct frivolous/irrelevant determination requires the separate reasons/information-needed notice. That exception is not imported into the CRA-notified route. Task checks |
| Resolve servicing fraud and identity-theft liability · Triggered when applicable | Supported fraud/nonfraud/held liability disposition with effective protections and accepted downstream correction instructions. Owner: Servicing fraud lead. | Keep a fraud finding and any treatment restriction on the exact account supported by the evidence. Human risk: Unverified report, conflicting identity, missed protection or erroneous hold. AI risk: A model can match similar records by name or description while linking the wrong identity. Require the source key and its corroborating record. Hold: No assumed liability or indefinite unowned hold. Suppress prohibited collection/furnishing/sale and tax treatment for debt not incurred. Preserve lawful correction/resume evidence. Task checks |
| Fulfill identity-theft application and transaction-record requests · Triggered when applicable | Free timely authorized application/transaction record production with delivery evidence, or supported lawful denial/nonexistence disposition. Owner: Identity-theft records response lead. | Match records to the authorized request and transaction. Similar identifiers do not justify releasing other records. Human risk: Missing request, ambiguous transaction, or unsupported third-party requester. AI risk: A model can fill a missing fact from a nearby record or a note. Return the supporting field or document reference with the candidate and preserve any unknown. Hold: Track30-day requirement for qualifying request. No charge or substitution of fraud case acknowledgment for records. Retain lawful refusal basis. Task checks |
| Apply servicing privacy notices, choices and rights dispositions · Parallel control | Accepted privacy applicability/content/choice or rights fulfillment, limited-denial or appeal case disposition, with evidence of required separately produced notice, choice and retention effects. Owner: Servicing privacy liaison. | Match the verified choice to the applicable disclosure or sharing scope. A privacy preference does not automatically change servicing-contact permissions. Human risk: Wrong population, new sharing practice, unverified agent or law conflict. AI risk: A model can fill a missing fact from a nearby record or a note. Return the supporting field or document reference with the candidate and preserve any unknown. Hold: Do not promise blanket deletion of legally retained loan records. Hold unverified agents. Changed practice or loss of notice exception triggers review. Task checks |
| Provide effective accessible servicing communication · Parallel control | Effective accessible servicing arrangement or supported exception with usable alternative and follow-through. Owner: Servicing accessibility lead. | Check the accessible format agreed for the specific request. Sending the usual format can leave the request unresolved. Human risk: Missing customer preference, ineffective aid or unresolved coverage. AI risk: A model can choose the latest-looking document or timestamp while missing the version effective for this event. Verify the governing revision and time from the source. Hold: No forced own-interpreter substitute. Ineffective format/callback remains open. Undue-burden/fundamental-alteration decision stays with authorized business owner. Task checks |
| Determine and implement military servicing protections · Triggered when applicable | Supported relief/denial/held determination with forgiven excess interest, actually returned excess paid amounts and revised installments/restrictions. Owner: Military benefits authority. | Preserve the verified service date used by the applicable protection and review the exact obligation and request before a release. Human risk: Missing/contradictory service proof, incorrect account or unverified representative. AI risk: A model can fill a missing fact from a nearby record or a note. Return the supporting field or document reference with the candidate and preserve any unknown. Hold: Do not defer forgiven interest or accelerate principal to offset it. Departure from cap requires applicable court relief. Preserve evidence and court-action handoff. Task checks |
| Validate covered collection debt and resolve disputes · Triggered when applicable | Accepted collection-validation information, verification/original-creditor or supported duplicate determination. Source-compliant initial oral provision where selected. And permitted-resume/continued-stop case disposition based on required actual written-send evidence. Owner: Collection validation authority. | Confirm the creditor for the specific debt and covered role. A servicing or placement record alone does not establish current ownership. Human risk: Unverified debt, wrong creditor, inconsistent itemization or unsupported recipient. AI risk: A model can fill a missing fact from a nearby record or a note. Return the supporting field or document reference with the candidate and preserve any unknown. Hold: Apply the actual collector/state-creditor route and validation-period facts. Complete initial oral provision can satisfy its alternative. Where a written response/notice is required, the “Deliver borrower account statements and event notices” process actually sends it before any source-permitted collection resumption. Missing proof remains held. Do not substitute a reporting dispute, consumer silence or prepared packet for the required result. Task checks |
| Administer bankruptcy restrictions, claims and case outcomes · Triggered when applicable | Protected case-linked account with actual claim/no-file, accepted distributions and court-supported stay/discharge/dismissal disposition. Owner: Bankruptcy servicing lead. | Match the case evidence to the actual docket, borrower and obligation. A restriction may stay in place while identity or case scope is unresolved. Human risk: Wrong debtor/docket, missing notice or uncertain deadline. AI risk: A model can match similar records by name or description while linking the wrong identity. Require the source key and its corroborating record. Hold: Stop prohibited collection/setoff. No general restart after discharge. Case deadline from actual notice/rules. Missing/contested claim held and corrected. Task checks |
| Resolve deceased-borrower, estate and remaining-liability cases · Triggered when applicable | Documented estate/liable-party treatment with accepted claim/benefit/payment/continuing or uncollectible disposition. Owner: Deceased/estate servicing lead. | Match the estate evidence and authority to the applicable case. A relationship to the borrower does not establish representative authority. Human risk: Wrong person, weak evidence or conflicting recent activity. AI risk: A model can return plausible amounts or a balanced total after dropping a line. Recompute exact amounts and reconcile individual records as well as the total. Hold: Do not demand personal payment from nonliable survivors. Hold unauthorized contacts and pending benefit/estate evidence. Preserve court/probate deadlines. Task checks |
| Implement sanctioned or restricted-payment dispositions · Triggered when applicable | Implemented cleared/held/blocked/rejected/authorized-release disposition with accepted holder/reporting feed and records. Owner: Servicing restricted-payment lead. | Apply the actual authorized restriction version to the exact transaction or account. An unverified match cannot decide the funds disposition. Human risk: Unverified identity/holder, mixed block and reject, unsupported release. AI risk: A model can match similar records by name or description while linking the wrong identity. Require the source key and its corroborating record. Hold: Do not release blocked property on ordinary servicing discretion. False-positive correction and release need actual authority. No universal nonbank AML/SAR rule imported. Task checks |
Collections, relief and recovery
| Process and type | Accepted output and owner | Critical failure and release check |
|---|---|---|
| Decide hardship assistance and agreed relief terms · Triggered when applicable | Documented approved/declined/incomplete/withdrawn assistance decision and offered/accepted terms with reasons and next owner. Owner: Lender assistance decision authority. | Match the assistance case to the actual loan. A case-note similarity cannot transfer a hardship hold or decision across accounts. Human risk: Duplicate request, unverifiable identity or missing case facts. AI risk: A model can match similar records by name or description while linking the wrong identity. Require the source key and its corroborating record. Hold: No automatic right to a particular modification inferred. Hold missing facts. New underwriting/refinance goes to origination. Do not label default-related exclusion as blanket notice exemption. Task checks |
| Implement arrangements and accept plan or settlement-condition performance · Triggered when applicable | Accepted versioned arrangement/settlement conditions and performing, completed, cured, re-aged, breached or withdrawn case disposition, with evidence of separately produced contractual, payment and reporting effects and lawful next treatment. Owner: Arrangement servicing lead. | Compare actual qualifying performance with the terms the borrower accepted. A superseded plan cannot establish completion. Human risk: Unaccepted terms, conflicting state or missing condition basis. AI risk: A model can fill a missing fact from a nearby record or a note. Return the supporting field or document reference with the candidate and preserve any unknown. Hold: Do not re-age or mark current before prerequisites. Monitor partial/failed payments, expiry and redefault. Modification is not forgiveness unless explicitly authorized Completion/cure closes the relevant plan/performance case and sends accepted state to the “Determine delinquency and release eligible collection treatment” process. It does not by itself close a continuing loan. Task checks |
| Authorize negotiated debt compromises and accepted settlement terms · Triggered when applicable | Approved/rejected/held debt-compromise decision and accepted settlement agreement with release conditions, expiry and remaining rights. Owner: Settlement decision authority. | The candidate offer amount must match the authorized settlement terms. An amount suggested in a draft is not an approval. Human risk: Debt, ownership or authority conflict, or adverse-litigation case in ordinary queue. AI risk: A model can read a request or prior approval as authority for the current action. Match the actual permission, its scope and its effective revision. Hold: Actual required lender/risk/owner/legal approval and borrower acceptance recorded, or supported refusal/hold. The “Implement arrangements and accept plan or settlement-condition performance” process accepts condition performance, the “Execute and receive permitted borrower payments” process/the “Apply, reverse and adjust loan payment entries” process execute and post funds, the “Confirm satisfaction and close servicing account state” process accepts final debt satisfaction. The “Authorize negotiated debt compromises and accepted settlement terms” process does not repeat these outputs. Task checks |
| Determine delinquency and release eligible collection treatment · Core servicing | Accepted collection treatment/work-queue assignment, closed/removed obsolete work or explicit hold/no-action with current owner and permitted action/channel/time. Owner: Collection strategy operations lead. | Recompute days past due from the accepted due state at the run cutoff. Hold the treatment if the staged value disagrees. Human risk: Stale or conflicting status, duplicate events or missing posting. AI risk: A model can choose the latest-looking document or timestamp while missing the version effective for this event. Verify the governing revision and time from the source. Hold: Close/remove obsolete collection work when accepted state no longer permits it. The “Conduct permitted collection communication and resolve promises” process cancels pending contacts/promises and the “Place, monitor and recall external collection accounts” process performs any actual external-agency recall. Re-evaluate a late payment reversal through current the “Maintain contractual balances, accruals, fees and due state” process/the “Apply, reverse and adjust loan payment entries” process and protected-state inputs before releasing new lawful work. No automatic resurrection of discharged/settled/restricted rights. Unresolved state conflict remains held. Task checks |
| Conduct permitted collection communication and resolve promises · Triggered when applicable | Recorded permitted contact/no-contact outcome and kept/broken/withdrawn promise or accepted specialist referral/next action. Owner: Collections contact lead. | Recheck current consent and restrictions at dispatch. A copied instruction may use the permission that applied before revocation. Human risk: Missing consent/history, unknown recipient, prohibited timing or obsolete treatment. AI risk: A model can read a request or prior approval as authority for the current action. Match the actual permission, its scope and its effective revision. Hold: Block prohibited attempt. Retain failed/undelivered outcome. Suppress and propagate opt-outs. Promise alone does not cure ledger delinquency Close or suspend the contact/promise case and cancel pending obsolete outreach when the “Determine delinquency and release eligible collection treatment” process accepts cure/payment/settlement/closed treatment. Retain history. A later reversal can generate new eligible treatment only after current state and legal restrictions are reaccepted. Task checks |
| Designate charge-off and approve operational exceptions · Core servicing | Dated operational charge-off/deferral designation and accepted Finance/account/recovery instructions preserving legal debt status. Owner: Servicing charge-off authority. | Compare the proposed date to the approved account-level policy outcome and exception decision. An accounting classification does not authorize recovery. Human risk: Stale aging, unconfirmed special status or use of another operator’s threshold. AI risk: A model can choose the latest-looking document or timestamp while missing the version effective for this event. Verify the governing revision and time from the source. Hold: Chargeoff does not itself forgive debt. No unsupported deferral. Retain actual centralized approval where operator requires it. Finance produces accounting journal/provision. Task checks |
| Maintain retained postcharge-off recovery and residual-rights disposition · Triggered when applicable | Accepted retained recovery/ceased/exhausted/transferred route and residual-rights state with reconciled recovery outcome. Owner: Retained recovery portfolio lead. | Use the current ownership and retained-rights record. A prior rights revision can incorrectly reopen collection activity. Human risk: Conflicting title, stale rights or work continuing after satisfaction/discharge. AI risk: A model can choose the latest-looking document or timestamp while missing the version effective for this event. Verify the governing revision and time from the source. Hold: No new recovery work on extinguished debt. Route status is not separate payment/contact execution. Shared monetary and contact producers reused. Task checks |
| Place, monitor and recall external collection accounts · Triggered when applicable | Acknowledged restricted placement/recall and reconciled returned account, status and proceeds without duplicate active owners. Owner: Collection placement lead. | Confirm the authorized placement destination and scope. A prior agency assignment cannot be reused after recall. Human risk: Unresolved role, prohibited debt, missing authority or duplicate active route. AI risk: An agent can replay an action when an earlier response timed out. Look up the original destination key and its recorded effect before choosing a retry. Hold: Unaccepted placement remains lender-owned. Stop/recall flags reach actual agent. Missing data or prohibited debt stays held. No implicit debt ownership transfer. Task checks |
| Authorize legal recovery and administer accepted court outcomes · Triggered when applicable | Accepted authorized/declined/held case and actual filed/court/judgment/enforcement/stop disposition with serviced-account effect. Owner: Lender legal-recovery authority. | Confirm title and authority for the specific debt before legal action is released. Hold where the chain or authority is unresolved. Human risk: Unsupported debt, expired limitation, improper venue or active protection. AI risk: A model can fill a missing fact from a nearby record or a note. Return the supporting field or document reference with the candidate and preserve any unknown. Hold: Prepared file is not filed action. Do not sue/threaten time-barred covered debt. Court/counsel required actions cannot be modeled as packet preparation. Task checks |
Loan owners, contractual reporting and servicing transfers
| Process and type | Accepted output and owner | Critical failure and release check |
|---|---|---|
| Complete authorized debt-transfer execution and servicing residuals · Handoff | Accepted authorized charged-off transfer instruments/chain-of-title and receiver handoff with correct servicing cessation or continuation, resolved buyer exceptions and reconciled residual rights/cash. Owner: Servicing debt-transfer lead. | Compare the named pool to the executed transfer scope. A proposal or older loan list cannot authorize the transfer. Human risk: Missing POA, incorrect pool or authority beyond actual terms. AI risk: A model can read a request or prior approval as authority for the current action. Match the actual permission, its scope and its effective revision. Hold: Hold prohibited or unauthorized debts and incomplete title. Retain actual delegated servicer execution of joinders/transfer/conveyance instruments where contract authorizes it. Contract/price authority is not presumed. Resolve recourse and misdirected proceeds. Capital raising and securitization structuring remain adjacent. Task checks |
| Reconcile and settle loan-owner proceeds and servicing deductions · Handoff | Reconciled owner deposit/remittance/servicing-fee settlement with accepted instructions and supported loan-level adjustments. Owner: Servicing remittance lead. | Match the owner at the relevant cash and contract cutoff. An otherwise reconciled transfer can still be directed to the wrong owner. Human risk: Wrong owner, conflicting election, missing trust or cut-off evidence. AI risk: A model can match similar records by name or description while linking the wrong identity. Require the source key and its corroborating record. Hold: No deduction without authority/substantiation. Held repurchase proceeds explicit. External bank/trustee actual withdrawal/distribution confirmed. Stale breaks remain owned. Task checks |
| Deliver portfolio, owner and contractual servicing reports · Parallel control | Accepted dated servicing portfolio-performance facts/report, contractual servicing notice or actual required Servicer officer-reviewed certificate, with disclosed exceptions and controlled correction lineage. Owner: Servicing reporting authority. | Verify the report recipient’s scope before dispatch. A copied distribution list can include an owner with no access to the affected population. Human risk: Wrong period, expired or unsupported source, unverified triggered recipient. AI risk: A model can choose the latest-looking document or timestamp while missing the version effective for this event. Verify the governing revision and time from the source. Hold: A generated report is not recipient acceptance or required personal review. Unresolved data defects/defaults disclosed. Do not misassign issuer certificates or trustee investor statements. Task checks |
| Substantiate debt-cancellation tax events and reconcile reporting evidence · Handoff | Tax-accepted servicing event data and reconciled actual filing/debtor-statement/correction or supported nonreportability evidence. Owner: Servicing tax-reporting liaison. | Match the underlying event to the applicable reporting determination. Charge-off alone supplies no proof of debt cancellation or reportability. Human risk: Invalid debtor/event/year, missing creditor/amount or duplicate feed. AI risk: An agent can replay an action when an earlier response timed out. Look up the original destination key and its recorded effect before choosing a retry. Hold: Chargeoff/1099-C is not automatic legal release. Do not report identity-theft debt not incurred. Unresolved Tax decision stays explicit and records reconstructable. Task checks |
| Maintain reconciled backup-servicer takeover readiness · Triggered when applicable | Actual backup-loaded/reconciled readiness file and required completeness/format/readiness certificate/test outcome with cured discrepancies. Owner: Contracted backup-servicer readiness authority. Primary servicing data owner retains source repair. | Reconcile the backup servicer’s reviewed sample to the approved loan population. Successful onboarding of a different sample does not prove readiness. Human risk: Missing docs/authority or invalid formats/loan IDs, unauthorized personnel. AI risk: A model can read a request or prior approval as authority for the current action. Match the actual permission, its scope and its effective revision. Hold: Missing authentic documents/data hold takeover readiness. Primary corrects source. Backup actually loads/reconciles/certifies. Test-data return/disposal tracked. Task checks |
| Transfer servicing with accepted successor and residual obligations · Handoff | Accepted effective successor/ownership servicing handoff with reconciled cutover, payment/customer routing and named residual owners. Owner: Servicing transition authority. | Use the actual receiver assumption and effective time applicable to cutover. Package delivery alone does not make the successor effective. Human risk: Missing actual notice/cure, invalid nominee or unproved transfer authority. AI risk: A model can read a request or prior approval as authority for the current action. Match the actual permission, its scope and its effective revision. Hold: Outgoing service continues for actual required transition. Appointment/prepared file alone is not successor assumption. Incomplete docs/holds block acceptance. Task checks |
| Resolve loan-owner instructions and servicing conflicts · Triggered when applicable | Documented permissible, rejected, held or no-change individual owner-instruction disposition, controlling priority and required conflict notice, with the affected execution owner identified. Owner: Servicing owner-instruction authority, functional reference allocation. | Compare the proposed instruction to the actual written direction from the authorized owner. A previously accepted direction cannot resolve a later conflict. Human risk: Missing instruction, unauthorized sender, wrong owner/pool, conflicted scope. AI risk: A model can match similar records by name or description while linking the wrong identity. Require the source key and its corroborating record. Hold: Complete when lawful direction/hold and required notice are recorded with the source-required evidence. Unresolved instructions remain specifically held/escalated. Other lawful unaffected work continues. Source does not establish automatic owner-response acceptance or an invented universal response SLA. Task checks |
| Review proposed loan transfers and pledges · Triggered when applicable | Documented prohibited/nonprohibited/held proposed-disposition status and source-required notices, description consent when actually required, and continuation-agreement conditions. No completed sale/pledge/successor is implied. Owner: Servicing loan-owner disposition authority, functional reference allocation. | Validate the exact proposal type before recording its decision. A pledge proposal does not create a servicing transfer or change custody by itself. Human risk: Missing/wrong owner or proposal, premature contact, unverified actual receipt. AI risk: A model can match similar records by name or description while linking the wrong identity. Require the source key and its corroborating record. Hold: Keep prohibited/incomplete proposals held or notice delivered as actually required. Do not invent an affirmative Servicer approval requirement for every disposition or treat silence as universal legal approval. A permitted proposal still needs actual owner transaction and any other applicable authority before records, cash or service rights change. Task checks |
Servicing controls, records and operational continuity
| Process and type | Accepted output and owner | Critical failure and release check |
|---|---|---|
| Maintain authoritative servicing file and custody lineage · Core servicing | Complete identifiable retrievable servicing file with authoritative-document/ownership/location lineage. Owner: Servicing records custodian. | Retain the exact accepted record version and custody receipt. A file with the right name can contain superseded evidence. Human risk: Conflicting title, flattened copy presented as authoritative, missing receipt. AI risk: An agent can interpret a submitted or queued event as a completed effect. Use the destination’s accepted status and evidence for the exact item. Hold: Missing or inconsistent document stays an owned deficiency. Ministerial change only within authority. Flattened copy is not authoritative e-contract. Task checks |
| Produce authorized servicing records · Parallel control | Complete authorized record production with delivery/access audit, or supported refusal/nonexistence/hold disposition. Owner: Servicing records response lead. | Limit production to the authorized scope. Even the correct case number does not allow disclosure of the whole file. Human risk: Missing authorization or no existing record, ambiguous subpoena or compulsion. AI risk: A model can read a request or prior approval as authority for the current action. Match the actual permission, its scope and its effective revision. Hold: Do not disclose beyond authority. Missing records return to actual custodian. Legal hold preserved. Theft-specific statutory acceptance remains the “Fulfill identity-theft application and transaction-record requests” process. Task checks |
| Retain, hold and securely dispose servicing evidence · Parallel control | Retained/held retrievable evidence or authorized secure disposition with destruction/access trail. Owner: Servicing retention owner. | Recheck current hold status against the retention decision before destruction. A scheduled disposal cannot override an active hold. Human risk: Wrong activity clock, unsupported destruction, legal hold or tax/furnisher exception missed. AI risk: A model can fill a missing fact from a nearby record or a note. Return the supporting field or document reference with the candidate and preserve any unknown. Hold: No disposal during applicable hold/retention. No blanket perpetual retention. Failed disposal or missing evidence remains owned and escalated. Task checks |
| Release applicable servicing policies, delegated authorities and configurations · Parallel control | Approved effective version of servicing rules, delegated authority, templates and configurations with implemented change evidence or a supported source-required review/no-change disposition. Owner: Servicing policy/change authority. | Activate the approved configuration at the effective time recorded in the policy decision. An earlier timestamp can alter already-processed work. Human risk: Contradictory authority, mistaken date/role or unqualified expansion. AI risk: A model can read a request or prior approval as authority for the current action. Match the actual permission, its scope and its effective revision. Hold: Hold conflicting law/owner instruction or untested rule. Source-of-authority priority resolved. Backfill/migration and training consumers explicit. Task checks |
| Monitor servicing controls and close systemic consumer remediation · Parallel control | Accepted control conclusion with owned corrective action and revalidated affected-consumer remediation or escalated open risk. Owner: Servicing control/remediation authority. | Compare the affected population to the approved remediation scope. A matching row count cannot hide omissions or extra cases. Human risk: Wrong case/cohort or unsupported authorization, actual contract scope unverified. AI risk: A model can match similar records by name or description while linking the wrong identity. Require the source key and its corroborating record. Hold: A report or fix deployment does not prove consumer repair. Identify full affected cohort and reaccept downstream changes. No unknown incidence treated zero. Task checks |
| Accept provider readiness and close delegated-service failures · Parallel control | Approved/conditional/held/terminated servicing provider relationship and accepted cure/performance decision. Owner: Servicing provider relationship authority. | Confirm the provider accepted the exact delegation and that readiness evidence covers that scope. An older approval cannot release new powers. Human risk: Unsupported delegate eligibility, missing contract terms or inappropriate shift of legal duty. AI risk: A model can fill a missing fact from a nearby record or a note. Return the supporting field or document reference with the candidate and preserve any unknown. Hold: Retain actual nondelegable legal responsibilities and the operative contract allocation of performance responsibility/liability, including a valid source-authorized agreement exception where applicable. A SOC report does not cover excluded tools/customer controls. Unresolved service harms retain the responsible operational owner. Task checks |
| Maintain qualified servicing staff and usable channel capacity · Parallel control | Accepted trained/authorized servicing roster and usable channel/queue capacity with owned unresolved gaps. Owner: Servicing operations readiness lead. | Match completion to the training required for the role and release window. Completion on a retired version leaves the capability unverified. Human risk: Missing check or training completion, stale role/access. AI risk: A model can choose the latest-looking document or timestamp while missing the version effective for this event. Verify the governing revision and time from the source. Hold: Do not assign restricted work to unqualified/untrained actor. Backlog/closure and access gaps require temporary routing. HR recruiting/payroll remain adjacent. Task checks |
| Restore and reconcile interrupted servicing capability · Parallel control | Accepted tested/restored service with reconciled replay/backlog, consumer/account effects and residual action owners. Owner: Servicing continuity lead. | Verify the actual restored checkpoint and compare interrupted transactions before replay. A successful restore message does not identify the checkpoint. Human risk: Failed test, stale or unavailable backup, unverified recipient or system recovery prerequisite. AI risk: A model can choose the latest-looking document or timestamp while missing the version effective for this event. Verify the governing revision and time from the source. Hold: Technical restart alone incomplete. Preserve balances/mandates/stays/consents through replay. Unresolved duplicate/missed events held before normal release. Task checks |
| Apply servicing access safeguards and security-control decisions · Parallel control | Accepted servicing access/safeguard implementation and exception evidence with enterprise security handoff. Owner: Servicing system/data owner. | Compare granted permissions with the approved scope, even when the policy revision matches. An admin grant exceeds a read-only decision. Human risk: Undefined role, broad access, missing written alternative or wrong framework. AI risk: A model can fill a missing fact from a nearby record or a note. Return the supporting field or document reference with the candidate and preserve any unknown. Hold: Do not treat servicing coordinator as QI. Actual required enterprise QI written report and governing-recipient receipt/presentation remain adjacent outputs under314.4(i), subject to its coverage/exemption. A separate governing-body review/approval requires its own actual source. Security tests remain actual enterprise producers and failed controls restrict affected access. Task checks |
| Coordinate security-incident servicing protection and recovery · Triggered when applicable | Accepted protected/restored servicing-account disposition and verified notification/response handoffs with consumer remediation. Owner: Servicing security-incident lead. | Preserve the verified discovery timestamp and timezone for the affected lane. A later escalation or ticket creation time is a different event. Human risk: Wrong account population, false alarm, missing actual authority or repeated harm. AI risk: A model can match similar records by name or description while linking the wrong identity. Require the source key and its corroborating record. Hold: No incident closure on technical containment alone. Actual institution/security reporter submits required notices. Incomplete impact/report facts remain owned and updated. Task checks |
| Resolve servicing complaints and regulatory inquiries · Core servicing | Substantive documented response with accepted individual correction/redress or reasoned no-error disposition and systemic issue referral. Owner: Servicing complaint-response authority. | Use the original intake timestamp when an applicable response clock depends on it. Later routing to an investigator does not reset receipt. Human risk: Dropped intake, wrong channel clock, unsupported successor/enterprise responsibility. AI risk: A model can fill a missing fact from a nearby record or a note. Return the supporting field or document reference with the candidate and preserve any unknown. Hold: Acknowledgment/referral alone incomplete. Keep unresolved legal/provider evidence owned. Verify actual consumer remedy and required regulator response. Task checks |
Build and verify one slice first
Start with payment posting. Preserve the original processor trace, amount, currency, event status and received time. Join on the exact transaction identity and account. Recompute the allocation against the effective policy. Confirm the ledger entry by reading it back, then reconcile it against independently verified cash. Route returns and corrections as linked events. A retry consults the original operation key and receipt before any permitted side effect.
Validate this servicing run for the authorized accounts at the recorded cutoff and timezone. Read the accepted account, approved rule revision, transaction and case evidence.
Match account and original event keys, source revisions, effective dates, amounts in minor currency units, approval scope and destination receipt. Return observed and expected values with source references, plus ready_for_review or hold, reason and owner.
Keep unknowns null. A pending processor status or an unverified restriction never releases the affected action. Reconcile prior effects before retries. Validate the corrected revision and its dependent outputs before release.
Use the payment posting exercise to reproduce exact-match, wrong-trace, missing-evidence, duplicate, corrected and uncertain-write decisions. The lab compares normalized synthetic evidence. A production implementation still needs verified processor and ledger interfaces, applicable lender policy and authority, and observed receipts.
Measure the operation before valuing it
Measure accepted serviceable accounts, eligible payment and borrower events, post-control errors, review touches, correction work, elapsed time to resolution and release exceptions. Charge continuing monitoring and incident repair to their actual owners. Count one terminal outcome or shared incident once, even when it crosses multiple processes. The published example supplies no lender volumes, staffing estimates or ROI.
Primary references
- CFPB Regulation E on preauthorized transfers
- CFPB Regulation V on direct disputes
- CFPB Regulation F on covered debt collection
- LoanPro payment processing overview
See Mage’s loan servicing and collections solution for the connected workflow.
