Behavioural scorecards after account opening. A practical lesson in credit risk models for banking and payments practitioners.
How to study this topic
Behavioural scorecards after account opening help a bank monitor existing customers using actual account conduct, repayment behaviour, utilisation, arrears movement, customer activity and risk changes over time. The phrase may sound technical, but the banking meaning is practical. A bank uses this concept to decide how much risk it is taking, how much loss it may face, how to price or limit exposure, when to refer a case, how to monitor portfolios and how to explain credit decisions.
Keep the scope banking-first. This chapter is about lending, borrowers, counterparties, credit obligations, portfolios, limits, collateral, repayment, arrears, provisions, capital, governance and customer treatment. It is not a payments chapter. Transaction behaviour may appear only as account conduct evidence where it genuinely supports credit risk.
A learner should finish this topic able to explain it to a credit analyst, business analyst, data engineer, model validator, underwriter, collections manager and auditor. Plain language matters because credit-risk models are not useful if only modellers understand the answer.
The plain language meaning
A behavioural scorecard answers the question: after the account is open, what does the customer's real behaviour tell the bank about current and future credit risk? It uses lived account conduct rather than only onboarding information.
The concept must always be tied to purpose. A number without purpose is dangerous. A PD used for application decisioning may differ from a PD used for IFRS 9, capital or portfolio monitoring. An LGD used for pricing may differ from a downturn LGD view. An EAD for stress testing may differ from day-to-day exposure monitoring. Scorecards also differ by product, portfolio and decision point.
Plain language does not mean weak explanation. It means the explanation should remain accurate without hiding behind formulas.
Source systems and evidence
Useful source areas include repayment behaviour, account conduct, limit utilisation, arrears history, missed payment pattern, overdraft usage, income movement, product usage, collections contact history, forbearance markers, bureau updates, and early warning signals. A credit-risk chapter depends on controlled data because every input can change the risk view. Income evidence, arrears data, utilisation, collateral value, limit data, customer identity, bureau information and case outcomes all need clear source ownership and timing.
The bank must know which source is authoritative. A front-end application may show what the customer declared. A document-verification platform may show what was evidenced. A core loan system may show what was booked. A collections system may show what happened after stress started. A risk engine may show the rating and score. Finance may show impairment treatment. These views must be mapped, not casually mixed.
The source question becomes sharper when AI and ML are used. If a model learns from a field, the bank must know whether that field is complete, stable, corrected, overridden, refreshed and legally usable for the purpose.
Key inputs and calculations
Key inputs include repayment behaviour, account conduct, limit utilisation, arrears history, missed payment pattern, overdraft usage, income movement, and product usage. The calculation must define the time horizon, observation period, product scope, customer population, exclusion rules and treatment of missing values. Without these, the same concept can be calculated in several different ways and produce conflicting results.
For behavioural scorecards, the bank needs post-opening account conduct, repayment history, arrears movement, utilisation, limit changes, income movement, collections contact, forbearance status and bureau refreshes. Behavioural models must separate temporary noise from meaningful deterioration.
The formula is only one part. Banking interpretation, data quality and governance decide whether the output is safe to use.
Controls before use
Controls should include behaviour window definition, repayment calendar alignment, arrears bucket quality, utilisation calculation, customer status changes, and collections outcome validation. These controls must run before the concept feeds a model, decision engine, risk report, provisioning process or strategy review. Technical checks are not enough. The bank must check whether the value makes sense in credit terms.
Important checks include population coverage, source reconciliation, date alignment, treatment of overrides, treatment of outliers, missingness analysis, validation against realised outcomes and monitoring of drift. Credit-risk models can fail slowly. They may not throw a system error. They may simply become less aligned with borrower behaviour or economic conditions.
The bank should also define who can override the model, how overrides are recorded, and how override performance is reviewed. Human judgement can improve credit decisions, but undocumented override culture can damage validation.
How it supports bank decisions
Common uses include limit management, collections prioritisation, early warning, credit line increase review, customer support triage, portfolio monitoring, IFRS 9 staging support, and behavioural PD update. Each use has a different control requirement. A model used for internal portfolio insight may need one level of governance. A model used for credit approval, regulatory capital, provisioning or customer treatment needs much stronger evidence, validation and monitoring.
Behavioural scorecards help the bank update risk after the relationship has started. They can identify improvement, deterioration, early warning and support needs. The bank should use them carefully because existing customer treatment can directly affect customer wellbeing.
The output should be explainable to the decision owner. If a credit manager cannot understand why the model changed a view, the bank has a governance problem even if the statistical result is strong.
Provisioning, capital and portfolio view
Credit-risk concepts matter beyond individual decisions. They feed portfolio monitoring, risk appetite, concentration review, stress testing, capital planning and expected credit loss. The same borrower-level evidence can roll up into management and regulatory views.
IFRS 9 expected credit loss uses credit-risk movement to decide whether a financial asset remains in a 12-month expected loss view or moves into lifetime expected loss because credit risk has increased significantly. That makes data timing, default definition and risk movement important. A poor risk signal can affect provisions and reported performance.
Capital frameworks also care about credit-risk parameters. PD, LGD and EAD are not classroom abbreviations. They are part of how banks understand expected and unexpected loss, risk-weighted assets, portfolio quality and risk-adjusted return.
Customer and conduct impact
Credit-risk models affect customers. They can approve credit, restrict credit, increase price, reduce limit, refer for manual review, trigger collections activity or identify customers needing support. A wrong model can harm a customer by granting unaffordable credit or by denying reasonable credit without a fair basis.
Fairness is not solved by removing obvious sensitive fields. Proxies can appear through income, geography, employment type, channel usage, account history, product access or missing data. The bank should review whether the model treats customer groups fairly and whether limitations are visible to decision-makers.
Human review is important where impact is high. A score can support judgement, but the bank remains accountable for the decision.
Validation and monitoring
Validation should review concept, data, methodology, calibration, discrimination, stability, limitations and use. It should not only ask whether the model predicts. It should ask whether the model is suitable for the bank's portfolio, product, policy, customer population and decision purpose.
Monitoring should compare predicted outcomes with actual outcomes. For PD, defaults should be compared with predicted default rates. For LGD, realised loss and recovery experience should be reviewed. For EAD, drawdown and exposure at default should be tested. For scorecards, approval quality, referral rates, override outcomes and portfolio performance should be monitored.
If performance deteriorates, the bank may need recalibration, redevelopment, policy adjustment, overlays, tighter controls or temporary restrictions. Model monitoring must have a response path.
Common mistakes
The first mistake is confusing the parameter with the whole credit decision. PD, LGD, EAD and scorecards are decision inputs, not the entire decision. The second mistake is using weak labels. The third mistake is ignoring data timing. The fourth mistake is assuming a score remains valid after products, policy or economic conditions change.
Another mistake is treating a model as neutral because it is quantitative. Credit models reflect historical data, policy decisions, source availability, customer access and economic cycles. The bank must challenge whether the model learned credit risk or learned old operating patterns.
The final mistake is failing to explain limitations to users. A model can be useful with limitations if users understand them. A model becomes dangerous when limitations are hidden.
Practical banking example
An existing credit-card customer begins using more of the limit, misses one payment, makes minimum payments only and shows reduced salary credits. A behavioural scorecard may identify increased risk and route the customer for support or limit review, depending on policy and fairness controls.
The practical point is that credit-risk models support judgement. They should make banking decisions more consistent and evidence-based, but they should not remove responsibility from the bank.
Bank-ready checklist
Before using this concept in production, ask: is the definition clear, is the population right, are sources authoritative, are time windows correct, are labels validated, are overrides governed, are missing values explained, are customer impacts reviewed, and can the result be reproduced later?
Then ask: is the model calibrated, independently reviewed, monitored, documented and approved for this exact purpose? Is it being used for approval, pricing, provisioning, capital, collections or internal insight? Has the bank matched controls to impact?
If the answers are strong, the concept is bank-ready. If not, the model may still produce numbers, but the bank should not treat those numbers as trusted credit-risk evidence.
Scoring an existing relationship
A behavioral scorecard assesses an existing borrower using observations gathered after account opening. It may help a bank monitor a credit portfolio, prioritize supportive outreach or review an account limit under approved policy. Its information set is different from an application scorecard's. It can observe actual repayment, utilization and account activity, but only as available at each score time. The bank specifies the product, eligible accounts, target event, horizon, score cadence and action. A score is an estimate for a defined use, not a license to infer why a customer behaved in a particular way.
An account that has been active for three years offers more repayment history than one opened last month. An account closed or transferred during a performance horizon needs explicit treatment. A behavioral model trained only on long-tenure accounts should not silently score new accounts with fabricated zeros. State minimum history and missingness rules. A customer with no missed payments in two observed months is not equivalent to one with none in thirty-six months.
Repayment observations
A scheduled instalment comes from a contract and its effective amendments; an actual payment comes from the ledger. A missed-payment feature needs a grace-period rule, partial-payment treatment, holidays, reversals and cutoff. A ledger posting delayed by a source outage might temporarily make a customer appear late. The model service can expose pending or stale states and use a referral or fallback. It should not turn a missing ledger feed into a delinquency count or a perfect payment record. Test transactions at the exact grace boundary and a payment corrected later.
For revolving credit, utilization depends on drawn balance and current limit at a defined time. A limit cut by the bank raises utilization without additional borrower spending. A model should distinguish an account action from customer behavior when interpreting a sudden change. Minimum-payment patterns, cash advances and balance growth can be candidate features, but their predictive value and fairness need validation for the particular product. A numerical trend does not reveal a customer's hardship, intent or ability to repay without further evidence.
Account and transaction context
Some behavioral scorecards use deposit-account activity, such as qualifying income credits or net cash flow, where lawful and relevant. A regular credit can be a transfer from another owned account rather than income. A customer may change the account into which salary is paid without losing employment. Define the relationship map, transaction classification, observation window and source coverage. If consented external data become unavailable, the model should distinguish withdrawn access, no history and technical failure. Do not turn absence of one channel's data into an adverse behavioral conclusion.
Bank account activity can reflect product design. A new mobile app may shift customers from branch to digital channels. A fee change may alter transfer timing. A repayment holiday can change scheduled cash flows. Validation compares source and policy changes with model input distributions before treating a shift as borrower risk. Behavior is observed through systems and interventions, not directly from a person's intentions. A feature contract should make those limitations visible to model users.
A monthly scoring snapshot
Suppose the bank scores active loans on the first business day of each month. The run manifest lists eligible facilities, the as-of date, source files, schedule versions, model artifact and policy version. A particular loan has one late payment under the approved rule, a moderate rise in utilization and an incomplete income feed. The feature service returns values and missingness reasons. The scorecard maps them to points or a risk estimate. Policy decides whether to make no change, refer for review or offer a supportive contact. It does not equate a high risk score with a default already occurred.
A second file arrives late after some accounts were scored. A corrected run should have a new identifier and explicit supersession. Operations needs to know which list was acted on and whether a customer was contacted. The original scores remain evidence of what staff saw. A subsequent replay can estimate how corrected data would have changed the action. A published batch without reconciliation of eligible, scored, skipped and failed accounts can give a false picture of portfolio risk.
Outcome and selection
Define the target, such as entering a particular arrears state in the next six months, with a stable rule and adequate follow-up. Recent scored accounts have not yet matured. A loan that prepays, is sold or restructures during the horizon needs an approved treatment. Behavioral models observe only customers who were approved and remained in the bank's portfolio. Their repayment outcomes can be influenced by earlier limit decisions, collections actions and assistance. An apparent model improvement may reflect a new intervention policy or safer account mix.
If a score triggers outreach that prevents a missed payment, labeling the alert false because no arrears occurred ignores the intervention. Record score, policy action, staff contact, assistance and eventual outcome. A bank can compare interventions on appropriately defined cohorts, but an observational before-and-after study may not identify a causal effect. Report uncertainty and avoid retraining on a short set of selectively contacted accounts as if they represented the full population.
Scorecard design and interpretation
Behavioral points tables can show how bands of repayment history or utilization contribute to a score. The bank should inspect stability of each band, missing categories, population counts and whether multiple correlated variables duplicate the same signal. A score-to-risk mapping requires calibration on the intended active-account population and horizon. The number of points is not a universal PD. Test exact band boundaries and changes due to source corrections. Preserve the fitted table, input definitions and policy version for each run.
Explanation is contextual. A score may fall because utilization rose after the bank reduced a limit, not because the borrower spent more. A reviewer should see the source balance and limit events. A model explanation identifies contributions under its own fitted structure, but it cannot determine whether the source is accurate or whether an action is fair. The customer-facing reason for a limit action or other adverse result must reflect the actual model and policy factors under applicable law and bank policy, with a route to challenge errors.
Portfolio and segment validation
Validate on later account vintages with mature target outcomes. Check discrimination, calibration, stability and operational action rates by product, tenure, channel and relevant customer group. A model can rank risk well on average and overstate it for newly migrated accounts. Compare a simpler baseline and an incumbent scorecard on the same dated cohort. Inspect whether features available in a complete warehouse were actually available at monthly score time. A performance report should include population counts, exclusions and outcome maturity.
Monitoring uses different clocks. Input missingness, source freshness, scores, referral volume and overrides are visible immediately; defaults and loss outcomes take longer. A sudden rise in high-risk scores can reflect a new repayment code or account mapping, not a deteriorating portfolio. Investigate source changes and bank actions before changing thresholds or retraining. Monitor human capacity and the timeliness of supportive outreach; a model with a useful ranking may have little impact if the queue cannot act.
Fairness and supportive use
A short bank history can be common among new customers, seasonal workers or people who previously used another provider. A behavior feature based on app use may reflect access, disability or product design. Review whether model actions disproportionately burden relevant groups where law and data permit, and whether a better evidence path exists. A bank may decide to refer uncertainty for human review rather than automatically cut a limit. Interventions should be proportionate and documented. A model should not turn a customer's request for hardship support into an unexamined punitive signal.
Human review needs access to the schedule, payments, missingness and customer circumstances within appropriate privacy controls. The reviewer can correct data, recognize a legitimate holiday and record a reasoned action. Track overrides and complaints to discover systematic issues. A high volume of automated referrals with insufficient staffing can delay help or make human checks superficial. Test workflow capacity when thresholds are changed and include customer outcomes in the review, not only bank loss metrics.
Limit management boundary
A behavioral score can inform a credit-line review, but a line change involves additional policy, affordability, notice and conduct considerations. A reduced limit itself increases utilization, which can feed the next score and create a feedback loop. The model owner should record limit interventions and test how the score behaves afterward. A policy should not treat a mechanically higher utilization caused by its own action as fresh independent evidence of customer deterioration. For a proposed increase, a strong score is not a substitute for affordability and eligibility assessment.
Consider a customer whose limit is reduced from 10,000 to 6,000 while the drawn amount stays 4,000. Utilization rises from 40% to about 67% with no new borrowing. A model that attributes the rise to changed behavior can compound the first action. A feature service can retain both balance and limit with their effective dates; validation can test this case. Product and risk owners decide whether a post-intervention feature or model restriction is needed.
Incident and replay
A servicing migration changes the meaning of a payment status code, causing valid holiday months to appear missed. The scorecard runs overnight and refers hundreds of accounts. Monitoring sees a sharp rise in the missed-payment feature and queue size. The incident owner stops the affected automated action under an approved fallback, preserves run IDs, source status and original scores, and corrects the mapping. A replay identifies customers whose referral or limit action would change. Servicing and product owners determine remediation; the model team verifies that no retraining absorbed the bad data.
A fix includes a semantic test for payment holidays, partial payments and reversals, plus a source-change notification to all consumers. The release compares old and corrected values by product and account age, then monitors the next run. The bank cannot close the incident merely because the batch succeeds again. It must reconcile affected decisions and contacts, including cases where a customer was not reached in time for supportive assistance.
Change control
A planned update might extend the repayment window from six to twelve months. Run both versions on the same dated accounts and compare feature values, scores, referrals, capacity and mature outcomes. A longer window can stabilize sparse histories but can also overemphasize older issues after a customer has recovered. Validate by tenure and product. Version the feature, scorecard and policy together; keep old snapshots for audit. A new model cannot automatically use the revised field without compatibility and validation evidence.
Before promotion, test an ordinary current account, a new account with short history, a legitimate holiday, a partial payment, a limit cut, a closed loan, a migrated account and a feed outage. Write expected feature values and policy actions. Confirm that out-of-population accounts are excluded or referred rather than scored confidently. Reconcile batch output counts and confirm that a correction creates a new run rather than silently overwriting yesterday's list.
A change in payment frequency
Repayment data can look different when a borrower changes from monthly to fortnightly payments, pays early, or switches the funding account. A naive thirty-day missed-payment count can misclassify a valid new schedule. The scorecard's source feature should derive due dates from the current approved contract and match receipts under the bank's payment allocation rules. If a customer pays an extra amount toward principal, that does not necessarily satisfy a later scheduled instalment; the product terms determine treatment. Test these cases with a servicing specialist before treating their patterns as risk signals.
The model may use a trend in account balance or income credits as a contextual signal, but a change in payment frequency can alter those patterns mechanically. A monthly score should identify complete periods and source coverage. A score just after a holiday weekend may see delayed postings from another bank. A valid absence of data is not the same as a failed feed. Preserve the feature value as served, availability timestamp and later correction so a reviewer can distinguish a genuine change in behavior from data timing.
Feedback from collections and support
Once a model prioritizes accounts for outreach, its own policy changes the distribution of future observations. A supportive reminder may prevent a missed instalment. A temporary payment arrangement changes scheduled amounts. A line freeze affects utilization. Training on later outcomes without recording those actions can teach the next model that contacted accounts naturally behave differently. Link each score to the intervention, its timing and a defined outcome. Evaluate model ranking and intervention effectiveness separately.
Investigators and servicing agents may write free-text notes that include sensitive hardship information and conclusions reached after the score. Those notes should not be copied wholesale into a live feature pipeline. If a narrowly defined indicator is justified, specify its source, timing, legal use and validation, and ensure it was actually available before the intended decision. A collections-stage code can leak the target if the model is supposed to forecast deterioration before collections. High predictive power from such a field should prompt a lineage check.
A portfolio migration test
Suppose the bank moves a credit-card book to a new servicing platform. Account IDs and transaction codes change, while credit limits and repayment contracts remain. Before the first production score, map old to new identifiers with effective dates, compare balances, schedules and payments on a frozen sample, and calculate old and candidate behavioral features side by side. A customer with three years of history should not suddenly have zero observed months. An overlimit flag must use the true limit, not a migration placeholder.
The model owner compares score distributions and actions for matched accounts, separately reporting unmatched and partial records. A perfect aggregate balance reconciliation does not prove individual feature values are correct. Test reversed payments, statement dates and joint holders. If the model service cannot establish parity, pause automated actions or refer cases under the approved contingency. Keep both platform records and original scores for later investigation. This is a source and feature release even if the model weights do not change.
Controlled challenger comparison
A new behavioral model may use richer transaction features than the incumbent scorecard. Score both on the same eligible dated accounts in shadow mode and record source coverage, missingness, latency and output disagreements. The challenger cannot be called better because it generates a wider range of scores. Compare mature outcomes and the effect of plausible policy thresholds, including outreach capacity and customer friction. Segment by tenure, product and relevant groups. If the richer inputs are unavailable for a material cohort, the deployment plan needs a referral or separate validated path.
For a controlled live change, define which model and policy each account receives, preserve assignment and intervention records, and set rollback triggers. The later outcome may be affected by different outreach under the two paths, so interpret performance with the experiment design in view. A model with slightly better ranking but much higher source failure or review burden may be the wrong operational choice. Approval should state the decision, population, feature versions, fallback and monitoring owner.
A decision trace for one borrower
Take a fictional borrower with a revolving line and an installment loan. On 1 April, the bank's behavioral model sees a rising card balance and one instalment that appears late. The servicing record later shows a valid grace-period payment. The original score and referral remain in the audit log. The corrected view removes the missed-payment flag and permits an impact review. If staff had offered a supportive contact, record whether it occurred, what evidence was used and whether the customer disputed the alert. Do not claim a later non-default proves the alert was wrong without considering the contact.
The model can help the bank notice emerging risk, but its usefulness rests on a precise contract for each behavior, a mature outcome definition and an action that respects the customer. A well-governed scorecard separates source events, score, policy and intervention; it can be reproduced at each monthly cutoff and corrected when evidence changes. That makes behavioral ML a practical tool for portfolio care rather than an unexplained judgment about a person.
An independent reviewer can sample a customer who was contacted, one whose case was referred but not reached, and one whose score stayed low after a source outage. Reconstruct the feature snapshots and confirm whether the intervention rules behaved as approved. The bank should report unresolved cases and late outcomes separately. This check makes the scorecard's value and its limitations visible across the whole portfolio, including customers whom the workflow failed to serve.
Banking practice note on data definition
For behavioural scorecards, data definition matters because credit risk is not only a model output. It is a banking responsibility that connects borrower behaviour, product policy, capital, provisions, customer treatment, collections, finance reporting, risk appetite and audit evidence. If one part is weak, the model result can look precise while the decision remains fragile.
The practical discipline is to keep source fact, credit-risk interpretation, model estimate and business action separate. Source facts come from areas such as repayment behaviour, account conduct, limit utilisation, arrears history, and missed payment pattern. The credit-risk interpretation explains what those facts mean. The model estimate turns that interpretation into a score, probability, severity, exposure or ranking. The business action decides approval, referral, monitoring, support, pricing, limit or reporting treatment.
This separation helps the bank challenge the result properly. When a score changes, the bank can ask whether borrower behaviour changed, source data changed, policy changed, model calibration changed or economic conditions changed. That is the kind of clarity serious banking AI needs.
Banking practice note on model use
For behavioural scorecards, model use matters because credit risk is not only a model output. It is a banking responsibility that connects borrower behaviour, product policy, capital, provisions, customer treatment, collections, finance reporting, risk appetite and audit evidence. If one part is weak, the model result can look precise while the decision remains fragile.
Banking practice note on customer impact
For behavioural scorecards, customer impact matters because credit risk is not only a model output. It is a banking responsibility that connects borrower behaviour, product policy, capital, provisions, customer treatment, collections, finance reporting, risk appetite and audit evidence. If one part is weak, the model result can look precise while the decision remains fragile.
Banking practice note on portfolio control
For behavioural scorecards, portfolio control matters because credit risk is not only a model output. It is a banking responsibility that connects borrower behaviour, product policy, capital, provisions, customer treatment, collections, finance reporting, risk appetite and audit evidence. If one part is weak, the model result can look precise while the decision remains fragile.
Banking practice note on validation
For behavioural scorecards, validation matters because credit risk is not only a model output. It is a banking responsibility that connects borrower behaviour, product policy, capital, provisions, customer treatment, collections, finance reporting, risk appetite and audit evidence. If one part is weak, the model result can look precise while the decision remains fragile.
Banking practice note on monitoring
For behavioural scorecards, monitoring matters because credit risk is not only a model output. It is a banking responsibility that connects borrower behaviour, product policy, capital, provisions, customer treatment, collections, finance reporting, risk appetite and audit evidence. If one part is weak, the model result can look precise while the decision remains fragile.
Banking practice note on governance
For behavioural scorecards, governance matters because credit risk is not only a model output. It is a banking responsibility that connects borrower behaviour, product policy, capital, provisions, customer treatment, collections, finance reporting, risk appetite and audit evidence. If one part is weak, the model result can look precise while the decision remains fragile.
Banking practice note on audit evidence
For behavioural scorecards, audit evidence matters because credit risk is not only a model output. It is a banking responsibility that connects borrower behaviour, product policy, capital, provisions, customer treatment, collections, finance reporting, risk appetite and audit evidence. If one part is weak, the model result can look precise while the decision remains fragile.
Banking practice note on economic conditions
For behavioural scorecards, economic conditions matters because credit risk is not only a model output. It is a banking responsibility that connects borrower behaviour, product policy, capital, provisions, customer treatment, collections, finance reporting, risk appetite and audit evidence. If one part is weak, the model result can look precise while the decision remains fragile.
Banking practice note on human judgement
For behavioural scorecards, human judgement matters because credit risk is not only a model output. It is a banking responsibility that connects borrower behaviour, product policy, capital, provisions, customer treatment, collections, finance reporting, risk appetite and audit evidence. If one part is weak, the model result can look precise while the decision remains fragile.
A dated account review
A behavioral scorecard uses repayment and account activity after origination. Choose a review date, eligible active accounts and history window. A payment made after the cutoff cannot improve the prior month's score, and a backdated correction must be kept separate from what the servicing team saw then. Define delinquency, utilization, limit change, cash flow and product state with source owners and missingness rules.
Consider a borrower with rising utilization and one missed installment who subsequently cures. A monthly score can detect the earlier deterioration without declaring permanent default. Compare the score with current hardship or collections policy before any customer action. A model may prioritize an account for review; trained staff decide outreach or assistance under approved rules. Monitoring includes performance on mature outcomes and the customer effect of interventions, because an intervention can alter later observed repayment.
Reconcile the batch population to servicing balances and accounts, including closed, restructured and migrated facilities. A missing branch extract can produce a deceptively small low-risk portfolio. Test a partial file, duplicate account, payment reversal and changed status code. Preserve score version, reason factors, action and override so a later reviewer can distinguish model drift from source and policy change.
Primary sources for further study
This application uses JavaScript for the full interactive experience. This text summary is served for accessibility and search indexing.