Digital Banking and FinTech Case Studies
Finance5 Case StudiesUpdated Sep 2026
Digital banking and FinTech firms move money faster than back offices were built to check it. The five cases below are drawn from fictional firms but reflect problems that reach finance teams regularly: a ledger that stops agreeing with the bank, a forecast broken by a funding shock, a pattern of transfers needing an escalation call, a credit request built on someone else's data, and an audit that ranks weaknesses. Each asks what the records show, how much risk they carry, what evidence supports the control, and where action should begin.
Case Study 1
Settlement breaks at a payments processor after a holiday cutover
Harrow Ledge Payments, a card acquiring processor that settles about 1,400 merchant batches a day through a pooled clearing account.
The situation
Harrow Ledge settled merchant card batches through a pooled clearing account and matched every batch against three records: its own ledger, the sponsor bank's daily statement, and the merchant payout file. In the first week after moving to a new settlement engine, the daily match rate fell from 99.9 percent to 97.6 percent. The gap was systematic. About 2,300 items a day appeared on the statement but not in the ledger, and the net unmatched balance passed 380,000 dollars by the fourth business day. Dana, head of payments operations, opened a break log and gave Ravi, a reconciliation analyst, the first forty items to trace by hand.
The challenge
The break log showed no single cause. Some items were refunds posted on the value date instead of the settlement date, some were reversals netted before export, and a handful were duplicates from a retry queue. The sponsor bank also applied its cutoff about two hours earlier, so a late batch could land in the next day's statement while its ledger entry stayed put. The team had to separate timing differences that would reverse from true errors that would not, before the month-end close.
Analysis
The first step was to treat the three records as independent witnesses rather than one source of truth. The sponsor statement is external, the ledger is internal, and the payout file shows what merchants actually received. Each break was sorted by age, value, and cause. Ravi applied two thresholds the team agreed before looking at totals: items under 500 dollars and less than two business days old could clear on matching evidence, while anything above 5,000 dollars or older than two days needed a named approver and a written cause. Fixing thresholds in advance kept the bar from moving.
Practitioners working in Digital Banking and FinTech tend to treat a break as evidence about the control, not only about the transaction. The engine change altered the refund posting rule, and no test compared that rule with the sponsor's cutoff schedule. The daily match had a tolerance set as a count, so 2,300 small items looked normal. The redesign moved the tolerance to a value limit, added a cutoff check across both systems, and required a parallel run of five business days before any future engine change goes live. Each control now leaves a dated artifact that the next reviewer can inspect.
- Classify breaks Sort by age, value, and cause before anyone clears an item.
- Trace evidence Pull the bank message or engine record that confirms the timing.
- Escalate or clear Items above threshold go to a named approver with a written cause.
- Harden the control Change the tolerance, add the cutoff check, and log the result.
Actions taken
- Freeze the break log by value and age, then clear every item under the agreed thresholds only where a bank message or engine record confirms the timing. Record the evidence reference beside each clearance.
- Escalate the 5,000 dollar and older than two day items to Dana with a written cause, a proposed entry, and the name of the approver who accepts the residual exposure.
- Replace the count-based tolerance on the daily match with a value limit, and add a cutoff check that compares posting dates between the engine and the sponsor schedule each morning.
- Require a parallel run of at least five business days, with results signed by the controller, before any future change to posting rules reaches production.
Results
Within the next month-end cycle, the unmatched balance fell below 40,000 dollars and stayed there through two further engine releases. The value-based tolerance flagged 61 items in its first week that the count rule had hidden, and 14 of them proved to be duplicate settlements worth about 22,000 dollars, reversed before merchants were paid twice. The parallel run also caught a timezone offset in the second release before it reached production. Dana now reports the match rate by value as well as count, and the controller checks the cutoff comparison every morning.
Questions for discussion
- Which of the three records would you trust first if they disagreed, and what evidence would change that ranking?
- How would you set a clearance threshold for breaks when waiting for confirmation carries its own cost in customer complaints?
- What would a parallel run need to show before you agreed to move a posting rule into production?
Case Study 2
Cash runway forecast after a lending partner pulls its line
Tidewater Neobank, a consumer digital bank with about 320,000 accounts that funds its card and personal loan book through a warehouse line from one partner.
The situation
Tidewater's plan assumed that deposit growth would cover the funding needs of its personal loan book through the year. In the second quarter, net deposit inflow ran about 35 percent below plan, while card spend rose faster than expected and pushed the cost of rewards above budget. Mira, the treasurer, built a thirteen-week cash forecast and found that the warehouse line's borrowing base would shrink if delinquent receivables crossed a set level. Two loan pools were already near that level. The partner had also signaled, in a routine review, that it wanted to renegotiate advance rates before renewal.
The challenge
The forecast had to answer a narrow question: would liquidity stay above the minimum buffer under the board's stress case, and if not, which lever could restore it within the quarter. The base case looked tight but workable. The downside case, where deposit inflow stayed flat and delinquencies rose two points, breached the buffer in the eleventh week. The team had to weigh a partner conversation that had not produced a term sheet.
Analysis
The team built three scenarios, not a single range. The base case carried the observed second-quarter shortfall forward against the plan's deposit curve. The stress case held inflow flat, let delinquencies rise, and assumed the partner cut advance rates. The severe case added a six-week freeze on new card limits, which Mira treated as a lever, not a forecast. Each scenario rested on three assumptions: whether the deposit shortfall was seasonal or structural, whether the delinquency rise sat in one loan pool, and whether the rate cut was written or only discussed. Only the first could be tested against prior quarters, and it did not hold.
The response was judged against thresholds fixed before the numbers were final. Liquidity had to stay above the minimum buffer in the base and stress cases, and the severe case was acceptable only if a lever could be pulled within five business days. Mira set a trigger: a stress breach inside eight weeks would mean cutting origination in the weakest pool and offering the partner alternative collateral. Anything less waited for the next forecast cycle. Each input was also tagged as ledger, management estimate, or partner draft, so the board could see how much of the plan rested on conversation.
Actions taken
- Rebuild the thirteen-week forecast from ledger balances each Monday, and mark every management estimate so it can be challenged in the review.
- Model the buffer breach date for each scenario and report the number of business days between that date and the earliest available funding lever.
- Cut origination in the weakest loan pool by a fixed share now, instead of waiting for the partner, so the buffer holds even if the renewal terms worsen.
- Ask the partner for written advance rate terms before the next committee, and treat any verbal indication as a scenario input rather than a fact.
Results
The team pre-committed a 12 percent cut in origination for the weakest pool, which held the base case above the buffer in every week. The stress case still breached in the seventh week, so the trigger fired and collateral talks began. The partner's written terms arrived near the end of the quarter with advance rates 3 points lower than the verbal indication, which the scenarios had already priced. Tidewater reported the forecast with its input tags visible, and the board approved the revised plan with a monthly buffer review.
Questions for discussion
- Which of the three scenarios should carry the most weight in the board's decision, and why?
- When is a verbal indication from a funding partner strong enough to change the base case?
- What would you need to see in the delinquency data to conclude that the rise is confined to one pool?
Case Study 3
Warning signs on a remittance corridor with rising velocity
Lantern Remit, a money transmitter that moves personal remittances through partner agents and digital wallets.
The situation
Lantern Remit flagged a corridor to a single receiving country after its monitoring rules showed transaction counts for a cluster of about 180 customers rising sharply over five weeks. Average send size fell while frequency doubled, and many transfers were made just under the internal reporting threshold. Several of the customers used new digital wallets that had been opened within the previous month. The compliance analyst, Hussein, pulled the account history for the 40 highest-velocity senders and found that 19 of them shared a funding instrument or a device identifier.
The challenge
The question for the compliance committee was whether these were warning signs of structuring or of something else, such as a commercial arrangement that paid senders to split transfers, or account takeover by a third party. Each explanation implied a different report obligation, a different customer outreach, and a different exposure for the corridor partner. The file also had to show that the decision to escalate, or not, was made by someone with authority and on a documented basis.
Analysis
Indicators were scored against a rubric weighted by how often each signal appeared in cases the firm had already confirmed. Velocity and sub-threshold clustering carried the most weight, shared device identifiers came next, and new wallets alone counted for little. Grouping was essential: 19 senders sharing one funding instrument represented a single decision, not nineteen. The exposure figure, about 1,240,000 dollars in transfers over the window, was reported net of amounts already held. No single customer crossed the filing threshold, so the cluster view carried the case.
Escalation turned on the documentation more than the scores. The file needed the customer-level evidence, the rule that fired, the analyst's reasoning, and the review by a second person who had not seen the first draft. The monitoring band for sub-threshold transfers was too wide, and device identifiers were not captured for transfers through partner agents. Escalation followed the firm's written procedure, which required a decision by the compliance officer within a set number of business days after an alert was confirmed. The officer's memo recorded the decision and the authority under which it was made.
Actions taken
- Freeze the cluster file and assemble customer-level evidence, including funding instruments, device identifiers, and send-size history, before any further outreach to the senders.
- Narrow the sub-threshold band in the monitoring rule and capture device identifiers for partner agent transfers, then backtest both changes against the prior two quarters.
- Route the cluster to the compliance officer with a memo that states the indicators, the exposure, the alternative explanations considered, and the recommended decision.
- Suspend the partner agent's onboarding for the corridor until its transfer records reconcile to its own ledger, and document the reason in the partner file.
Results
The cluster review ended with a suspicious activity report filed inside the required window, and the corridor moved under enhanced monitoring for the following quarter. Nine of the 19 shared-instrument senders were confirmed as unrelated, and their accounts were restored once the evidence was documented. The backtest of the narrower band flagged 14 transfers that had been missed, and three of those led to a second referral. The memo has become the template for cluster reviews, and the officer now signs off on any corridor where two or more indicators coincide.
Questions for discussion
- Which alternative explanation for the cluster would you test first, and what data would separate it from structuring?
- How should a firm weigh a shared device identifier against new account age when the scores conflict?
- What should a second reviewer be able to see in the file that the first reviewer might have missed?
Case Study 4
Growth loan request from a regional merchant lender under review
Halvard Line, a small lending platform that funds working capital loans for independent retailers and sells most of them to institutional buyers.
The situation
Halvard Line originates working capital loans to independent retailers and sells most of them to two institutional buyers within sixty days. It asked for a 12,000,000 dollar warehouse facility to fund new originations ahead of those sales. Its last two years of reported results showed growth in originations and a modest profit, but a rising share of loans had been renewed rather than repaid. Chidi, the credit officer at the digital bank, received audited statements, a loan tape of 6,800 open loans, and a summary of buyer agreements. The tape showed that about 14 percent of balances were past their original maturity.
The challenge
The committee had to decide whether the facility could be repaid from buyer purchases under a downside case, not only the base case the applicant presented. Renewals were the central issue. If renewals were absorbing new capital, the real loss rate was hidden inside a growing book. The committee also needed conditions specific enough to withdraw approval without dispute, if the tape deteriorated.
Analysis
The evidence was tested in three layers. The reported profit was checked against the loan tape, since a profit line can look healthy while rolled balances carry the risk. The tape was reconciled to the audited balance sheet, and the gap of 1.8 percent of principal was large enough to question. Downside scenarios started from the tape: buyer purchases delayed by thirty days, a renewal rate that rose by five points, and a loss rate on renewed loans equal to its worst recent quarter. Under the combined case, the borrowing base covered the facility only if buyers kept buying at current prices, which the agreements did not require.
Suitability for a facility of this kind depends on who bears the loss if the book deteriorates. Here the bank would hold a senior claim on the borrowing base, but the applicant controlled the loan tape that defined the base. That made the tape's integrity the real collateral. The committee set written thresholds: advances at no more than 70 percent of eligible receivables, with renewed loans excluded after one renewal, and monthly loan-level data access. Approval would be rejected outright if the reconciliation gap widened beyond 2.5 percent or if buyer agreements remained unsigned at closing.
- 1Verify the tapeRebuild the borrowing base from loan-level data, not summaries.
- 2Test downside casesDelay buyer purchases, lift renewals, and apply the observed worst loss rate.
- 3Size the advanceCap advances at 70 percent of eligible receivables with renewal exclusions.
- 4Set rejection triggersReconciliation gap above 2.5 percent or unsigned buyer terms ends the request.
- 5Approve with conditionsAdvance only after signed agreements, with monthly loan-level reporting.
Actions taken
- Obtain loan-level data for the full book and rebuild the borrowing base independently, using the bank's own eligibility rules rather than the applicant's definitions.
- Require signed buyer agreements with purchase commitments and pricing floors as a condition precedent to the first advance, not after closing.
- Run the combined downside case through the committee with the renewal rate and buyer delay set at the assumed levels, and record the advance rate that keeps coverage above 1.25.
- Write the rejection triggers into the term sheet, including the reconciliation gap limit and the unsigned agreement test, so both parties know the terms before funds move.
Results
The committee approved a facility of 8,500,000 dollars rather than 12,000,000 dollars, with advances capped at 70 percent of eligible receivables and renewal exclusions in place. Two buyer agreements were signed before the first draw, and the third became a condition of the second advance. In the first full month of reporting, the reconciliation gap held at 1.6 percent and renewals fell as the applicant tightened its own underwriting. Chidi noted that the approval was sized to the tape the bank could verify, not the one the applicant had described.
Questions for discussion
- Which downside assumption would you stress first if buyer pricing changed, and how would the facility size respond?
- When does a renewal stop being a sign of customer health and become a sign of hidden loss?
- What evidence would convince you to approve a facility on the applicant's own loan tape?
Case Study 5
Internal audit findings on a neobank's access and payment controls
Marrowgate Digital Bank, a deposit bank with about 410,000 customer accounts and a payments platform run by a small operations team.
The situation
Marrowgate's internal audit team reviewed its payments operations over most of a quarter after a customer complaint about a duplicated transfer. The review sampled 300 payment instructions and tested user access to the core ledger, the payments queue, and the reconciliation tool. Four findings came out of the work. Two concerned access, one concerned manual journal entries posted without second approval, and one concerned an exception queue that had no aging rule. Adaeze, the internal audit manager, classified the findings and sent them to the chief risk officer with a request for owners and dates.
The challenge
The findings did not carry equal weight. Access rights that let two operations staff approve their own payment releases could reach customer funds directly, while the aging gap in the exception queue mainly affected how quickly errors were seen. The committee needed a ranking it could defend to the board, with owners who had the authority to fix each issue and closure tests that an auditor could rerun without relying on management's word.
Analysis
Materiality was judged on three measures: the value of customer funds that the control could affect, the likelihood that the weakness would be exploited or cause error, and whether the weakness had already produced a loss. The self-approval access gap ranked first. It touched payment releases of about 2,100,000 dollars in a typical month and had already allowed one release that was reversed only after a complaint. The manual journal gap ranked second because it could misstate the ledger without touching customer balances directly. The aging gap ranked last, though it was not trivial, since errors sat unseen longer than the policy allowed.
Owners were assigned by authority over the system, not by who had raised the issue. The access findings went to the head of payments operations, who controls role assignments, and the journal finding went to the controller, who approves posting rules. The aging finding went to the operations manager who runs the queue. Each closure test could be rerun by an auditor: a full role extract checked for self-approval pairs, a sample of manual entries with second approver timestamps, and a queue report with no item past the policy limit. Management's statement that a fix was made was not accepted as closure evidence.
- Days one to fiveRemove self-approval rights and start the daily role extract.
- Week twoJournal approvals enforced and aging rule live in the exception queue.
- End of first monthQueue cleared of overdue items; escalation log reviewed by the operations manager.
- Next audit cycleAuditor reruns all three closure tests from raw extracts, not summaries.
- Following cycleLegacy role owner confirmed, or the role is removed and the finding closes.
Actions taken
- Remove self-approval rights from all payment release roles within the week, and run the role extract daily until the exception list is empty.
- Require a second approver on every manual journal above a set value, with the approval captured in the system rather than in email.
- Add an aging rule to the exception queue that escalates any item older than the policy limit to the operations manager and logs the escalation.
- Report each finding's status to the audit committee at each meeting with its owner, its closure test, and the date the test was last rerun.
Results
Self-approval pairs fell to zero within two weeks, and the daily extract has shown none since. The journal sample reached full second-approver coverage, though one entry needed a retrospective approval that the controller documented. The aging rule cleared the queue of overdue items in the first month. Three findings closed on their tests, while the access finding stayed open one more cycle because a legacy role still had no clear owner.
Questions for discussion
- How would you decide whether a legacy role with no clear owner should be removed or reassigned?
- What would make a closure test weak enough to reject, even when management says the problem is fixed?
- Where should the line sit between a finding that goes to the board and one that is tracked only at committee level?
Learning Resources
Finance CASE STUDIES
China, India, United States, Indonesia, Pakistan, Nigeria, Brazil, Bangladesh, Russia, Japan, Mexico, Philippines, Vietnam, Egypt, Turkey, Germany, United Kingdom, France, Italy, Canada, South Korea, Australia, Spain.