BANK FEE REFUNDS · 8 MIN
Recoup plugs into a user's checking account, parses transaction history for predatory fees, and fires off structured refund claims in under 30 seconds. To date, founder Michael Dean Ferguson claims more than $5 million clawed back, averaging $325 per user. A few steps upstream, Brigit (built by Zuben Mathews and Hamel Kothari) uses continuous liquidity monitoring to kill the overdraft before it ever posts. Both are hunting on the exact same high-margin battlefield: an open regulatory vacuum created when Congress gutted the CFPB’s $5 overdraft cap in May 2025, leaving the banks' $35 toll wide open for extraction.
For decades, getting a bank fee reversed required manual social engineering: find the penalty on a monthly PDF, wait twenty minutes on hold, plead an accidental shortfall to a tier-one customer rep, and rely on a courtesy waiver matrix. If the customer stays silent, the bank pockets pure margin.
Recoup, founded by Michael Dean Ferguson in 2018, turns that manual friction into an automated batch process.
The pipeline is dead simple. The user authenticates via Plaid across a network of 16,000+ institutions. An ingest parser scans the ledger for fee transaction codes (NSF, overdraft, maintenance charges, and late penalties). An agent generates and dispatches formal written claims within 30 seconds. According to Recoup, internal bank review cycles run about 30 days, with institutions routinely granting 30% to 100% of the prior 12 months in goodwill reversals to protect customer retention.
Recoup claims $5,012,253 recovered to date, averaging $325 per customer.
The business model reveals an interesting customer acquisition funnel. While its primary landing page promotes a free tier with zero cuts, tips, or commissions, the broader site gates high-volume claim filings behind paid tiers: $20 a month, $45 a quarter, or an $84 annual subscription. It is a textbook loss-leader engine: hook the user with a frictionless introductory clawback, capture the account credential via Plaid, and upsell ongoing financial dispute automation.
The economic toll here is pure retail extraction: standard fee schedules triggered the millisecond a ledger balance drops below zero.
While institutions like Capital One and Citi eliminated overdraft fees, and Bank of America slashed its penalty to $10, legacy megabanks continue to enforce high-margin caps:
• Chase: $34 per transaction on deficits over $50 (capped at $102 a day)
• Wells Fargo & TD: $35 per item (up to three times daily)
• PNC, Truist, & U.S. Bank: $36 per item
The CFPB tracked $5.83 billion in overdraft and NSF fees in 2023. More importantly, the revenue profile is hyper-concentrated: historical CFPB data shows that roughly 9% of account holders shoulder 79% of all fees. The model survives by repeatedly penalizing a narrow band of liquidity-constrained customers.
The structural vulnerability rests on a friction tax. Under Regulation E, statutory dispute protections apply to accounting errors, but a valid overdraft fee requires an explicit courtesy waiver. The bank retains 100% of unprompted fees simply because human customers face hold times, intimidation, and friction. Yet asking works. LendingTree data shows a 92% waiver rate for customers who request relief.
The landscape shifted across two axes:
The Regulatory Rollback: In December 2024, the CFPB finalized a rule capping overdraft fees at a $5 benchmark or operational breakeven. On May 9, 2025, Congress dismantled it under the Congressional Review Act (Public Law 119-10). The $35 toll remains fully sanctioned.
The Infrastructure War: Financial aggregators turned bank ledgers into machine-readable JSON, enabling software agents to detect fee postings and generate dispute tickets at line speed. Incumbents responded by taxing the perimeter: in September 2025, Plaid agreed to pay JPMorgan Chase bilateral access fees for customer data, while regulators reopened Section 1033 open banking rules for reconsideration.
The arbitrage is clear. Banks rely on human silence to bank billions in pure margin, but an autonomous agent writes the dispute for every user simultaneously.
A mention here is listed, never sold.
LISTED · BANK FEE REFUND APP
Recoup
• The Team: Michael Dean Ferguson (2018), operated by Riddlic, Inc.
• The Vector: Ingests ledger telemetry across 16,000+ banks via Plaid, greps the statement for punitive fee codes, and fires off formal dispute claims in under 30 seconds.
• Traction: Claims $5,012,253 clawed back from bank vaults, averaging $325 per customer.
• The Play: A free, zero-commission front door paired with claim credit subscriptions running $20 a month to $84 a year.
LISTED · OVERDRAFT PREVENTION APP
Brigit
• The Team: Zuben Mathews and Hamel Kothari. Upbound Group acquired it in January 2025 for up to $460 million.
• The Vector: Intercepts the disaster before it hits the ledger. Tracks cash velocity in real time to spot a zero-balance trajectory, then fronts $25 to $500 in micro-liquidity before the bank can swing its hammer.
• Traction: Over $1 billion in avoided overdraft penalties reported, saving roughly $425 per member.
• The Play: Pure recurring SaaS ($8.99 to $15.99 a month). No contingent fee slicing.
Comms · X @hellobrigitapp · hellobrigit.com
Run one transaction the way a CFO would. A Wells Fargo customer overdraws by a few dollars and pays $35. The CFPB's voided rule would have let the bank charge $5 without showing its costs, so measured against the regulator's own benchmark, the bank's spread on that item is $30, and a customer who gets the whole fee refunded takes all $35 back out of the bank's fee income.
On the refund side, old-school bill negotiators love contingency cuts. Rocket Money charges 35% to 60% of whatever they save you. That means on a $35 bank refund, they pocket up to $21 and leave the user wondering why their savior took more than half the loot. Recoup goes the opposite direction: they set their take rate to exactly zero percent on the entry page, letting the customer keep all $35, and then monetize by selling claim credits for about $7 a month on an annual plan.
Brigit looked at the entire dispute model, decided dealing with customer service reps sounded miserable, and bypassed it completely. They charge a flat subscription of up to $191 a year in exchange for keeping the account above zero. The customer avoids $425 in predatory fees, Brigit collects predictable recurring software revenue, and nobody has to write an angry letter to a branch manager.
The second term is cost to serve: what it costs to handle one customer. Your primary operational cost is reading the user's transaction history. Plaid loves to keep their exact per-connection pricing hidden behind a sales team, but make no mistake: those API calls are not free. Now that Jamie Dimon finally got his way and JPMorgan Chase charges Plaid for customer data access, a tiny meter runs every time your background worker refreshes a ledger. If your software polls an inactive account every six hours just to check if someone bought a soda, aggregators will bleed your margins dry before you ever send a dispute.
The third is contribution margin: what each sale leaves after the costs that rise with it: $7 a credit, less the account connection, less the model calls that draft the letter, less support. On an $84 annual claim package, you collect seven dollars a month. Now subtract the Plaid authentication fee, the continuous webhook syncs, the LLM tokens to draft a convincing dispute letter, and the customer support ticket when the bank drags its feet. If your script runs cleanly on the first pass, you clear a gorgeous margin. If the bank ignores the claim and you have to retry three times, you just worked for free.
A quick reality check from the startup graveyard: Cushion launched in 2016, raised over $21 million in venture capital, refunded $15M in bank penalties, and signed up over a million users. In early 2025, they quietly turned off the lights and sold the salvage to LendingClub.
Why does a company with a million users die?
Because an automated complaint letter is the ultimate churn machine. A user signs up with a flaming account, your bot claws back $325 from Chase, the customer throws a party, and then they immediately cancel their subscription because they don't plan on overdrawing their account next week. The customer acquisition cost never recovers.
If you want to build a venture-scale business here, do not just build a fee negotiator. Build a continuous reason to hold the keys to their ledger. The moment the refund check clears, your churn countdown begins.
If you build this for consumers, you will spend your life chasing college kids who cancel their subscriptions the second you claw back their eighty bucks. Build it for small business checking instead. Wells Fargo charges the same $35 fee on commercial accounts, except they bump the daily ceiling to four items instead of three: a cool $140 toll extracted in twenty-four hours because an invoice cleared three hours behind payroll. The business owner absorbs it as a routine cost of doing business, and nobody on the payroll has forty minutes to sit on hold arguing with a branch manager.
The wedge is not the founder. It is the fractional bookkeeper who reconciles twenty or thirty client ledgers every month. They stare at these fee codes all day, hate seeing them, and lack the bandwidth to dispute them. Pull transactions via Plaid for the modern clients, or stand up a dead-simple parser that ingests monthly PDF statements for the paranoid owners who refuse to link an API token.
Your actual moat here is not the AI model. It is a clean routing table that maps each bank's fee schedule to its specific intake channels: secure portal messaging endpoints, compliance web forms, or back-office operations desks. A request sent to general support gets buried, but a request routed to the retail operations queue gets settled. The agent matches the fee, cites the client's average deposit volume and tenure, and dispatches the request from the client's authenticated portal or queues it for a one-click digital signature.
Ditch the SaaS recurring fee. Charge the bookkeeper a flat $5 bounty per successful refund that actually lands on the ledger. No recovered cash, no invoice. Your cold-outreach target is a regional bookkeeping firm managing fifty SMB accounts across Wells Fargo, PNC, and U.S. Bank. Offer to audit their last ninety days of statements for free, show them three grand in unrecovered bank penalties sitting on the table, and let the software run.
RECON · TWO TERRAINS WHERE THE TOLL STILL STANDS
Most founders spend their time building another AI wrapper for project management while billions of dollars in legally obligated payouts rot in bureaucratic queues.
If you want an asymmetric wedge, look for the quiet corners of the economy where an institution legally owes someone money, but banks on human exhaustion and paper forms to keep the cash. The blueprint is always the same: asymmetric administrative friction on one side, zero automated opposition on the other.
Here are two operational battlefields where someone is keeping money that does not belong to them, the paper trails are public, and no builder has claimed the perimeter yet:
Drop in the details, sources, or rough notes for the two specific terrains you picked this week, and we will turn them into sharp, actionable dossiers.
RECON · MORTGAGE INSURANCE
Cancel the PMI on time: The Homeowners Protection Gap
Under the Homeowners Protection Act, a borrower can demand cancellation of private mortgage insurance the day their loan balance hits 80% of the original value, but servicers are only legally required to drop it automatically at 78%. That silent 2% gap costs homeowners $90 to $210 a month on a standard $300,000 note (Freddie Mac estimates PMI at $30 to $70 per $100k borrowed), and the servicer has zero legal obligation to remind them. Additionally, 12 CFR 1024.17 mandates servicers refund escrow surpluses of $50 or more within 30 days of an analysis, yet unmonitored escrow balances routinely sit swollen. The build: an agent parses the closing disclosure, monitors the amortization schedule, and automatically dispatches the statutory cancellation request to the servicer's escrow desk the day the loan crosses 80%. Charge a flat $49 bounty. Distribution is a single phone call to high-volume mortgage loan officers who want an automated retention tool to save their past 3% down clients thousands in phantom fees.
RECON · RENTAL SECURITY DEPOSITS
Get the deposit back
Property managers treat tenant security deposits as free operational float, relying on human exhaustion to pocket bogus paint and cleaning deductions. The statutory trap is strict calendar deadlines: California Civil Code § 1950.5 gives landlords exactly 21 calendar days post-vacate to deliver an itemized disposition or face statutory bad-faith damages up to twice the deposit, while New York General Obligations Law § 7-108 sets a lethal 14-day fuse where missing the clock forfeits the entire right to retain a single dime. Property managers juggling dozens of month-end turns constantly miss these deadlines. The build: an agent ingests the move-out date, time-stamped walk-through photos, and lease metadata, running a countdown that auto-dispatches a formal statutory demand letter via certified mail and portal the morning after the clock expires. Charge a 20% contingency fee on the cash recovered ($400 on a standard $2,000 deposit). The wedge: partner directly with university off-campus housing offices or student legal services, where thousands of graduating seniors are systematically robbed of their deposits every June and have zero time to take a landlord to small claims court.
plaid/quickstart is Plaid's own set of sample apps for linking a bank account and pulling its transactions, with backends in Node, Python, Ruby, Go and Java and a React front end. It has 749 stars, an MIT license, and a last commit on September 27, 2026. It is worth an hour because it takes you from nothing to a list of real transactions, which is the input every fee detector needs, and the fee-matching rules you write on top of it are the part we could not find in any open source project this week.
From a counterintelligence and fincrime perspective, an automated fee-recovery app is a magnet for two attacks: cloned credential harvesting and weaponized authorization. Threat actors will spin up lookalike front doors with fake Plaid modals and ticking recovery counters to siphon raw bank logins from desperate consumers. On the flip side, malicious actors will abuse legitimate recovery infrastructure: pointing an automated dispute bot at accounts they do not own, or weaponizing the bot's structured bank correspondence as a script to socially engineer tier-one branch reps.
The defensive baseline is absolute credential isolation and identity verification. Never touch or store a bank password: ingest ledger data exclusively through bank-issued OAuth tokens locked strictly to read-only transaction scopes. Match the legal account title to the verified app user before drafting a ticket, and route every outgoing request through the customer's authenticated portal or an auditable digital signature.
This boundary is exactly why we built FLINT. When you deploy autonomous software to challenge institutions, negotiate fees, or touch live ledgers, the agent itself becomes an attack surface. Without cryptographic identity, strict verification, and hard execution bounds, you haven't built an autonomous fintech company: you've built an unmonitored liability waiting to be compromised.
PASS IT ON
Who do you know that paid a $35 overdraft fee this year and never asked for it back? This issue is for them.
NEXT ISSUE
Working on it 😀
SOURCES
Recoup fees-back page · Recoup FAQ · Recoup pricing · Crunchbase Recoup founder · Brigit $1 billion saved · Brigit plan pricing · Upbound Brigit deal terms · Upbound Brigit closing · Wells Fargo service fees · Chase overdraft practice · PNC fee schedule · Truist fee schedule · U.S. Bank pricing · TD Bank fees · Bank of America fee schedule · Capital One overdraft options · Citi overdraft announcement · CFPB 2023 overdraft revenue · CFPB frequent overdrafters · Federal Register overdraft rule · GovInfo Public Law 119-10 · eCFR Regulation E 1005.11 · Fox 5 LendingTree fee survey · PaymentsJournal Plaid and Chase · Federal Register 1033 reconsideration · Plaid pricing page · Rocket Money negotiation fee · Cushion company history · PR Newswire LendingClub buys Cushion · Freddie Mac PMI costs · CFPB PMI cancellation · eCFR escrow rule 1024.17 · California Civil Code 1950.5 · New York GOL 7-108 · GitHub plaid/quickstart · Y Combinator Coverage Cat launch · Insurify Muse block release
