Between 2021 and 2025, consumers filed 97,028 complaints against major US fintechs about money transfers, digital wallets, prepaid cards, checking products, and consumer installment loans. We classified every complaint by its underlying cause.
7.26% of them, roughly 1 in 14, describe patterns consistent with a reconciliation failure: a payment that moved but was not recorded, matched, or settled correctly across systems. From 2021 to 2024, the share of reconciliation-related complaints climbed steadily from 11.55% to 14.43% of all fintech payment complaints, before being diluted in 2025 by a regulatory-event-driven surge in non-reconciliation complaints. When these complaints reach the CFPB, they receive monetary relief at 10.39%, nearly three times the 3.51% relief rate for the rest of the dataset.
Reconciliation failures are usually invisible until a consumer files a complaint. A duplicate charge, a transfer that leaves one account and never arrives in another, a refund that never posts: each is a consumer-visible pattern consistent with two systems disagreeing about the same transaction. This report quantifies how often that pattern appears across the largest consumer fintechs, how its share has shifted, and how it resolves compared to other complaint types.
Key findings
- 7.26% of fintech payment complaints (7,048 of 97,028) describe patterns consistent with reconciliation-related failures, distinct from fraud, account access, or service quality complaints.
- Reconciliation-related complaints rose from 11.55% of all fintech payment complaints in 2021 to 14.43% in 2024, before the share compressed in 2025 due to a surge in non-reconciliation complaints tied to a single regulatory event. In absolute terms, reconciliation-related complaints grew from 677 in 2021 to 3,109 in 2025.
- Block (Cash App), Chime, and PayPal account for the largest raw complaint counts, with 2,012, 1,655, and 1,300 reconciliation-related complaints respectively. These are raw counts that scale with platform size, not failure rates.
- Reconciliation-related complaints close with CFPB-recorded monetary relief at 10.39%, nearly three times the 3.51% rate for non-reconciliation complaints in the same dataset. When these disputes reach a regulator, they are recognizable and remediable in a way that fraud and account-access complaints often are not.
What counts as a reconciliation failure, and what does not
Reconciliation is the process of confirming that two or more records of the same money movement agree. A payment processor says a transfer settled. An internal ledger says it did not. A bank feed shows one amount, a provider report shows another. When those records disagree and no one catches it, the money is effectively lost in the gap between systems. That gap is where reconciliation failures live.
The distinction that drives this analysis is between a payment that failed to reconcile and a payment that simply went wrong for another reason.
| Reconciliation failure (counted) | Not a reconciliation failure (excluded) |
|---|---|
| “I sent money, it left my account, and the recipient never received it.” The transaction was recorded as sent by one system and never recorded as received by another. Two systems disagree about the same event. | “Someone hacked my account and drained it.” Money moved, both systems recorded it correctly, but the transfer was unauthorized. Nothing failed to match. This is a fraud and authorization problem, not reconciliation. |
| “I was charged twice and only refunded once.” A duplicate that was never reconciled against the original, leaving a balance the two systems do not agree on. | “I cannot log in and support will not respond.” An access and service problem. No transaction-level matching or recording error is described. |
The test is not how the consumer labels the problem. It is whether the sequence of events they describe is consistent with a recording, matching, settlement, or transaction-ledger error.
How the classification works
The classifier does not read a complaint and trust the consumer’s diagnosis. Consumers rarely know what happened inside a fintech’s ledger, and most have no idea their problem is a reconciliation issue at all. The model reads the sequence of events the consumer describes and asks a single question: is this sequence consistent with a recording, matching, settlement, or transaction-ledger error somewhere in the system?
Each complaint also carries structured metadata: product type, sub-product, issue category, sub-issue, the company, and the company’s response. That structured context helps the model reason about the most likely root cause even when the consumer’s account is vague. The classifier output was validated against a hand-labeled sample of 80 complaints; full agreement metrics are in the methodology section.
Reconciliation-related complaints rose steadily as a share of fintech payment complaints from 2021 to 2024
Across all 97,028 fintech payment complaints filed between 2021 and 2025, 7,048 describe reconciliation-related failures. The share by year tells a story in two parts.
| Year | Reconciliation-related | All fintech payment complaints | Reconciliation share |
|---|---|---|---|
| 2021 | 677 | 5,862 | 11.55% |
| 2022 | 701 | 5,867 | 11.95% |
| 2023 | 1,128 | 7,780 | 14.50% |
| 2024 | 1,433 | 9,934 | 14.43% |
| 2025 | 3,109 | 67,585 | 4.60% |
From 2021 to 2024, the share of reconciliation-related complaints climbed from 11.55% to 14.43%, while overall fintech payment complaints in the dataset roughly doubled. Reconciliation complaints were not just keeping pace with the rise in complaint volume. They were outpacing it.
2025 looks like a structural break. Total fintech payment complaints in the dataset jumped roughly sevenfold in a single year, driven in large part by template complaints tied to the CFPB’s December 2024 lawsuit against Early Warning Services, the operator of Zelle. Many of these complaints follow near-identical wording and describe service grievances rather than transaction-level errors. The denominator inflated faster than the numerator, and the reconciliation share compressed even as the absolute count of reconciliation-related complaints more than doubled from 1,433 to 3,109.
The 2021 to 2024 trend, isolated from the 2025 event, is the cleaner signal. The share of fintech payment complaints that describe reconciliation patterns has been rising. The acceleration is concentrated in product categories where transaction complexity is highest: money transfer, digital wallet, prepaid card, checking, and installment loan products. These are the same areas where money moves across external banks, processors, wallets, ledgers, and internal records.
Where reconciliation-related complaints concentrate
Reconciliation-related complaints cluster in the highest-volume consumer platforms, where transaction complexity and the number of connected systems are greatest. The counts below reflect platform scale and complaint volume, not failure rates: the CFPB data does not provide transaction denominators, active user counts, or operational quality measures.
Block (Cash App) leads the raw count at 2,012, followed by Chime at 1,655 and PayPal (including Venmo) at 1,300. Early Warning Services, which operates Zelle, and Coinbase round out the top five. These are platforms that move money across many counterparties, multiple payment service providers, and external bank feeds. That environment is precisely where multi-source reconciliation can break down. The concentration of these complaints in money-transfer products is consistent with the reconciliation pattern rather than with fraud or access complaints, which distribute differently across the dataset.
Reconciliation complaints resolve with relief nearly three times more often than other fintech payment complaints
We compared CFPB closure outcomes for reconciliation-related complaints against the 89,980 non-reconciliation complaints in the same dataset.
| Closure outcome | Reconciliation-related | Non-reconciliation |
|---|---|---|
| Closed with explanation | 87.10% | 93.78% |
| Closed with monetary relief | 10.39% | 3.51% |
| Closed with non-monetary relief | 2.51% | 2.71% |
Explanation-only is the default CFPB closure for most complaint types, including fraud, and on its own says nothing about reconciliation quality. The comparison is what matters. Reconciliation-related complaints receive monetary relief at 10.39%, while the rest of the fintech payment complaints in this dataset receive relief at 3.51%. Reconciliation-related complaints are nearly three times more likely to end with the consumer being made whole.
Two reasonable interpretations of that gap, not mutually exclusive. The first is that reconciliation-related complaints describe verifiable transaction-level errors: a duplicate that can be matched against the original, a transfer with two conflicting records, a refund the merchant can document as sent. Once these reach a regulator, both sides can usually trace what happened. Fraud, authorization, and access disputes are typically much harder to adjudicate from documentary evidence alone. The second is that companies treat reconciliation disputes as operational errors they are obliged to correct, while categorizing fraud and access complaints as user-side problems.
The operational concern sits upstream of the CFPB. Every complaint in the recon-flagged 7,048 is one consumer who escalated. The larger and unmeasured population is the consumers whose discrepancies were absorbed quietly, the duplicates that closed without a refund, and the settlement breaks that became write-offs.
Why reconciliation failures are an operations problem, not a support problem
The instinct when complaints rise is to add support staff. The data argues against that. A reconciliation failure is not resolved by a better support reply. It is resolved by being able to trace a transaction across every system it touched and prove where the records diverged. When that capability exists, the dispute resolves with relief, as the data above shows. When it does not exist, the dispute either closes with an explanation or never reaches a regulator in the first place.
Payment reconciliation software exists to close that gap: to ingest transaction data from every source, match it into a single source of truth, and surface the discrepancies that would otherwise reach a consumer as a complaint. Two capabilities separate fintechs that resolve these issues quietly from those that let them reach a regulator. The first is exception management: when a transaction fails to match, it has to be identified, routed, investigated at the transaction level, and resolved before the consumer notices. The second is revenue leakage detection, which catches duplicate payments, settlement breaks, and unrecovered fees before they accumulate into write-offs. Both depend on transaction-level visibility and an audit trail.
This is the operating layer that consumer fintechs, neobanks, payment companies, and marketplaces increasingly need as they add payment service providers and scale transaction volume. It is the problem Rexi was built to solve: an agentic reconciliation platform that ingests transaction data from any source, reconciles it across fragmented systems, and investigates the exceptions before they reach a consumer. The rising share of reconciliation-related complaints from 2021 to 2024 is one measurable signal that the underlying gap is real and growing.
Methodology
We pulled every consumer complaint filed with the CFPB between January 1, 2021 and December 31, 2025 against a defined set of consumer fintechs: PayPal and Venmo, Block (Cash App), Chime, Wise, Zelle (Early Warning Services), Coinbase, Revolut, Klarna, Affirm, SoFi, and Robinhood. We limited the set to money transfer, digital wallet, prepaid card, checking, and consumer installment loan products, and to complaints that include a consumer narrative. That produced 97,028 complaints. Each complaint was classified by a large language model using the logic described in “How the classification works” above. Complaints whose root cause was fraud, unauthorized access, account suspension, fee disputes, or service quality were classified as not reconciliation-related, even when reconciliation-adjacent words appeared in the text.
Validation
To assess classifier accuracy, we hand-labeled a stratified sample of 80 complaints, drawn to include both classifier-positive and classifier-negative cases, and compared the human labels against the classifier output. The classifier reached 78.75% overall accuracy, with precision of 70.00%, recall of 94.59%, and F1 score of 80.46%. The classifier is high-recall and over-flags rather than under-flags: it captures nearly all genuine reconciliation-related complaints but applies the label to some complaints that on closer reading describe account-management or policy disputes. The reported 7.26% share is therefore best read as an upper bound, with a precision-adjusted estimate of approximately 5%. The labeled validation set is included with the audit data below.
The classification is a judgment about the most likely root cause, not a forensic determination of what happened inside any company’s ledger. It should be read as a classification of complaint patterns rather than confirmed internal system failures.
Limitations
This analysis is based on consumer complaint narratives and CFPB complaint metadata, not internal company ledger data. A complaint can show a pattern consistent with a reconciliation, settlement, recording, or matching breakdown, but the public CFPB record does not prove the exact internal cause.
Company-level counts are raw complaint counts. The CFPB dataset does not include each company’s transaction volume, active user base, or total payment flow. The company comparison should not be read as a ranking of operational quality, and no claim is made that any named company has unresolved reconciliation failures.
Audit the data
- Classified dataset (narratives removed): download the CSV. Includes Complaint ID, date received, company, product, year, and the reconciliation flag for all 97,028 records.
- Validation set: 80 hand-labeled complaints with classifier output, used to compute the agreement metrics in the methodology.
- Original source: CFPB Consumer Complaint Database. This pre-filtered query reproduces the exact base dataset of 97,028 complaints.
- Classification prompt: available on request.