BaaS, or banking as a service, separates the entity holding the licence from the entity holding the record of who owns what. The Synapse Financial Technologies bankruptcy tested that separation in front of a court. The trustee's report then put a price on the failure. I think the kind of number matters more than its size, because it is not a loss on a loan book, which is how most people picture a failed bank partnership, but the gap between cash sitting in pooled accounts and the balances a ledger said should be there.
Deposit-taking through a partner bank is still the standard route for a US neobank. The Synapse file is the only complete public record of a middleware failure. Anyone doing diligence on a BaaS target in 2026 meets the questions the trustee could not answer quickly. Who holds the funds, and who holds the ledger? Do the two agree on any given day?
Synapse filed for bankruptcy in late April 2024. A Chapter 11 trustee was appointed on 24 May 2024. The trustee's report in Case 1:24-bk-10646-MB, docket 287, was filed on 20 June 2024, and it sets out the reconciliation position in more detail than I expected from a filing made less than a month after the appointment.
The first estimate was larger and simpler. The report records "an estimated shortfall of $85 million between the cash held in FBO Accounts among all Partner Banks and the Synapse trial balance ". That figure was later refined to $65m to $96m. The trustee calls it the position "according to reconciliation progress to date".
The structural cause is stated plainly: "identifying which end-users' funds are held in FBO Accounts is complicated by the structure of those accounts". A pooled custodial account has one name at the bank. Its many owners sit on a ledger held somewhere else. When the ledger operator stopped, the bank was left holding cash it could not allocate.
So I read the ledger as the asset that matters here. The licence is closer to something rented. A partner bank that cannot reconstruct beneficial ownership without the middleware is a single point of failure, whatever its capital position.
The candour of the report surprised me. It says that "Lineage, the partner banks, and the Trustee presently lack a clear understanding of the exact amount or the origin of this cash shortfall". It adds that "Synapse's final trial balance reflects millions of missing funds in the Synapse ecosystem".
There is one passage I find hard to read past. According to the trustee, "one partner bank's distribution of over $50 million primarily to one fintech's end users prior to the appointment of the Chapter 11 trustee complicates any attempt to fairly allocate the available funds among the affected parties".
For an acquirer, the file stops being history here. Reconciliation evidence belongs in diligence next to the loan tape. A target that cannot produce a same-day match between bank cash and ledger balances is carrying the Synapse exposure, whatever its deck says about compliance.
My arithmetic asks two plain things. What sets the width of the trustee's range, and how much of the gap has public enforcement money since filled? The trustee report, Case 1:24-bk-10646-MB, Doc 287, gives three inputs: a low estimate of the FBO shortfall of $65m, a high estimate of $96m, and approximately $31m of unattributed interest and rebate funds at partner banks. I took the other two from the CFPB's Civil Penalty Fund allocations page, where $46,248,291 is recorded as allocated on 28 November 2025 and $8,965,767 on 29 May 2026.
Take $65m from $96m and the range is $31m wide. The trustee attributes it to "approximately $31 million in Synapse-related funds from interest and rebate payments" whose ownership was undetermined. So the uncertainty in the shortfall is not a modelling band, and I think that is the most useful line in the report. It is one specific unreconciled pool, and it accounts for the spread exactly.
A sponsor bank should read an unattributed balance like that as a reconciliation defect. It is not a rounding item.
I start from the midpoint of the range, $80.5m. The two CFPB allocations add up to $55,214,058. That is 68.6% of the midpoint. Against the low end, coverage reaches 84.9%. Against the high end it drops to 57.5%. At the midpoint, $25,285,942 of the shortfall stays uncovered.
The coverage ratio swings 27 percentage points across the trustee's own range, and the whole of that swing comes from the $31m attribution question, the same unreconciled pool of interest and rebates, rather than from anything the CFPB decided about who gets paid.
These are my figures, worked for M&A House. They carry limits I cannot remove. Neither the trustee nor the CFPB has published a coverage percentage. I have assumed the allocations are applied to the FBO shortfall, which the allocation notice does not state. I have also ignored recoveries from the estate and from partner bank distributions, which would cut the uncovered figure by an amount that is not yet public.
The money that reached Synapse end users came from none of the parties. It did not come from Synapse, from its partner banks, or from its shareholders. It came from the Civil Penalty Fund, which is funded by penalties paid by unrelated firms in unrelated cases, and the same CFPB page lists allocations to victims of Key Credit Repair, DMB Financial, Integrity Advance and others.
So the loss was socialised across an enforcement pool. The parties to the arrangement did not bear it. Any incentive argument that assumes a middleware failure is paid for by the participants does not survive that fact.
Enforcement followed the licence, not the software. On 14 June 2024 the Federal Reserve moved against Evolve Bancorp and Evolve Bank and Trust over "deficiencies in the bank's anti-money laundering, risk management, and consumer compliance programs". It found that "Evolve engaged in unsafe and unsound banking practices by failing to have in place an effective risk management framework for those partnerships".
It also found that "Evolve did not maintain an effective risk management program or controls sufficient to comply with anti-money laundering laws and laws protecting consumers". Examiners assess the partnership, not the partner. Evolve was sanctioned for the framework around its fintech programmes, not for any single programme. Related reading on this site covers AML obligations and regtech tooling. The M&A deals base records comparable transactions.
Greg Buchak, Gregor Matvos, Tomasz Piskorski and Amit Seru published "Fintech, regulatory arbitrage, and the rise of shadow banks" in the Journal of Financial Economics in 2018, volume 130, issue 3, pages 453 to 483. The Crossref record for it checks out. The paper asks how much of the shift to non-bank providers comes from regulation rather than technology.
The authors run a set of difference-in-difference tests, and I think the design is the strong part: the tests exploit geographic variation created by four specific increases in regulatory burden, so the effect of regulation can be told apart from the effect of technology. The finding is that "traditional banks contracted in markets where they faced more regulatory constraints; shadow banks partially filled these gaps". Shadow bank share of residential mortgage origination nearly doubled between 2007 and 2015.
The pricing result is the one I would carry into any bank partnership model. Fintech lenders in the sample charged a premium of 14 to 16 basis points and "appear to provide convenience rather than cost savings to borrowers". A model built on convenience, with a cost base that leaves out prudential supervision, will attract that supervision eventually. Synapse shows it arriving through a court rather than through an examination.
The FDIC published "Recordkeeping for Custodial Accounts" on 2 October 2024 as a notice of proposed rulemaking. It extended the comment period on 20 November 2024. Nothing further has appeared under that title. A proposal, however detailed, is not yet a rule, and a fintech that has been pricing compliance cost against an assumed final rule has been pricing against a document that binds nobody.
Deposit insurance attaches to the account as recorded at the bank. I would weigh pass-through recordkeeping above headline coverage limits for that reason. Almost two years after publication, the rule meant to stop a repeat of Synapse is still a draft, and I still cannot say who would carry a $25,285,942 gap if the next ledger went dark tomorrow.