The intuition is close to universal and it is wrong. More banking relationships feel safer, in the way that more exits feel safer, and for the same reason: the cost of each additional one is assumed to be near zero.
It is not near zero. An account is a standing obligation, and beyond a small number the marginal relationship consumes more than it returns.
What a relationship actually costs
The visible price of an account is the fee schedule. The real price is administrative and it recurs.
- Onboarding. Corporate documents, ownership charts, certified identification for every beneficial owner and signatory, source of funds, business rationale, and answers to questions that arrive in instalments.
- Periodic review. Every institution re-papers its clients on a cycle. Each cycle re-requests documents the institution already holds, because the request is generated by a system rather than by a person.
- Signatory management. Every change of authorised person, in either direction, must be effected separately at each institution, with its own forms and its own timetable.
- Reconciliation. One more statement, one more format, one more set of value dates and fee postings to be matched.
- Reporting. Additional lines in group reporting, and in several regimes an additional disclosure obligation.
- Attention. The scarcest input. Someone senior must hold a view on whether each relationship is healthy, and attention divided across seven relationships is thinner than attention across three.
None of these costs appears on an invoice, which is precisely why the marginal account is approved so readily.
Compliance surface
Each relationship is a separate window through which the group is examined, and each examiner reaches its own conclusions using its own risk model.
That matters more than it used to. A relationship at an institution that later withdraws from the client's sector, or from a jurisdiction in the group's ownership chain, does not simply end quietly. It ends with an exit that other institutions can see, and questions about it recur in every subsequent onboarding. A structure with seven relationships has seven opportunities to acquire that history. A structure with three has three.
Compliance surface scales with the number of relationships. Capability does not.
There is a second-order effect. Where documentation is spread across many institutions, inconsistencies accumulate: an ownership percentage stated differently, a registered address updated in one place and not another, an outdated director listed on one mandate. Any single inconsistency is trivial. Discovered during onboarding at a new institution, the pattern reads as carelessness at best.
The dormant account problem
Dormant relationships are the worst category, because they carry the full compliance and administrative load while delivering nothing at all.
They persist for predictable reasons. The residual balance is too small to justify attention. Nobody is certain whether a standing instruction still points at the account. The entity that owns it may be needed again. Closure requires a board resolution that nobody wants to draft.
So the account remains, generating a periodic review every eighteen months until somebody eventually closes it under time pressure during an unrelated transaction, at which point it is discovered that a counterparty has been paying into it for two years.
The test
One question resolves most cases. What would break if this relationship were closed at the end of next month?
Acceptable answers are specific. Payroll in a named market. Receipts from three named customers who cannot pay elsewhere without renegotiating contracts. Guarantees issued by this institution and accepted by this counterparty. Custody of securities that cannot be transferred without a tax event. A settlement route that no other relationship in the structure reaches.
Unacceptable answers are general. Redundancy. Flexibility. It is useful to have options. Our auditor likes it. We might need it. Each of those is a statement that the relationship has no function, phrased so as to avoid the conclusion.
Pruning method
- List every relationship with owning entity, currencies, average balance and stated function.
- Apply the closure test to each. Record the answer in one sentence.
- Group the survivors by function. Where two occupy the same function in the same jurisdiction with the same correspondent route, one of them is redundant.
- Before closing anything, trace inbound instructions: standing orders, direct debits, counterparty payment files, card and merchant settlement, tax and payroll authorities.
- Close in sequence, not in parallel, and leave each account open and monitored for one full reporting cycle after the last expected receipt.
How many is right
There is no universal number, but there is a rule for deriving one. Count the functions the structure genuinely performs, then add redundancy only where a failure would be material and no other relationship in the structure could absorb it.
For many internationally active mid-sized groups that produces a low single-digit count of primary relationships, with one deliberate second route for cross-border settlement. Larger groups with genuine multi-market operations carry more, and they carry them because each is doing identifiable work, not because the number grew.
The failure mode to avoid is symmetrical. Too many relationships is expensive and opaque. One relationship is a single point of failure that has been rationalised as simplicity. Neither is a structure.