Financing agent risk through a captive when the market will not write it
When no market will write a coverage at a sensible price, a large enterprise insures itself. That is a familiar mechanism, and it moves the question from what will an underwriter accept to what will our own actuary accept.
How do you finance agent risk when nobody will write it?
You retain it, and then you own the loss control problem. Commercial insurers price unbounded autonomous-system liability out of reach or decline it, because the exposure is undefined, there is no loss history, and attribution between agent, operator and third party is unsettled.
- A captive converts an external underwriting problem into an internal one, with the same need for verifiable loss control.
- Actuarial credibility requires exposure definition and loss data; for agent risk the exposure is poorly bounded and the history is short.
- Instrumenting authorisation gives the captive both a loss control and the data to price the retained layer.
Part of Cyber insurance and risk transfer
Why the commercial market hesitates
Three reasons, all reasonable from an underwriter's position.
| Reason | Detail |
|---|---|
| Undefined exposure | What an autonomous system can do is bounded by its credentials, not by a policy schedule |
| No loss history | The technology is too new for credible frequency and severity estimates |
| Aggregation risk | A defect in a widely used component could produce correlated losses across many insureds |
| Attribution difficulty | Determining whether a loss was caused by the agent, the operator, or a third party is unsettled |
The last row is the one that makes wordings hard to draft. An insurer needs to know what triggers the cover, and 'the agent did something wrong' is not yet a definable trigger.
What a captive changes and what it does not
It changes who bears the risk and who prices it. It does not change the underlying exposure or the need to control it.
In fact the discipline is often greater. A captive must satisfy a regulator, an auditor and its own actuary, and those parties ask the same questions a commercial underwriter would — with the difference that the answers determine your own capital requirement.
The actuary's problem
Pricing a retained layer needs an exposure base and loss data. For agent risk both are thin.
# What an actuary needs, and what is usually available
Exposure base ? number of agents — varies by definition
? actions per period — rarely measured
? value at risk per action — usually unbounded
Frequency ? little internal history
Severity ? little internal history
Control credit ? no evidence that controls operatedInstrumenting authorisation fills three of these directly. Actions per period becomes countable, value at risk per action becomes bounded by the delegation scope, and control operation becomes measurable.
Bounding the exposure with delegation
The single most useful thing an enterprise can do for its own risk financing is to make the agent's maximum exposure a configured number rather than an open question.
- Per-action value ceiling in the delegation
- Aggregate ceiling per period
- Counterparty and action-class restrictions
- Expiry, so a stale delegation does not persist
- Revocation that is enforced at the point of effect
With those, the theoretical maximum loss from a compromised agent is computable. An actuary can work with a computable number; they cannot work with 'whatever it can reach'.
Loss control that produces its own evidence
A captive's loss control programme has to be demonstrable to its auditor, the same way a commercial insurer's warranty has to be demonstrable at claim.
| Captive requirement | What the instrumentation provides |
|---|---|
| Documented loss control | Defined scope and gates, with measured coverage |
| Evidence it operates | Receipts per in-scope action |
| Exposure measurement | Countable actions with bounded values |
| Claims adjudication | Per-action evidence of what was authorised |
| Reinsurance or excess placement | Evidence to present to the commercial market |
The last row is where this eventually pays. A captive that has accumulated three years of instrumented loss data is in a position to buy excess cover commercially, because it can present the one thing the market lacks.
The realistic path
- Retain the risk in the captive, because there is no alternative at a sensible price.
- Bound it with enforced delegation scopes, so the maximum loss is a number.
- Instrument authorisation on the consequential subset, producing per-action evidence.
- Accumulate loss and near-miss data with the control's coverage rate alongside it.
- Approach the commercial market for excess layers with evidence rather than assertions.
Step five is a multi-year proposition and it is how this category eventually becomes insurable. The enterprises that instrument first will be the ones that can buy cover first, which is a commercial advantage that has nothing to do with security.
This describes risk financing structures at a general level and is not insurance, tax or legal advice. Captive formation and regulation are jurisdiction-specific; take specifics to your advisers.
Bounding the exposure so it can be priced
| Instrumented | Actuarial value |
|---|---|
| Actions per period | An exposure base that did not exist |
| Per-action value ceiling | A computable maximum loss |
| Aggregate ceiling per period | A bound on correlated loss |
| Control coverage rate | A credit that can be justified |
Without these, the theoretical maximum loss from a compromised agent is ‘whatever it can reach’, which no actuary can work with. With them it is a number.
Objections and honest limits
“A captive is just moving money between pockets.” It is, and the discipline is often greater: a regulator, an auditor and your own actuary ask the same questions a commercial underwriter would, and the answers set your capital requirement.
“This is a multi-year play.” It is. The enterprises instrumenting now are the ones that will be able to buy excess cover commercially first, because they will hold the loss data the market lacks.
Financing agent risk
- Retain it — there is no alternative at a sensible price. Start from that honestly.
- Bound it with enforced delegation scopes. So maximum loss is computable.
- Instrument authorisation on the consequential subset. Producing per-action evidence.
- Accumulate loss and near-miss data. With coverage rate alongside it.
Terms used here
- Captive
- An insurer owned by the organisation it insures, used where the commercial market will not write a risk sensibly.
- Exposure base
- The measurable quantity a rate is applied to — which for agent risk does not currently exist.
- Excess layer
- Cover above a retention, which becomes purchasable once loss data exists.
Frequently asked questions
Why won't commercial insurers write agent liability? Exposure is undefined, loss history is short, aggregation risk is real, and attribution between agent, operator and third party is unsettled — which makes triggers hard to draft.
Does a captive reduce the risk? No. It changes who bears and prices it. The discipline is often greater because the regulator, auditor and actuary ask the same questions.
What does the actuary need? An exposure base and loss data. Instrumented authorisation makes actions countable and bounds the value at risk per action, which is what turns this into a priceable layer.
How does this eventually reach the commercial market? Accumulated instrumented loss data is the thing the market lacks. An enterprise with three years of it can present evidence rather than assertions for excess layers.
Where this fits in Manav
Manav produces the artefact underwriting, claims and forensics all lack: a per-action receipt verifiable without the insured's cooperation, and a measurable coverage rate.
Sources and further reading
- NAIC — captive insurance and risk retention groups
- Actuarial literature on pricing emerging risks with limited loss data.
- Insurance market commentary on autonomous system and AI liability coverage.
- NAIC — cyber risk resources