Software warranties require a defect you can define
Offering a financial guarantee against fraud is a marketing decision until someone claims on it. At that point it becomes an underwriting problem, and the question is whether your product can tell whether it failed.
What makes a software warranty credible?
A failure condition that can be determined from artefacts rather than argued. “We guarantee you will not be breached” depends on the customer's configuration, their users and an adversary. “A payment released through our gate without a valid authorisation is covered” is a fact about the product.
- Warranties fail commercially when the failure condition is subjective, because every claim becomes a negotiation.
- A deterministic control gives a binary failure condition: the gate operated and produced evidence, or it did not.
- Reinsurance for such a warranty is available in proportion to how measurable the exposure is.
Part of Cyber insurance and risk transfer
Why most software warranties are hollow
They promise an outcome the vendor does not control. "We guarantee you will not be breached" depends on the customer's configuration, their users, their other systems and an adversary.
So the warranty comes with exclusions that swallow it, and everyone treats it as marketing. That is a reasonable reading of most such offers.
What a warrantable claim looks like
| Unwarrantable | Warrantable |
|---|---|
| You will not suffer fraud | A payment released through our gate without a valid authorisation is covered |
| Our product is secure | Our gate fails closed; if it fails open and a loss results, we cover it |
| We prevent account takeover | An action taken without a verifiable receipt, where our control was enabled, is covered |
The right column shares a property: the failure condition is a fact about the product's behaviour, determinable from artefacts, without arguing about the customer's overall security posture.
The three requirements
- A binary condition. The gate either produced a valid receipt for the action or it did not. No judgement.
- Evidence generated by default. Not on request, not reconstructible — produced as a by-product of the control operating.
- A bounded exposure. The vendor must be able to compute maximum loss per customer and in aggregate.
The third is what makes reinsurance possible. A reinsurer pricing a warranty programme needs the same thing an insurer needs anywhere: a defined exposure and a determinable trigger.
Computing the exposure
# What a vendor needs before offering the warranty
Per-transaction cap the warranty covers up to X per event
Per-customer aggregate capped annually
Portfolio aggregate capped, and reinsured above a retention
Trigger action executed without a valid receipt,
control enabled, no customer bypass
Exclusions customer disabled the control
customer's own credential compromise
loss outside the gated action setThe exclusions must be narrow and objective, or the warranty returns to being marketing. The first exclusion is fair and verifiable from configuration; the second is where most disputes would concentrate and needs careful drafting.
The commercial logic
A warranty is a signal, and a credible one is expensive to imitate. A competitor without a deterministic control cannot offer the same term, because they cannot determine whether they failed.
| Effect | Mechanism |
|---|---|
| Differentiation | Competitors cannot match without the same control architecture |
| Procurement acceleration | The customer's risk team has a concrete answer rather than a questionnaire |
| Pricing power | The warranty is a feature with quantifiable value |
| Internal discipline | A vendor exposed to its own failures builds differently |
The last row is understated and real. Engineering behaviour changes when a defect in a control costs the company money directly rather than costing a customer money indirectly.
Where this goes wrong
Two failure modes worth anticipating.
- Adverse selection. Customers with the worst fraud exposure adopt the warranty most enthusiastically. Underwrite per customer rather than offering a flat term.
- Moral hazard at the customer. A warranted control may reduce the customer's own vigilance. Structure it with a customer retention so the incentive remains shared.
Both are ordinary insurance problems with ordinary insurance answers. That is the point: once the trigger is determinable, this becomes a known discipline rather than a novel one.
This describes commercial structures at a general level and is not insurance or legal advice. Warranty drafting, regulatory treatment and reinsurance arrangements are specialist matters; take them to advisers.
What a reinsurer needs before pricing it
| Requirement | Why |
|---|---|
| A binary trigger | Determinable without judgement |
| Evidence generated by default | Not on request, not reconstructible |
| A bounded exposure | Per event, per customer, per portfolio |
The third is what makes reinsurance possible. Without a computable maximum loss the programme cannot be placed, whatever the trigger looks like.
Objections and honest limits
“Adverse selection will ruin the economics.” It is a real risk — the worst-exposed customers adopt most enthusiastically. Underwrite per customer rather than offering a flat term, which is ordinary insurance discipline.
“A warranty reduces the customer's own vigilance.” Moral hazard, also real, also standard. Structure it with a customer retention so the incentive stays shared.
Designing a credible warranty
- Define the trigger as a product behaviour. Not as a customer outcome.
- Cap per event, per customer and per portfolio. So the exposure is computable.
- Keep exclusions narrow and objective. Or it returns to being marketing.
- Underwrite per customer. Flat terms attract adverse selection.
Terms used here
- Trigger
- The condition under which a warranty pays — credible only when determinable from artefacts.
- Adverse selection
- The tendency for the worst-exposed parties to adopt a guarantee most readily.
- Retention
- The portion of a loss the customer bears, which keeps their incentive aligned.
Frequently asked questions
Why are most software warranties hollow? They promise outcomes the vendor does not control, so exclusions swallow the promise and everyone treats it as marketing.
What makes a claim warrantable? A failure condition that is a fact about the product's behaviour, determinable from artefacts, without arguing about the customer's overall posture.
Why does this need a deterministic control? Reinsurance requires a determinable trigger and a bounded exposure. A probabilistic control gives neither.
What goes wrong commercially? Adverse selection — the worst-exposed customers adopt first — and moral hazard. Both have standard insurance answers: per-customer underwriting and a customer retention.
What does a reinsurer need? A determinable trigger and a computable maximum exposure per event, per customer and per portfolio.
What is the unexpected benefit to the vendor? Engineering behaviour changes when a control defect costs the company money directly rather than costing a customer indirectly.
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
- Federal Acquisition Regulation
- Insurance market commentary on technology errors and omissions cover.
- Reinsurance structuring guidance for affinity and embedded programmes.
- NAIC — cyber risk resources