An Umbrella Layer Is Not a Higher General Liability Limit

An umbrella limit looks very tidy in our policy system. It is a number above the general-liability limit, with a carrier name and a policy number beside it. Add the rows and the account appears to have a total available limit. Exactly the view our reporting likes. It answers a question nobody asked.

An umbrella layer is not a higher general liability limit. It is a conditional promise whose shape depends on everything beneath it.

The corpus gives us a four-layer placement. A $2M primary CGL with a $4M general aggregate. A $10M umbrella attaching at $2M. A $15M following-form excess above $12M. A $25M following-form excess above $27M. The umbrella also carries a $25,000 self-insured retention, for a claim it covers that no underlying policy does. A screen reporting $52M made the tower look simpler by deleting the facts that decide whether the fourth row ever pays.

Those facts are not edge cases.

They are the tower we sold.

We need to know what the CGL promises.

What it has paid.

What the umbrella schedules.

How each layer treats defence.

Which wording each excess layer follows.

And which condition turns a loss from a watching brief into an obligation.

A total-limit field cannot carry an argument.

The primary is not merely the first number in the stack

We start at the bottom.

The Canadian CGL reference point is IBC 2100, and its US cousin is ISO CG 00 01.

Neither identifier substitutes for the placement’s endorsement stack.

Both make clear what the primary layer is trying to answer. Coverage A takes bodily injury and property damage caused by an occurrence.

Coverage B takes personal injury and advertising injury.

Coverage C handles medical payments without regard to fault. Canadian bureau forms put tenants’ legal liability in its own section. That structure matters, because an excess policy doesn’t sit above an abstract category called liability. It sits above a contract with an insuring agreement, definitions, exclusions, exceptions, limits, and conditions. The primary CGL is not a warranty on the insured’s work. Its business-risk exclusions separate the cost of redoing that work from the damage the work does to other property. A products-completed operations loss means finished work or products away from premises the insured owns or rents. It carries its own aggregate where the declarations provide one.

Take the water-loss sequence in the corpus.

A coupling installed in May 2023 fails in March 2026. On occurrence-based Coverage A, the relevant policy is the one in force when the damage happens. Not the one in force when the work was done. Not the one in force when a pleading arrives. So a completed-operations requirement two years past substantial performance is an operating control.

It is not a contractor’s administrative preference.

The tower record holds bodily injury, property damage, and products-completed operations separately.

It also has to carry the trigger.

A claims-made underlying paired with an occurrence-based excess, on a different retroactive date, is not a metadata inconsistency.

It is a gap across time.

The word “liability” does not repair it.

The Canadian decisions show where this comes from.

Progressive Homes Ltd. v. Lombard General Insurance Co. treats defective construction as capable of being an occurrence, and applies the pleadings-based duty to defend. Alie v. Bertrand & Frère Construction is cited regularly on allocating progressive damage across successive policies. Neither tells us to guess a result from a case name. Both remind us that date, wording, damage, and layer connect.

Attachment is stated, and exhaustion is discovered

Here is the distinction I would put at the centre of the data model.

Primary exhaustion records what has happened.

Excess attachment records what the layer says must happen first.

Related. Not one field.

The schedule of underlying insurance names the policies, insurers, and limits beneath a layer.

It defines attachment by its stated figures.

Not by what the insured ultimately bought, and not by what the insured recovered. Replace a scheduled $2M primary with half that mid-term, leave the umbrella schedule alone, and the umbrella still attaches at $2M. The maintenance condition makes the missing million the insured’s problem.

That isn’t a drafting curiosity.

In the four-layer example, two unrelated claims consume $2.8M of the general aggregate.

Only $1.2M remains for the fire.

The umbrella still attaches at $2M under maintenance.

That leaves an $800,000 gap the insured carries.

After that contribution the umbrella pays its limit and the first excess pays $4.2M.

The second excess never attaches at all.

The reported tower did not get smaller.

The usable path through it did.

Some umbrellas use a reduced-aggregate drop-down instead.

They respond above the reduced underlying aggregate, and can continue as underlying insurance after it is exhausted. That difference can’t live in a boolean called “umbrella.” It is a specific conditional relationship between a primary aggregate, a scheduled attachment, and the umbrella’s own grant.

Insolvency is where our screens turn dangerously optimistic.

Modern umbrella and excess wordings ordinarily say that insolvency of an underlying insurer does not alter attachment.

The layer applies as if the underlying were still in force.

PACICC may matter to the failed insurer here.

It doesn’t turn a $10M umbrella into a first-dollar CGL.

A system reporting it that way manufactures coverage with arithmetic.

Defence costs decide when the tower is real

The question isn’t how much loss has been reported.

It is what kind of dollar moved.

The standard Canadian CGL treats defence costs as supplementary payments, in addition to the limit.

In the corpus fire, $2.1M on counsel, origin-and-cause expertise, and an accounting review doesn’t erode the CGL attachment.

Indemnity does.

Early contents settlements of $340,000 do.

Occurrence count matters too: the amount hits the per-occurrence limit if the fire is one occurrence, and the aggregate regardless.

Reverse that convention and the file changes shape.

Defence-inside-limits treatment is common in professional liability, in directors-and-officers, and in many environmental placements.

A layer like that burns its limit while counsel works.

An umbrella with defence inside its limit isn’t equivalent to one with defence outside.

Both display $10M.

This is a poor place to inherit a field from a US template and assume our primary behaves the same.

IBC 2100 is only the reference point.

The endorsement stack and the excess wording govern the placement.

So we store defence treatment per layer, with its source form and edition.

Otherwise a reserve dashboard describes a tower as intact while its usable limit is already being spent.

Following form is a relationship, not a label on a row

We use “umbrella” as a synonym for excess.

We shouldn’t.

A true umbrella is standalone.

It has its own insuring agreement, definitions, and exclusions, and it can be broader than the CGL.

Where it covers something the CGL excludes, it responds above its retention.

In the example that retention is $25,000.

A following-form excess layer cannot make that move.

It carries no coverage grant of its own.

And following form does not mean identical.

A follow-form clause adopts the underlying terms, except premium, limits, and whatever the excess policy otherwise provides.

The layer’s own asbestos, abuse, communicable-disease, or cyber exclusion can displace the underlying grant.

A narrower definition of insured, a tighter territory, an occurrence-based excess above a claims-made underlying: each changes the promise.

A second relationship needs our attention.

Does the layer follow the first underlying policy, or the policy immediately below it?

Under the second convention, a narrowing at layer 2 propagates all the way up.

Under the first, layer 3 can be broader than layer 2 and still fail, because layer 2 denies and layer 3’s attachment never happens.

Non-concurrency is normal here.

It isn’t evidence that somebody failed to tidy up.

It gets worse when one apparent layer is a quota share.

The corpus describes a $25M layer written across three carriers in unequal shares.

That is three contracts.

If they do not follow form the same way, we do not have one layer with three participants.

We have three coverage relationships sharing a position.

A total-limit calculation has nowhere to put that, so it pretends otherwise.

Exhaustion, notice, and consent are three separate controls

An attachment point alone tells us nothing about exhaustion.

Modern excess wording can require payment of the full underlying limit.

Sometimes payment by the underlying insurers themselves.

So a primary settlement below limit strands the excess layers, even where the insured will fund the difference.

Zeig v. Massachusetts Bonding and Qualcomm v. Certain Underwriters at Lloyd’s mark the competing lines of authority.

The controlling answer is the wording in that layer.

Notice does not travel up the tower either.

Each layer sets its own condition on us.

Excess notice is required when an occurrence is reasonably likely to reach that layer.

Brokers notify each layer when carried reserves reach half the attachment beneath it.

Fatalities, brain or spinal injuries, multiple claimants, class proceedings: reportable at any amount.

So a primary notice date should not populate every excess notice field by default.

Consent is a third record we keep.

A settlement reaching the umbrella can require its written consent, and a following-form layer can inherit that condition.

In the fire example, mediation produces demands of $19.4M and a mediator’s number of $16.2M.

The decision isn’t whether the total sits below the tower.

It is whether the primary can tender.

Whether the umbrella consents.

Whether the following-form excess consents.

And whether the exhaustion condition has been satisfied.

A system reducing that to “claim amount versus limit” has automated the wrong question.

After the primary is exhausted, what happens next?

The board sees a diagram of four stacked rectangles.

We need the question the diagram avoids.

Once the primary is gone, can the system say exactly what happens next?

The answer we need is modest enough.

Per layer, keep the policy period, agreement type, attachment, limit, and aggregate.

The schedule of underlying insurance. Defence treatment. Retained limit. Exhaustion wording. Maintenance condition. Follow-form reference point. Its own exclusions. Notice and consent conditions. Its drop-down treatment.

Then link them to the CGL’s Coverage A trigger, its products-completed operations aggregate, and the endorsements.

Including any additional-insured and primary-and-non-contributory arrangement.

OSFI supervises solvency and capital federally.

Provincial regulators supervise licensing and market conduct.

None makes layer 4 concurrent with layer 3.

None tests how we applied an attachment condition to a loss.

Commercial liability wording isn’t subject to prior approval the way automobile forms are.

The contract record has to carry the discipline no filing process will supply.

AI compares schedules, spots an absent endorsement, flags a primary limit that no longer matches an umbrella schedule.

That is useful policy archaeology.

It should not determine that a layer follows form, that defence is outside the limit, or that a settlement has exhausted underlying insurance.

“The assistant added the limits” reads badly at mediation.

The tower is reliable only while each layer still knows why it is there.