Procedure clauses are read last, when they are read at all. That is an error of sequence, because they do not merely decide how a dispute is resolved: they decide the amount, exactly as a definition of cover does. Who decides, under what rules, with access to which documents, and what may be retained by way of offset: each of these questions moves the outcome by millions without touching a single line of cover. Five clauses make up this ground, and the most important is not the one naming the judge, it is the one giving access to the files.
The arbitration clause refers disputes to an arbitral tribunal, usually of three members two of whom are party-appointed, and traditionally requires the arbitrators to be practitioners of insurance and reinsurance, serving or retired. Many wordings invite the tribunal to decide having regard to market custom rather than to the letter of the contract alone. What it solves is real: a strictly literal reading by a non-specialist court produces outcomes neither cedant nor reinsurer had contemplated. Arbitration also brings confidentiality, valuable where the dispute concerns a cedant's underwriting quality.
The defect of arbitration is the one least often cited and it is the heaviest: no award makes precedent. Costs are high, appeal is close to non-existent, but above all the same questions are relitigated indefinitely from one treaty to the next. A cyber aggregation dispute can be decided in eleven months, for 1.4 million euros of costs, by a tribunal of practitioners holding for aggregation by vulnerability exploited rather than by insured affected. Two years later a different tribunal holds the opposite on neighboring wording, and both awards stand. An underwriter therefore cannot rely on an earlier decision to know what its own wording will produce: it has only the text, which points back to wording discipline rather than to any hope of clarification through litigation.
The access to records clause is the instrument that gives practical reality to contractual good faith, and it is the one that pays best. It allows the reinsurer to inspect, at the cedant's premises, the documents relating to ceded risks and notified losses: underwriting files, adjusters' reports, correspondence, accounting records. It answers the structural informational asymmetry of reinsurance, where the reinsurer underwrites on information it does not produce and handles losses it never sees. Without a right of access it has no means of detecting a drift in underwriting, a claims classification artificially inflating an event, or a questionable allocation between two layers. The friction lies less on the principle than on its limits: legal privilege, policyholder data protection.
The errors and omissions clause deals with the reverse problem, the disproportion between the fault and the sanction. A treaty covering thousands of policies is administered by systems and by people: error there is statistically certain, and destroying cover worth millions over one mistyped line would turn reinsurance into an administrative lottery. The clause therefore provides that a good faith error in administering the treaty does not deprive the cedant of its cover, provided it corrects on discovery. It never covers bad faith, however, nor anti-selection dressed up as error, nor an omission that would have changed the reinsurer's decision. The whole difficulty sits there, on the boundary between the protected clerical error and culpable non-disclosure.
The offset clause allows each party to settle only the net balance of sums mutually due, premiums against losses, and often across treaties where the drafting permits. It simplifies cash flow in normal times and becomes decisive on default, where it functions as security without collateral. A simple case shows its reach. Take a reinsurer owing 14 million euros of losses to a cedant in liquidation which owes it 9 million of premiums across three separate treaties: with transversal offset recognized by the applicable law, it pays 5 million; without it, or where the law confines offset to connected debts under the loss-bearing treaty alone, it pays 14 million and recovers its 9 million at the liquidation's distribution rate.
That leaves the service of suit clause, American in origin, by which the reinsurer agrees to submit to a designated jurisdiction and appoints there an agent authorized to accept service. It answers a very concrete difficulty: getting a reinsurer established abroad to appear, then enforcing a decision against it, can take years and cost more than the receivable. American regulators frequently require it as a condition of reinsurance credit. Its interaction with arbitration demands rigor: well drafted, the two coexist, arbitration settling the merits and service of suit serving to compel arbitration and then to have the award recognized; badly drafted, they cancel each other and open a preliminary dispute on forum, before the merits have been reached at all.
A reinsurer exercises in 2026 its access to records right on a cyber treaty whose loss ratio has moved from 62 to 141 percent over two years. The program carries a per-risk attachment and a per-event layer. Inspection of 180 files reveals that 23 losses arising from a single supply-chain event were notified separately, under the per-risk attachment. What does regrouping them change, and what has this clause actually produced?
Regrouping changes which layer answers, and with it the amount owed. Notified separately, each of the 23 losses is measured against the per-risk attachment: the smaller ones stay wholly with the cedant, the larger ones exceed the attachment only slightly, and the total recovery is modest. Regrouped as a single event, they add into one charge that crosses the per-event attachment and enters the corresponding layer. The account is corrected by 4.6 million euros. It should be seen that the correction can run either way and that the inspection did not predetermine it: regrouping may equally reveal a recovery due to the cedant or one wrongly collected, depending on the program structure and the size of the individual losses. What the clause has produced can then be stated precisely. It created no new right: the 23 losses fell within the per-event layer from day one, and the definition of event did not change. It merely made visible a classification that was not, and without it the reinsurer would have had only aggregates supplied by the cedant, in which 23 separate lines are indistinguishable from ordinary attritional frequency. The ratio moving from 62 to 141 percent told it something was happening; it did not tell it what, and no ratio ever does. Two practical consequences follow: the clause is read before renewal rather than after the loss, and its survival after a commutation must be expressly settled in the wording, failing which the right of access dies with the relationship it existed to verify.
- 01Procedure clauses decide the amount as much as a definition of cover does: who decides, with access to which documents, and what may be offset.
- 02No arbitral award makes precedent: the same questions are relitigated from treaty to treaty, and two tribunals can hold the opposite on neighboring wordings.
- 03The access to records clause creates no new right, it makes visible a classification that was not, and a deteriorating ratio never says what happened.
- 04The errors and omissions clause protects the good faith clerical error, never bad faith nor anti-selection dressed up as error: the whole litigation sits on that boundary.
- 05The offset clause works as security without collateral, and its reach depends on insolvency law, which may confine it to connected debts under a single treaty.
- 06Service of suit and arbitration coexist when well drafted, and cancel each other when badly drafted, opening a dispute on forum before the merits are reached.