Tolerances and Service Level Agreements

An impact tolerance and a service level agreement are different kinds of commitment, and confusing them is a costly error. A tolerance is a harm limit set by the firm for its own service: the point beyond which disruption causes intolerable harm to customers or to the wider system. A service level agreement is a contractual commitment from a supplier about the performance of a supplied component, such as availability or response, usually tied to commercial remedies. A supplier can meet its agreement while the firm's service still breaches its tolerance, because the agreement measures the component, while the tolerance measures harm to users of the whole service.

The illustration here is a lending platform that depends on a hosted infrastructure provider. The provider agreement may describe availability of the hosting platform in general terms, while the lending service depends on that platform plus application software, data, payment connections, and customer channels. If the hosting platform performs as agreed but another component fails, or if the platform degrades in a way the agreement does not capture, customers of the lending service can still suffer intolerable harm. The firm remains responsible for the tolerance, even where a supplier is responsible for a component.

The difference begins with perspective. A tolerance is written from the perspective of harm to users of the service. An agreement is written from the perspective of a supplier relationship, describing what the supplier will provide and what happens commercially if it does not. Those perspectives can overlap, but they are not the same, and an agreement that is strong in commercial terms can still be weak in harm terms if it measures the wrong thing, measures it over the wrong period, or excludes the degraded states that actually harm users.

Meeting an agreement can still mean breaching a tolerance for several reasons. The agreement may cover only part of the service chain, leaving other components unmeasured. It may measure availability in aggregate, hiding harm to a specific user group or harm from inaccurate data. It may allow for maintenance or for degraded performance that, in the context of the firm's service, still causes intolerable harm. It may also measure performance over a period that smooths out a severe but brief disruption, so that the period average looks acceptable while users at the moment of disruption suffered seriously.

Alignment begins by mapping the tolerance to the components that support the service. For each component, the firm can ask what performance is needed from that component for the whole service to remain inside its tolerance, and whether the existing agreement requires that performance. Where the agreement falls short, the firm faces a choice: negotiate a stronger agreement, add monitoring or workaround arrangements, change the service design to reduce dependence, or accept and record the gap with eyes open. Each option is legitimate if it is chosen consciously and recorded, while an unexamined gap is not.

Alignment goes wrong when the firm copies agreement language into its tolerance, or tolerance language into its agreement, without translating between harm and contract. Agreement language about availability percentages and measurement periods does not, on its own, describe intolerable harm, and tolerance language about harm does not, on its own, create an enforceable supplier obligation. Each document needs its own language, with a clear mapping between them that explains how supplier performance supports, or fails to support, the harm limit.

A reader can test alignment by asking, for each supplier that supports the service, what the agreement actually measures, whether that measure relates to the harm limit, and what happens when the agreement is met but the service still fails. Further questions concern monitoring and communication: can the firm see supplier degradation in harm terms, and can it inform affected customers promptly? If those questions have no answer, the alignment is incomplete, however strong the individual agreements may appear. A mapping that is reviewed when agreements change, and not only when they are signed, keeps the link between contract performance and harm visible over time.