Writing service levels that can actually be measured

Every measure needs a population, a window and a source

Take despatch performance. The population might be all orders, clean orders, or clean orders received before a cut-off. The window might be a day, a week or a rolling period. The source might be the provider's system, the client's, or a carrier scan. Change any one of those and the same operation produces a different figure. Settle all three per measure, and state which system is authoritative when the numbers differ, because they will.

Exclusions decide most disputes

Events outside the provider's control are normally carved out: severe weather, industrial action, utility failure, client system unavailability, stock that was never delivered, and customer sites that refuse access. The list should be finite, since an open-ended exclusion for anything the provider considers beyond its control removes the commitment entirely. Agree also the notification duty attached to invoking an exclusion, so a claim cannot be made retrospectively at the review meeting.

Remedies, and what they are actually for

Service credits rarely compensate for the commercial damage of a poor month, and they are not meant to. Their function is to make failure visible and to trigger structured attention: a corrective action plan, an escalation to more senior management, and, after repeated or serious failure, a right to terminate. Designing remedies as a ladder is more effective than pressing for a large credit that a provider will price into its rates anyway.

Measure the client too

Most provider failures have a client-side cause somewhere in them: a forecast that never arrived, orders released after the cut-off, inbound vehicles arriving unbooked, master data that misstates the size of a product. A schedule that measures only one party invites an argument in which every failure is contested. Including a small number of client obligations, measured on the same cadence, changes the tone of the review from blame to diagnosis.

Frequently asked questions

How many service levels should a schedule contain?

Few enough that each one is reviewed and acted on. A short set covering despatch timeliness, order accuracy, stock record accuracy and responsiveness to queries carries most of the weight. Long schedules tend to produce reports nobody reads and failures nobody notices.

Are service credits worth negotiating hard?

Only to the point where they command attention. Beyond that, providers price the exposure into the rate, so the client pays for its own credits. Escalation rights and termination triggers for sustained failure usually deliver more leverage than a larger credit.

Which system should be authoritative when figures disagree?

Name one in the agreement, and set out how a discrepancy is investigated. Most disputes are timing differences or definitional gaps rather than errors, and they resolve quickly when there is an agreed order of precedence and a defined reconciliation routine.