Manage vendors through the service experienced by users

99.8% can describe a service unusable at the wrong moment

This retrospective analysis of 2019 is published in 2026. The ERP case, participants, volumes and availability levels are illustrative. They demonstrate a method for interpreting service and do not describe a CYTIZEN result. The analysis does not argue that contractual commitments are useless: it explains what an aggregate measure can leave outside its scope.

Émilie, the finance director, does not dispute the provider’s report. It announces 99.8% monthly availability. Yet she observes three interruptions during closing, when accountants must post entries and reconcile flows. The same incidents would be inconvenient on a Sunday morning; at this time, they push tasks into the night and delay approval. The total number of minutes does not express that difference.

Marc, the service manager, must reconcile two views: the contract and operations. Changing an indicator without respecting its contractual definition would create another problem. He retains the existing measure and adds a view by journey, critical period and verified impact. The service can be discussed through facts without reinventing an SLA after every incident.

Break down the service before adding indicators

Marc lists finance journeys: entry, approval, transmission, reconciliation and closing. Each depends on applications, interfaces, data, permissions and teams. A stopped interface can make reconciliation impossible even when the ERP responds. A data defect can prevent approval without an infrastructure incident. Events are linked to responsibilities rather than automatically attributed to the main provider.

Critical periods are defined with the business before measurement. For closing, they cover certain dates and hours; for an industrial site, they may follow shift starts. The scope includes affected users and relevant dependencies. A window is not selected retrospectively simply because it makes the supplier look bad: the definition must remain stable and explainable.

The contract is reread: service periods, exclusions, maintenance, reporting mechanisms and measurement sources. If the definition no longer covers the current need, the contract manager prepares a change or negotiation. Operational management can reveal the gap, but does not itself constitute a signed amendment.

Put both availability measures in the same table

The illustrative contractual calculation is measured time minus recognised downtime, divided by measured time. Across 43,200 monthly minutes, 86.4 minutes correspond to 0.2% downtime. This representation does not say where those minutes concentrate or what users can accomplish. Exclusion rules must be retained exactly when comparing against the contract.

The proposed business view measures the proportion of time during which the essential journey is executable within a previously defined window. In a 600-minute closing example, three interruptions totalling 60 minutes produce 90% within that window. This does not cancel the contractual 99.8%, established over a different scope. It exposes a value and risk question that the monthly average is not intended to resolve alone.

Incidents are dated and classified. A failure preventing every login, an interface rejection and an approval blockage have different mechanisms. The team states observational limits and avoids calling every slowdown a complete outage. The business confirms whether activity stopped, degraded or continued with manual recovery.

A measure both parties can understand

Turn an impact into a supplier request

Marc prepares a case covering all three incidents. The provider identifies a scheduled task consuming capacity at the wrong time. The business confirms its closing window; operations examines dependencies and whether the task can move. The correction becomes a scheduling and control decision rather than a vague request for “better performance”.

The provider proposes rescheduling; the client checks that this does not degrade another country’s service. The change is authorised under the agreed procedure and then observed over several windows. If incidents stop but delayed processing compromises the following day, the result is insufficient. Evidence of success must cover dependent activities that the correction might displace.

A second hypothesis may require additional capacity. Marc distinguishes sizing of the contracted service, volume growth and operational defects. The contract manager handles commercial implications. The team does not accept a new cost without this classification, but neither does it block an urgent measure because a responsibility discussion remains open.

Restoring a component does not always end the incident

At 19:10, the interface restarts. The technical dashboard turns green. Accountants must still identify pending messages, verify processing and reconcile entries. If the incident is closed at 19:10 without recording that effort, the service appears faster to recover than it really is.

The model distinguishes technical restoration, journey recovery and final checking. Each stage has an owner; the necessary people must be available within the window. The supplier may own restoration and contribute to recovery while the business validates the result. This allocation must be known before the next incident rather than negotiated during the emergency.

The procedure includes duplicate-processing risks and cases where automated replay is prohibited. It identifies the escalation mechanism if checks fail. Service quality is therefore more than an availability curve; it includes the ability to restore a consistent business state after a disruption.

Ticket rates do not measure satisfaction on their own

Fewer tickets may mean fewer defects. They may also indicate that users stop reporting difficulties or that a coordinator now groups requests. Marc compares ticket volumes with usage volumes, major incidents and business feedback. He retains changes to counting rules in the report.

Satisfaction is examined in relation to a service and period, with a concrete context. “Are you satisfied with IT?” confuses hardware, tools, support and organisation. A question about completing closing or recovering an order is more useful, but does not replace technical events. Both types of data must be examined without treating either as absolute proof.

Workaround cost is estimated from observed times and people actually involved. Multiplying incident counts by the company’s entire workforce produces spectacular and rarely defensible amounts. Specify what is measured, what is estimated and what cannot reliably be quantified.

Establish responsibilities that last

The service owner maintains the link between commitments and journeys. The business manager confirms impact and critical periods. The provider contributes events and treatment options. The contract manager governs commercial consequences. Security or quality participates where an incident affects its requirements. Marc coordinates these decisions without replacing the authorities of each domain.

The service review examines trends and recurring problems. An incident task force works at another cadence; it must not wait for the next meeting. Corrective actions have evidence of closure and a date. If scope changes, the measurement model is explicitly revised and historical comparisons annotated.

A supplier cannot compensate for absent client decisions. If a window cannot be reserved or the business has not named its validator, those dependencies are made visible. The relationship becomes more demanding because it is more precise, and fairer because each party can demonstrate what belongs to the other.

What actually changes with the activity

In pharmaceuticals, service recovery must preserve the integrity and traceability of affected operations. Quality checks may form part of recovery; high availability is insufficient if nobody can explain which records were affected. Particular obligations are examined with the competent functions.

In finance, a closing or payment window changes an event’s significance. The case tracks transactions and reconciliations with appropriate segregation of permissions. In manufacturing, local availability, shift starts and connections to site systems may dominate. Measurement then concerns a journey genuinely executable by the operators present, with the planned fallback mode.

In services, a queue may remain invisible behind available applications. End-to-end lead time and abandonment must be examined alongside utilisation. The aim is not to add all these indicators everywhere, but to select those that trigger a useful decision for the service in question.

Sources and method

Official pages consulted in 2026 provide management and measurement guidance. Their British administrative rules are not automatically transposed to a private French contract. Windows, business-view formulas and case figures are original and educational. The year 2019 locates the issue studied; no contemporary prediction or result is claimed.

• GOV.UK — The Sourcing Playbook, current version consulted in 2026
• GOV.UK — How to set performance metrics for your service

Links verified on 4 October 2026. The scenarios and thresholds proposed in this dossier are educational; they describe neither a client engagement nor a result achieved by CYTIZEN.