MVNO SLAs are moving beyond monthly platform availability toward service outcomes that the tenant can observe and the host can isolate. The critical measure is not whether a BSS/OSS estate remained reachable, but whether subscribers attached, authenticated and sustained usable connectivity. PDU session continuity provides a more commercially relevant anchor for 5G and NSA service commitments, alongside attach success, latency, incident handling and a credit model that allocates accountability across the MNO, MVNE and MVNO.
Availability Is a Control Metric, Not the SLA Outcome
Platform availability should remain contractual because it establishes whether the managed estate was reachable and functioning. It cannot stand alone. Subscriber-facing failures can originate across the RAN, CN, IMS, provisioning and policy layers while every monitored application remains nominally available. The contract therefore needs to distinguish an infrastructure control metric from the outcome sold by the MVNO.
A 99.9% monthly commitment permits roughly 43 minutes of unavailability. That arithmetic says little about whether the loss occurred in one continuous event or hundreds of short interruptions, how many IMSIs were affected, or whether the event coincided with an activation campaign, payroll window or enterprise shift change. A short outage at a commercially sensitive time may carry more revenue impact than a longer event during a low-usage period.
The operating schedule should separate management-plane availability from service-plane outcomes. CRM access and reporting availability do not substitute for successful registration, PDP or PDU establishment, or charging continuity. The principal control questions are distinct:
Which service boundary covers the MNO RAN and CN, MVNE BSS/OSS, HLR, HSS, UDM, AuC, OCS and eSIM RSP?
Which denominator supports each KPI: eligible attach attempts, active IMSIs, attempted data sessions or affected subscriber minutes?
Which exclusions apply, who validates them, and how quickly must supporting evidence be supplied?
How are home-network failures separated from visited-network degradation, IPRN faults and transit effects in multi-IMSI or roaming-led propositions?
Without those boundaries, the SLA becomes a recurring negotiation over fault domain rather than a record of customer impact. Aggregate network availability is particularly weak as a denominator because it can remain high while a defined tenant population, APN or geography is effectively out of service.
In one operating review involving a Tier-2 MNO, Southeast Asia, ~18M subscribers, the material issue was not loss of platform reachability. A tenant-specific provisioning path failed while the wider estate remained available. The case illustrates the contractual gap: an estate-level measure can pass even when the MVNO cannot activate or serve a commercially relevant cohort.
Set KPIs at the Subscriber Transaction Level
Attach, session establishment and session retention track transactions that an MVNO can monetise and support. Attach success should be defined as successful registration or attach procedures divided by eligible attempts. Rejections, timeouts and retries need separate treatment. Exclusions should be limited to documented device, coverage and subscriber-status conditions, with evidence retained at a granularity that permits tenant reconciliation.
For EPC domains, the schedule should track PDP context activation success and unexpected bearer release. For 5G SA, it should specify PDU session establishment success, modification success and abnormal release rates. Initial attach is only the first control point. A subscriber who attaches successfully but repeatedly loses the data session remains commercially impaired.
Continuity should therefore combine session-drop rate, retained-session duration and re-establishment success after mobility, policy or charging events. Mobility-driven releases should be reported independently from network-initiated abnormal releases. The same distinction applies when a policy update or balance event interrupts traffic even though the access registration remains intact.
- Attach success
- Successful eligible attach or registration attempts divided by eligible attempts; define rejected, timed-out and retried attempts separately.
- PDU session establishment
- Successful PDU session establishments divided by valid requests, reported by DNN/APN, access type and geography.
- Session continuity
- Unexpected release rate and successful re-establishment rate, with mobility-driven releases reported independently.
- Latency
- Percentile-based round-trip latency from an agreed measurement point; avoid monthly averages as the sole commitment.
- Affected subscriber minutes
- Unique affected IMSIs multiplied by confirmed impact duration, used as the commercial severity input.
Reporting should segment results by access technology, geography, APN, enterprise versus consumer tenant profile, and time window. In 5G SA environments, the equivalent DNN segmentation should be explicit. A blended national result can conceal a severe cluster in a launch city, enterprise APN or recently migrated subscriber pool.
Latency and packet loss require equal discipline. Device probes, GTP-interface counters and synthetic tests observe different parts of the service path. The SLA should identify the measurement point, test population, sampling interval and percentile. A monthly mean can absorb short, severe periods that materially affect applications while still producing a compliant result.
A baseline period is prudent before credits take effect, particularly when the MVNO is migrating IMSIs, introducing eSIM activation flows or moving traffic between EPC and 5GC. The baseline should validate counters and exclusions, not suspend accountability indefinitely. It needs a fixed duration, an agreed acceptance process and a route for correcting measurement defects.
Incident Clocks Need Operational Evidence and Commercial Consequence
Response and restoration commitments work only when the counterparties pre-agree severity, notification, evidence and escalation. The incident clock may begin at NOC validation, an automated threshold breach or MVNO ticket acceptance. These triggers are not equivalent. Manual validation can defer the recorded start even when subscriber impact is already visible, while ticket acceptance can expose the tenant to delays in the host’s intake process.
Severity should reflect affected IMSIs, inability to activate or attach, sustained PDU session failures, regulatory exposure and the presence of a viable workaround. Alarm count is a poor proxy. One policy fault can affect a large tenant population with few alarms, while a noisy but redundant component can produce many alarms without measurable service loss.
A shared incident record should capture detection, notification, acknowledgement, engagement, workaround and restoration timestamps. It should identify affected MCC/MNC and APN populations, relevant SS7, Diameter or GTP traces, remedial actions and restoration confirmation. Evidence does not need to expose unrelated tenants, but it must be sufficient to support attribution and credit calculation.
The timetable should separate acknowledgement, technical engagement, workaround and restoration. A rapid acknowledgement does not compensate for a delayed correction to provisioning, PCRF/PCF policy or OCS behaviour. Tenant communications also need a defined cadence, accountable operational lead and conditions for executive escalation.
Chronic degradation requires a separate mechanism. Repeated sub-threshold failures can depress activation conversion, session duration and usage without any event crossing the individual P1 threshold. Rolling KPI misses, recurrence counts and unresolved root causes should therefore trigger escalation independently of major-incident classification.
An incident framework used by an MVNE servicing 12+ tenants in EMEA illustrates the structural issue. A host may coordinate one technical fault across several tenants, but each tenant can have different APNs, subscriber volumes and commercial exposure. The common incident record must support tenant-level impact calculations without fragmenting the underlying technical investigation.
Credit Mechanisms Should Price Material Failure, Not Create Perverse Incentives
Service credits should apply to the recurring wholesale charge associated with the affected tenant and service component. Applying them to unrelated traffic, roaming or pass-through charges weakens the connection between failure and remedy. The credit base, calculation period and eligible charge categories should be fixed in the commercial schedule rather than reconstructed after an incident.
A tiered schedule can combine severity with affected subscriber minutes. The monthly cap should be commercially meaningful but should not convert the SLA into uncapped consequential-loss cover. Separate remedies can address repeated failure, material breach, chronic KPI misses and termination rights where credits no longer provide an adequate incentive to correct performance.
Planned maintenance should be excluded only when notice periods, maintenance windows, rollback procedures and maximum duration are contractual. Broad exclusions allow recurring engineering activity to remove service from measurement. Emergency maintenance also needs a defined approval and notification path, followed by evidence that the intervention was necessary and limited in scope.
Credits should remain contingent on verified service failure, but verification cannot become an open-ended barrier. Evidence deadlines, dispute windows and deemed-acceptance rules reduce the risk that a valid claim remains unresolved across billing cycles. Where host and tenant counters differ, the contract should identify the authoritative source and the reconciliation procedure.
Root-cause analysis is a separate deliverable. It needs an owner, due date, corrective actions and closure evidence. Credit issuance without corrective-action governance can normalise recurring defects across provisioning, charging and policy domains. Conversely, a credit waiver tied to an unverified promise of remediation removes the commercial signal before the fault is controlled.
The commercial centre of an MVNO SLA is shifting from estate uptime to observable subscriber outcomes. Hosts that define denominators, fault domains and credit triggers before launch will reduce dispute cycles and give tenants a clearer basis for forecasting activation, usage and retention risk.