RESOURCES / ARTICLE · STEERING & ROAMING

Why network-side steering breaks at scale — and what SIM-level steering fixes

TELKOA · AUGUST 2026

Steering of roaming is one of the oldest levers in the wholesale playbook: keep outbound roamers on preferred partner networks, protect margins, honour bilateral commitments. The standard tools — signalling-based steering in the core, OTA updates to preferred-network lists — work well enough in the average case. The problem is that roaming economics aren't decided by the average case. They're decided by the edge cases, and at fleet scale the edge cases are a population.

FAILURE MODE 1Losing the retry battle

Signalling-based steering works by rejecting registration attempts on non-preferred networks until the device tries a preferred one. But the visited network has a say too. Anti-steering mechanisms, aggressive re-registration handling, and plain protocol timing mean a meaningful share of devices end up camping exactly where you didn't want them. The steering platform reports the rejection it sent; it doesn't report the network the device finally settled on. You find that out from the wholesale invoice.

FAILURE MODE 2The forbidden-network trap

When a device is rejected repeatedly, it writes the network into its forbidden list (FPLMN) — and stops trying. Now the device may sit with degraded or no service, and the standard remedy is an OTA update to clear the SIM file. Which must be delivered… over the network the device can't properly reach. For consumer handsets a user reboot eventually clears things. For an unattended IoT device in a locker, a vehicle, or a meter cabinet, "eventually" is a truck roll.

FAILURE MODE 3OTA at fleet scale

Preferred-network lists change with every renegotiated agreement. Delivering those changes by conventional OTA means SMS campaigns across millions of SIMs: delivery windows, retries, devices offline, devices roaming, devices whose radio is asleep. Campaigns take days to converge and never quite reach 100%. Meanwhile the old priorities keep steering traffic to the old partner. The gap between "policy decided" and "policy live in the field" is pure margin leakage.

FAILURE MODE 4Steering blind

Core-based steering knows which network a device registered on. It doesn't know what the device experienced there: signal quality, attach failures, data speeds, time without service. So steering policy optimises cost against a quality picture that doesn't exist — and the first sign that a "cheap" partner is a bad partner is a churn curve or an enterprise SLA dispute.

What changes when steering logic lives in the SIM

The SIM is where network selection actually happens — the preferred and forbidden lists are SIM files, the subscriber identity is a SIM file. Placing steering logic there, as an applet, changes the failure modes structurally:

No retry battle. The SIM updates its own network priorities and refreshes — the device selects the preferred network because its own configuration says so, not because a distant core rejected it into compliance.

Self-rescue from forbidden states. SIM-resident rules detect no-service and registration-timeout conditions locally and act — resetting FPLMN, cycling network priorities, or switching to a different IMSI — with no server connection required. The device escapes the trap by itself; the platform hears about it afterwards.

Policy changes in near real time. New steering rules go to the applet over encrypted IP (with SMS fallback), per SIM or per segment, with delivery tracked and confirmed. Renegotiated a rate yesterday? Steer to it today.

Steering with eyes. The same applet measures what each device experiences on each network — signal, attach success, service availability, achieved data speeds. Cost policy and quality policy finally use the same data, and partner-performance conversations happen with your own evidence.

None of this replaces network-side steering where it works. It closes the gaps where network-side steering structurally can't: the retry battles, the trapped devices, the convergence lag, and the blindness.

The economics, briefly

Every mis-steered device-day is either wholesale premium paid or subscriber experience burned. At small scale that's noise; across a large roaming or IoT estate it's a line item. SIM-level steering attacks the line item at its source — with decisions made where they take effect, and an audit trail behind every change.

RELATED READING

Scope a steering use case.