A private 5G business case is usually built with care. Radio units are counted, spectrum is priced, integration is scoped, and a five-year total cost of ownership emerges that compares favourably against the Wi-Fi estate it replaces. The board approves it. The network goes live on schedule.
Then year two arrives. A production line moves twelve metres and throughput on three machines collapses. Nobody in the building can say why. The integrator’s service desk confirms the network is available, because by every measure in the contract it is. The plant manager’s problem is not availability. It is that the network no longer works for the use case it was justified by.
This gap has a name, and it is not a technical one. It is a line missing from the business case: the cost of operating a mobile network, which the enterprise has now acquired along with the radios.
What the enterprise actually inherits
Deploying a private network transfers five operational disciplines to an organisation that has almost certainly never performed any of them.
- RF planning and spectrum coordination. Coverage is designed once, for a physical layout that changes. In shared or licensed-light bands, coordination with neighbouring users is ongoing, not a launch task.
- RAN performance monitoring and optimisation. Counters, KPIs, neighbour relations, interference analysis. This is a specialist discipline in every mobile operator, staffed by a dedicated team.
- Core network lifecycle management. A cloud-native core means container upgrades, patch cycles, rollback procedures and compatibility testing against the RAN. It is software operations with mobile-specific failure modes.
- SIM, device and identity management at scale. Provisioning, credential lifecycle, device certification, and the question of what happens when a device vendor changes firmware.
- Fault isolation across the IT/OT boundary. When an automated guided vehicle stops, the cause may be radio, core, application, or the vehicle. Someone must be able to tell the difference, quickly.
An enterprise IT team is generally strong on the third discipline and has no exposure to the other four. That is not a criticism; there was no reason to build those skills before.
Why the managed-service model thins out around year two
The standard answer is that the systems integrator handles operations, and for the first year that is broadly true. The difficulty is what the contract actually guarantees. Managed-service SLAs are written around availability, incident response times and fault resolution. They are not written around performance against a use case, because the use case belongs to the customer and changes without notice.
So the network is up, and the SLA is met, and the throughput problem on those three machines is a change request. Multiply that across a few dozen layout changes, device refreshes and new applications over five years, and a recurring cost appears that no one modelled. Meanwhile the enterprise has no internal capability to challenge the integrator’s diagnosis, which weakens its position in exactly the negotiations where it needs leverage.
Consider a case that will be recognisable to anyone working on German campus networks. It is drawn from patterns common across those deployments rather than from a single site.
A tier-one automotive supplier licenses 100 MHz in the 3.7-3.8 GHz band for a plant of roughly half a square kilometre, deploys a standalone network for AGV control and torque-tool telemetry, and hands day-to-day operations to its integrator. Eighteen months in, the plant reconfigures a body-shop cell and moves a line by a few metres. AGVs begin dropping control sessions in one aisle. The monitoring dashboard shows the network available and the cells within normal parameters, because at cell level they are. The actual cause – a new metal enclosure creating a coverage hole at AGV antenna height, compounded by an unfavourable TDD frame alignment with an adjacent cell – takes several weeks to isolate, because nobody on site can read RAN counters and the integrator is working to an SLA the network has not breached. Production runs degraded throughout.
Nothing in that sequence is a technology failure. The radios worked, the core worked, the contract was honoured. What was missing was someone who could look at the counters in week one and say what had changed.
Three operating models, compared honestly
There is no single right answer here, but the trade-offs are more predictable than most business cases assume.
Fully managed. The integrator retains all five disciplines. Lowest internal headcount, highest recurring fee, longest mean time to resolution for anything outside the SLA, and maximum vendor dependency at renewal. Appropriate for single-site deployments with stable layouts and undemanding applications.
Hybrid. The enterprise employs one or two people who genuinely understand RAN behaviour and core operations, and retains the integrator for depth. This is the model that changes outcomes most, because the internal specialists are not there to fix everything. They are there to diagnose accurately, to know which problems are real, and to hold the integrator to a technical standard. Mean time to resolution falls sharply for the ordinary problems, which are most of them.
Fully in-house. Justifiable at multi-site scale or where the network is genuinely mission-critical and downtime carries a large per-hour cost. Below that threshold the headcount rarely pays back.
The figures below are indicative planning ranges for a mid-sized single-site European deployment, not measured results from a specific project. They are offered to show the relative weight of each line rather than to price any particular network.
Start with what the spectrum costs, because it anchors everything else. Germany assigns local 3.7-3.8 GHz licences under a published formula: 1000 + B x t x 5 x (6a1 + a2), where B is bandwidth in MHz and t the licence term in years. For 100 MHz over ten years across a site of that size, the fee lands in the region of EUR 16,500 – about EUR 1,650 a year. Against a deployment whose infrastructure and integration typically runs EUR 800,000 to EUR 1.5 million, spectrum is a rounding error.
Operations are not. A full managed-service arrangement commonly prices at 15 to 20 per cent of initial CAPEX annually, or roughly EUR 150,000 to EUR 250,000 a year on that deployment. Over five years that is EUR 750,000 to EUR 1.25 million – an operating cost comparable to the entire build, and one that many business cases carry as a single unexamined line.
The hybrid model reallocates rather than adds. One RAN-literate engineer costs perhaps EUR 80,000 to EUR 110,000 fully loaded in Western Europe, and typically allows the managed-service scope to be narrowed to depth support in the 8 to 12 per cent range. The net five-year difference is often modest. What changes materially is time to resolution: the coverage problem described above is a matter of hours for someone who can read the counters, and weeks for an organisation that cannot.
Two smaller lines deserve their own budget rather than being absorbed as incidents. A meaningful layout change triggers an RF re-planning exercise in the region of EUR 15,000 to EUR 30,000. And a plant running continuously has no natural maintenance window, so each core upgrade carries scheduling and risk cost that a nine-to-five site does not.
Where the hidden costs actually sit
The unmodelled costs are individually small and collectively material over a five-year horizon.
- Re-planning after physical changes. Every significant layout change is an RF planning exercise, not a configuration change.
- Device certification cycles. New devices, new firmware, and the testing required before either reaches the production floor.
- Upgrade windows in a continuous operation. A plant that runs 24/7 has no natural maintenance window, which makes every core upgrade a scheduling and risk exercise.
- Spectrum coordination. In shared bands this is a standing obligation with compliance consequences, not a one-off filing.
- Knowledge continuity. Whoever learns the network becomes a single point of failure the day they resign.
A readiness checklist before signing
These questions cost nothing to ask and are considerably cheaper to answer before deployment than after.
- Who diagnoses a throughput complaint that the network monitoring reports as normal?
- What is the escalation path at 02:00, and who has authority to act on it?
- Who owns spectrum compliance, and what happens if a neighbouring deployment causes interference?
- What happens to coverage when the physical layout changes, and who pays for the re-plan?
- Who certifies a new device type, and how long does that take?
- What is the rollback procedure for a failed core upgrade, and who has rehearsed it?
- If the integrator’s diagnosis is wrong, who inside the organisation is able to say so?
- What is the plan when the one person who understands the network leaves?
Reframing the model
The correction is not complicated. It requires moving a defined share of the five-year model out of deployment and into a standing operating-capability line, covering managed services, internal specialist time, training, and a genuine contingency for re-planning as the site evolves.
A business case built this way looks less attractive on day one. It is also the one that survives contact with year three, when the deployment is finished, the integrator’s attention has moved to the next customer, and the network has become something the enterprise simply has to run. Private networks do not fail because the technology underperforms. They disappoint because the organisation that bought them was never resourced to operate them.
Mohamed Idrissi is CTO and founder of 5GWorldPro.com, a training platform for telecom and IT engineers. He has spent more than 18 years across radio, core, transport and IT architecture, on both operator and vendor sides.
