Decisions that changed what our network costs and how it fails

Replacing a restricted supplier's radios on 4,100 sites, one visit each

In 2021, 4,100 of the 9,400 radio sites we then operated, and parts of our packet core, used equipment from a supplier that had come to be treated as high-risk. Three months into my tenure, the board asked me for a removal plan. The core was the simpler part: we moved it to two other suppliers by September 2022, running old and new in parallel for four months. The radio network was the expensive question.

Our chief financial officer, Marco Pellandi, proposed keeping the radios until their normal end of life around 2029 and removing the supplier's remote access in the meantime, at about €130 million spread over eight years. I proposed replacing them between 2023 and 2025 for €196 million, combining each swap with removal of the 3G layer, new batteries wherever the old set was more than seven years old and the removal of cabinet cooling. Our marketing director feared visible service dips in our strongest regions. Sestriva Telecom, whose traffic runs on 1,300 of the sites we host under our rural sharing agreement, asked for a veto over swap windows, and we gave it one.

The board chose my plan in October 2021. We gave up a year on our 5G standalone launch, froze new builds in two regions for nine months, deferring about 180 sites, and wrote off €61 million of equipment. The work was split between two suppliers. Because I had worked for one of them, Halvorsk Networks, until 2012, I declared the connection and stayed out of the commercial scoring.

The last site was finished in November 2025, at a cost of €214 million. In 21 of 23 clusters, dropped-call rates returned to baseline within ten days; two clusters took six weeks. Swapped sites draw 27% less electricity for the same traffic, about 38 gigawatt hours a year. During the July 2025 heatwave, 212 of them stayed on air through power cuts their old batteries could not have carried.

Every supply contract we sign now prices the supplier's exit support at signature. What I would do differently is to have done that for this programme in 2022. In the final nine months, spare-part lead times on the last 900 sites stretched to three weeks, because I relied on the outgoing supplier's goodwill instead of negotiating its end-of-programme terms while we still had something it wanted.

Switching off 3G two and a half years early, and keeping 2G

Our electricity bill rose from €54 million in 2021 to €121 million in 2022. The 3G layer carried 1.6% of our data traffic but ran on every site and used about 11% of the radio network's electricity. Our plan had been to switch it off at the end of 2025. In September 2022 I proposed doing it region by region between February and June 2023.

About 310,000 lines still relied on 3G: 190,000 handsets without voice over 4G and 120,000 machine connections, among them payment terminals, lift emergency phones and alarm panels. Our enterprise director, Paolo Zennaro, argued that business customers needed until 2025 to replace devices and that we would lose some of them. Customer care expected a surge of calls from older customers. I agreed with both predictions. My proposal kept 2G running until at least 2029 as a fallback for voice and simple machine traffic, which costs us spectrum and maintenance, because many older phones that could not make voice calls over 4G could still fall back to 2G.

The executive committee approved the earlier date on three conditions: an €11 million device-replacement credit, letters and calls to every affected customer at least four months ahead, and a regional sequence that allowed us to stop if something went wrong. Alongside it we enabled night-time carrier and symbol shutdown on 7,200 sites.

Something did go wrong. In the first region, Marche, we found after switch-off that about 2,100 lift emergency phones, registered to building managers rather than to our enterprise accounts, used 3G-only modules with no 2G fallback. We switched 3G back on in two provinces for five weeks while installers replaced them.

Across the network about 21,000 lines left us, against the 30,000 we had assumed. Network electricity fell from 448 gigawatt hours in 2022 to 362 in 2025 while data traffic rose 71%, so energy per gigabyte fell by 53%. The 2025 bill was €76 million, although lower prices explain much of the fall from 2022.

We now keep a register of machine connections by device type rather than by account. What I would do differently is the search itself. We trusted account data when we should have scanned the network for 3G-only device identifiers months earlier, which would have found the lift phones before anyone was stuck in a lift.

Testing failover with live traffic after a six-hour voice outage

On 18 July 2024, during a heatwave, a cooling failure at our main core site near Bologna forced part of the voice core to shut down at 14:20. Failover to our second site in Rome should have taken minutes. It failed, because a routing change made at Bologna in March had never been applied in Rome. Voice calls and SMS were unavailable for 2.6 million customers for six hours and forty minutes, while data worked, slowly, for most. We notified the authorities as the rules require and credited affected customers.

At the audit and risk committee two directors asked for a third core site, which my team costed at €48 million over two years. Our operations director, Davide Stramazzo, wanted a period of stability and no risky testing. I argued that we did not yet know whether we had two core sites, only that we owned two buildings, and that the first priority was to prove failover by moving live traffic between them every quarter, at night, with customers' calls on it.

That meant accepting the risk of causing an outage ourselves, with the reporting and press attention that might follow. Davide set out his objection in writing for the committee, and it went with my paper. The committee approved quarterly live failover and automated checks for configuration drift between the sites, and deferred the third site to the 2027 budget.

The first live failover, in October 2024, found 14 configuration differences between the sites. The second, in February 2025, interrupted voice for 160,000 customers for 19 minutes at ten past two in the morning; we reported it and published an explanation. By September 2026 we had completed eight failovers, the last five with no interruption customers could see, and switchover averaged under four minutes. Drift checks now run nightly and have caught 37 differences before they mattered.

Resilience is now reported to the committee as tests passed and failed, not as equipment owned. What I would do differently is our communication on the day. The first message to customers went out two hours and ten minutes after the failure, because nobody below the chief executive could approve it. We now hold pre-approved messages that the duty manager can send within thirty minutes, including a reminder that a 112 call can usually be placed through another network.

Moving 168,000 copper customers by radio capacity, not by margin

Corvalta resold the incumbent's wholesale copper lines to about 205,000 of our 540,000 fixed broadband customers. Between 2024 and 2026 copper was switched off in exchange areas covering 168,000 of them, under the procedure supervised by AGCOM, the communications regulator. Each customer had to move to wholesale fibre to the home, move to our own 5G fixed-wireless service or leave.

Our commercial director, Chiara Benvoglio, proposed fixed wireless as the default wherever our signal reached, because it earned about €6 a month more margin per line than wholesale fibre and needed no technician visit at a third party's pace. On her figures, a fixed-wireless default would have put about 95,000 homes on our radio network. I opposed it on capacity grounds. A fixed-wireless home uses many times the data of a mobile customer, in the same evening hours and on the same mid-band spectrum. I proposed a cap per radio sector, set from our busy-hour model at 18 fixed-wireless homes, and fixed wireless only where an indoor signal test passed.

The executive committee adopted the cap. We gave up roughly €14 million a year of margin against Chiara's plan and accepted longer migrations where fibre installers were scarce. Chiara asked for the cap to be reviewed every six months, which was fair; in two reviews it has moved sector by sector, up in some places and down in others.

By September 2026, 112,000 customers had moved to fibre and 39,000 to fixed wireless, and about 17,000 had left us. Median evening download speed for fixed-wireless homes is 118 megabits per second, and their monthly churn is 1.1%, close to our fibre base. We also supplied battery units to about 9,400 customers who had told us of telecare or alarm devices, because neither fibre nor fixed wireless works in a power cut without one.

The sector cap is now part of how we plan spectrum and sites. What I would do differently is the starting cap in Puglia, which we set at 24 in the first months. Evening speeds fell below our published figure in 140 sectors there, and we moved about 2,300 customers to fibre after telling them fixed wireless was their service. I would also have made the battery offer opt-out from the first day.