Decisions that cost time and money to protect customers

Stopping a billing migration after its third wave sent 7,900 wrong bills

In November 2024 Lonmara was moving 1.01 million mobile accounts from Lonmara Mobile's 2006 billing platform onto the convergent platform Tolvarra had implemented in 2021. The programme had sixteen planned waves and a business case that assumed €14.2 million a year of savings from closing the old platform by December 2025. The third wave included 64,000 households holding both our broadband and our mobile services, whose mobile billing dates we had aligned with their broadband dates. Three weeks after cut-over, 7,900 of them received bills overcharged by between €11 and €64, because the converged discount was prorated twice across the realigned cycle.

The stakes were the customers the merger most needed to keep: converged households left us at about a third of the rate of mobile-only customers. Halvecor, our integrator, proposed fixing the defect in the next release and continuing on schedule. Our finance director, Tomás Redfern, feared delayed savings and a precedent for pausing programmes that were on time. I stopped for seven weeks to change the test, accepting a quarter's delay and about €6.4 million of extra cost.

We refunded every affected customer within nine days, with a €20 credit and a letter that described the error plainly. From the fourth wave, every account's first bill on the new platform was compared with that account's last three bills, and any difference over twelve per cent or €8 was held until an analyst could state the reason in a sentence. Each bill run also needed signatures from revenue assurance and customer operations, neither of them measured on the schedule.

Across the remaining thirteen waves we compared 912,000 first bills, held 2,140 and corrected 530 before issue. Billing complaints per thousand migrated customers fell from 4.8 in the third wave to an average of 0.6 afterwards. The last accounts left the old platform in March 2026 instead of December 2025.

Full first-bill comparison is now required for any change to rating or discount logic. What I would do differently came before the programme began: I knew from 2010 that samples miss unusual combinations, yet accepted a five per cent sample because the integrator called full comparison unaffordable, without asking its price. When we finally did, the extra bill runs cost about €1.1 million; most of the €6.4 million was the pause itself.

Bringing billing knowledge back from the integrator

When the merger closed in March 2023, Lonmara Mobile employed eleven people in rating and billing. Halvecor Solutions had about 180 on the account under a managed-service contract priced per change request. The contract met its service levels. It failed the first time we needed something new: planning the consolidation required a precise account of how roughly 2,300 legacy tariff and discount variants behaved, and our own people could explain about a third of them.

The stakes were the merger's largest programme, worth €118 million over four years. Halvecor's first estimate for migrating the mobile catalogue was 9,400 person-days, a figure nobody on our side could test.

In February 2024 I proposed bringing the design, configuration and acceptance of rating and billing rules in-house, leaving Halvecor with platform engineering, regression testing and infrastructure. Tomás Redfern, our finance director, opposed adding permanent headcount when our owners expected merger costs to fall. Ostrevan's operating partner preferred a renegotiated contract with stronger knowledge-transfer clauses, an option about €2.2 million a year cheaper. The executive committee backed my proposal on condition that I report the cost difference every quarter until consolidation finished. We gave up the predictability of a price per change and accepted paying for idle capacity in quiet years.

We created 138 roles over eighteen months, from billing architects to a small team owning the tariff catalogue. Fifty-two went to engineers from our Halvecor account, under a transfer agreed with Halvecor. Every role had to pass one test: would we need this person to judge whether the integrator's answer was right?

In the first full year the in-house team cost €3.1 million more than equivalent change requests would have. Halvecor's revised estimate for the mobile catalogue fell to 7,300 person-days after our architects questioned it. Tariff lead time fell from 41 working days to 12, and our own analysts found the defect behind the November 2024 billing error.

The tariff catalogue now has a named owner inside Lonmara. What I would do differently is the transfer itself. We did not pair the Halvecor engineers with our own staff for long enough, and eight of the fifty-two left for other integrators during 2025 with knowledge nobody had written down. A retention award tied to documenting their rules would have cost less than replacing them.

Switching off the old mobile billing platform instead of keeping it as an archive

Once the last mobile accounts moved in March 2026, the 2006 billing platform and the prepaid charging system beside it held nothing live. They did hold twenty years of bills, payments, disputes and usage records. Our general counsel, Orla Brannick, and the director of customer operations both proposed keeping the platform running read-only for seven years, to answer disputes, data access requests and requests from authorities entitled to records. That would have cost €2.9 million a year, on an operating system its supplier no longer patched.

The stakes ran both ways. An unpatched dormant system was exactly the exposure our audit and risk committee had asked us to remove. But if we could not retrieve a bill for a customer disputing a 2023 charge, or answer a lawful request on time, the saving would be false and possibly a breach of our obligations.

I proposed extracting what we were obliged or likely to need into an archive we controlled, then decommissioning the platform completely. Orla disagreed at first. She did not want the company's ability to answer lawful requests to depend on an extraction whose completeness nobody could demonstrate. We agreed a condition: before switch-off, revenue assurance and legal would run 1,500 retrieval scenarios drawn from five years of real disputes and requests against both the old platform and the archive, and the platform would stay until every scenario matched. We gave up a quick saving and several months of an engineering team's time.

The first run matched 91%. The gaps were mostly scanned correspondence attached to disputes and manual adjustments made outside the rating engine. Closing them took ten weeks. The second run matched every scenario, and we switched off the platform in July 2026, with retention periods agreed with legal and data protection for each record type and a deletion schedule the old system had never applied. Running costs fell by €2.6 million a year after the archive's own costs.

Every decommissioning at Lonmara now begins with a retrieval test built from real requests rather than a data inventory. What I would do differently is minimisation. We archived some records because extracting them was easier than deciding, including old usage data for which none of us could name a current purpose. I should have pressed for decisions on those before extraction, when deleting them would have been simple.

Linking cable and mobile accounts only when the customer confirms it

After the merger Lonmara had 436,000 broadband households on one side and about 1.02 million mobile accounts on the other, with no common customer identifier. A cable account usually represents a household; a mobile account usually represents a person. Our commercial director, Aisling Prendeville, and her marketing team asked for a customer-data platform that matched the two by name, address and payment details, so that converged offers and the discount for holding both services could be applied automatically. A probabilistic match by our data team found about 118,000 likely pairs.

The stakes were commercial and personal. The converged discount was the merger's main argument to customers. But matching by address treats a household as one unit when it often is not. Our tests found former partners at the same address, adult children paying their own mobile bills and a landlord's broadband matched with a tenant's phone. An offer that mentioned services held by another member of a household could disclose something that person had not chosen to share.

I proposed that we link accounts only when a customer confirmed the link, through our app or a call, and that automated matching be used only to invite customers to check, never to apply offers or display another person's services. Aisling disagreed in writing. She estimated that confirmation-only linking would reach about half of converged households in the first year and cost around €7 million of revenue the plan assumed. Our data protection officer supported my proposal, and the chief executive accepted it with a quarterly review of take-up.

We built the customer-data platform around confirmed links, which made it smaller and simpler than the original specification. Invitations went to the 118,000 probable pairs from April 2025. By June 2026, 71,000 households had confirmed a link and 6,400 had told us the match was wrong, a figure that on its own justified the design. Revenue from converged discounts ran €4.8 million below the original plan over the period.

Linking is now the customer's decision across all our products, and the platform records who confirmed each link and when. What I would do differently is the invitation. Our first message read like a sales offer and drew a weak response until customer operations rewrote it as a service question. I should have asked them to write it from the beginning.