Most software companies want more customers. Liberty IT has spent almost 30 years going in the other direction. The Belfast-rooted company builds technology for Liberty Mutual Insurance, its parent and effectively its only client. That sounds cosy until you consider the assignment: keep modernising the machinery of a global insurer while real claims, payments and policy decisions continue to move through it.
This is not a startup hunting product-market fit, nor a consultancy arriving with an immaculate deck and departing before maintenance gets awkward. Liberty IT sits inside the consequences. Its engineers, data scientists, designers, product owners and security specialists build customer journeys, cloud platforms, fraud and risk tools, internal data products and AI systems. If an experiment graduates, the same organisation must make it observable, secure, governable and boring enough to run.
That one-customer model is the fact from which nearly everything else follows. Liberty IT can learn insurance at a depth that a rotating project team rarely achieves. It can reuse a successful architectural pattern across business units. It can also spend years improving capabilities whose value appears as fewer calls, quicker decisions or lower infrastructure burden rather than a neat subscription line. The trade is concentration: there is no diverse customer book to soften a change in the parent's priorities.
A £3.5 million foothold in Belfast
Liberty Mutual announced the Northern Ireland operation in January 1997: a £3.5 million investment to produce insurance software for internal use across the American insurer's offices. The local Industrial Development Board supplied £837,000 in grant aid. The plan called for 35 graduates at first, training in New Hampshire, and 124 jobs within five years. Four years later, roughly 100 people were already in place and another 100 jobs were proposed, with e-commerce among the targets.
Today the company occupies an interesting market slot. Against Belfast neighbours such as Allstate's and Aflac's technology operations, it competes for engineers. Against Accenture, Cognizant and Deloitte, it represents the build-it-inside option. Against Guidewire and other insurance platforms, it is the team that integrates, extends or creates what Liberty Mutual needs. But Liberty IT is not really trying to beat those firms in a public sales contest. Its benchmark is whether Liberty Mutual becomes easier to operate and easier to deal with.
The chatbot was the least interesting part
The company's Virtual Assistant makes the model tangible. Built with Amazon Lex and Amazon Connect, it brought a serverless, AI-backed conversational system into Liberty Mutual's customer service. It operates around the clock for routine queries. Liberty IT says the platform saved its parent millions of dollars, earned engineer Chris Gormley a Liberty Mutual Chairman's Award and became an AWS re:Invent reference architecture.
The important artifact was not simply a bot that could talk. It was a reusable serverless approach. Managed cloud services reduced the infrastructure that teams had to own, while engineers carried the resulting patterns into other work. That is how a useful internal project compounds: the first application delivers an outcome; the architecture makes the next application cheaper or quicker.
“Innovation is simply delivering new value.”Andy O'Sullivan, Liberty IT solutions engineer
That definition is unusually stern for an innovation group. A blockchain prototype, a chatbot or a dazzling model is only an idea until it changes the economics or experience of the business. Liberty IT's own formulation is blunter still: in an enterprise with billions in revenue, innovation should make or save millions. Novelty is welcome, but it does not get to mark its own homework.
They saw GenAI early - then waited
Liberty IT's Generative AI story includes a detail most corporate origin myths edit out. Teams identified the technology in late 2022 and wrote a white paper. Then they did not immediately move. When ChatGPT went mainstream in early 2023, they reopened the work. The public shock changed the urgency; the earlier research meant they were not starting cold.
They stood up an internal ChatGPT-like service, created a reusable GenAI architecture and built a platform able to serve a portfolio of foundation and frontier models. Engineers adopted GitHub Copilot. Knowledge assistants began retrieving information scattered through tools such as SharePoint and Slack. Liberty IT says the internal assistant has saved hundreds of thousands of hours, and that GenAI has compressed database-conversion work that once took months. In one conversion exercise, teams tested models until achieving 100 percent accuracy, avoiding significant outside spend. Neither the spend nor the evaluation set has been disclosed, so those results are best read as company-reported operating evidence, not a general benchmark.
What failed first was not the mathematics. Early experiments ran into limited model access, throughput problems and the mundane indignity of asking for more tokens. The harder obstacles were organisational: data, process, privacy, procurement and security. Liberty IT's answer was “Responsible AI by Design,” a multi-step review built into product development, supported by specialists from each of those functions.
There is a practical lesson here. A company can copy the sequence without copying the insurer's scale: scan early, write down the opportunity and risks, protect a small experimentation budget, build one shared access layer, choose use cases tied to actual work, and put governance beside the product team before launch. Add model evaluation, monitoring and a path for humans to override the machine. Buying licences for a famous model is the easy square on this board.
The factory builds people, too
The other half of Liberty IT's operating system is talent. In 2025, more than 850 employees across Belfast, Dublin and Galway logged over 25,000 hours of formal training. The company runs specialist technical education, AWS certifications, leadership paths, monthly GenAI office hours, promptathons and an annual internal conference. Its first TechCon gathered 600 employees in 2024. The Ignite Hackathon gives teams three days to turn ideas into working prototypes; in 2025, Liberty IT teams won second place overall, the Tech Excellence Award and the People's Choice Award.
The culture programme can occasionally sound like the technology programme wearing a party hat. LibertyGPT, an internal assistant, has even helped employees write nominations for “Culture Stars,” a peer-recognition scheme. Yet there are measurable commitments beneath the language. Women in Tech, Pride, Able and Unity groups run mentoring and accessibility work. Flexible hybrid arrangements accompany offices intended for collaboration rather than daily attendance theatre.
Outside the company, STEAM Studio puts 13- to 15-year-olds into hands-on technology sessions at the Ulster Museum, in partnership with National Museums NI. Students build a game during the day. Liberty IT covered school travel, expanded after uniformly strong pilot feedback, and later tested a Galway version. Its education and outreach activity reached more than 5,500 young people in 2025. This is community work, but also patient ecosystem building in cities where every technology employer needs the next cohort.
The advantage
One customer, shared incentives, deep insurance context and the chance to reuse platforms across a global organisation.
The constraint
One parent's priorities, enterprise controls, legacy complexity and no outside market to independently price the work.
The privilege - and price - of one customer
Liberty IT's playbook works because several conditions line up. The parent is large enough to fund reusable capability. The problem set repeats across many policyholders and teams. Senior leaders tolerate experiments, while engineering teams remain responsible for production. Deep domain knowledge has time to accumulate. Remove those conditions and the model weakens.
A smaller insurer may not generate enough repeat use to justify an internal platform. A company whose technology needs change wildly may be better served by vendors. Serverless is less attractive when workloads are steady, highly portable infrastructure is essential or a managed service cannot satisfy regulation. GenAI assistants disappoint when source knowledge is stale, evaluation is vague or employees do not trust the retrieval. Hackathons become expensive theatre if prototypes have no owner on Monday.
There is also a measurement problem. Liberty IT can point to millions saved by a virtual assistant and hundreds of thousands of hours saved by internal GenAI, but its financial statements do not offer outsiders a clean product margin or customer-retention chart. The supplied revenue estimate of £5.328 million is hard to interpret for a captive subsidiary and should not be confused with the value of the software it creates. The fairest scorecard is operational: systems shipped, costs avoided, decisions improved and capabilities reused.
By that measure, the company's distinction is not that it uses AI, AWS or agile teams; almost every technology organisation says the same. It is that Liberty IT has been allowed to keep learning one complicated business since 1997. The original Belfast bet bought an enduring seat inside Liberty Mutual's problems. The useful lesson is gloriously unflashy: know the domain, make experiments answer to economics, keep the architecture reusable, and stay around for what breaks after launch.