TL;DR

Government clients in the UK are still wary of AI, according to Brendan Mooney, chief executive of the Belfast software company Kainos. He spoke to BBC News NI on 9 November as the firm reported a record half-year, with revenue of £269m and £36m in pre-tax profit. Mooney said that the hardest part of the job is not building AI systems but getting them live and keeping them there.

Two conversations at once

Mooney described “parallel conversations” with clients: excitement about what AI can do, and what it takes to deploy it in big organisations. Those working in sensitive or regulated fields such as healthcare and defence are “super cautious”, he said.

His explanation for the difficulty is technical. “An AI system runs on top of a model, that model drifts, either through changes to the model or changes to how the model interprets the data,” he told the BBC.

Kainos’s own AI spending has doubled in a year, Mooney said, partly from wider internal use and partly because the major AI companies have raised their prices. He said productivity gains in software development are “well proven” and generating value for the firm.

The results follow a £111m Home Office “digital borders” contract that Kainos will deliver over four years.

Not replacing graduates

Mooney dismissed the argument that AI makes junior roles redundant as “one-dimensional”. Graduates can and should take on more complex work than in the past, he said, adding that treating them as fit only for simple tasks “shows limited imagination”. This year Kainos expects over 200 graduate and school-leaver recruits, which Mooney called its strongest intake yet. Of its workforce of nearly 4,000, about 1,400 are based in Northern Ireland.

What it means

Kainos works directly for UK government, including the Home Office, so its chief executive’s view is a useful check on ministerial ambitions for public sector AI. We have covered the firm’s earlier work on what holds back AI at scale and getting AI into production.

In our view, the model-drift point deserves close attention in procurement. A government AI service is not finished at launch: a change to the underlying model, or to the data it sees, can alter its behaviour. Contracts that fund build but not ongoing monitoring leave that risk with the department. Mooney’s rising-prices comment adds a second cost that buyers should plan for.