Government AI engineers will be embedded in Liverpool City Region for 12 months to build tools that identify 16 to 24-year-olds at risk of falling out of work, education or training. The partnership between the Incubator for Artificial Intelligence, known as i.AI, and the Liverpool City Region Combined Authority was agreed today by AI Minister Kanishka Narayan and Mayor Steve Rotheram.

It is the first partnership of its kind in England, and work begins in January 2027.

What the engineers will actually do

The brief is narrower than the headline suggests, and more interesting for it. The team will look at whether information already held across different public services can be brought together so that careers advisers get a clearer picture of where help is needed.

  • Spotting risk earlier. A young person who stops attending college, drops out of a training scheme and stops engaging with support services often appears in three separate council systems. Joined up, those signals look like a pattern.
  • Giving advisers better information. The output is meant to be a prompt for a trained professional, not an automated decision.
  • Testing what data can lawfully be combined. Much of the work is likely to be legal and technical rather than model building.

i.AI engineers will be embedded inside the Combined Authority and its six constituent local authorities, working alongside existing teams rather than delivering a finished product from London.

The government is explicit that the technology supports decision-making rather than replacing it, with trained advisers keeping responsibility for what support each young person receives. That distinction is the one to hold on to, because the failure mode for this kind of project is an algorithm quietly ranking teenagers.

Why Liverpool, and why now

The partnership is part of the government's response to the Milburn Report, which argued for stronger support for young people at a point when around one million people nationally are not in employment, education or training.

Narayan's framing is blunt: nearly a million young people out of work, education or training is an enormous waste of talent, and the system that leaves them behind is not acceptable.

Liverpool City Region also has the governance to make it work. A combined authority with six constituent councils and a directly elected mayor is a single counterpart for a national team, which is easier than negotiating with six separate authorities. If the model proves out, the government intends to explore rolling it out across England's other mayoral strategic authorities.

The uncomfortable questions

Three things will decide whether this is a success or a cautionary tale.

Data quality. Records held by a council about a teenager who has already disengaged are often the least reliable in the system. A model trained on patchy data produces confident answers about the people it knows least about.

Consent and legal basis. Combining social care, education and employment data raises data protection questions that are not solved by good intentions. The Information Commissioner has been consistent that public bodies need a clear lawful basis for this kind of matching, not just a policy aim.

What happens when the prompt is wrong. If an adviser is told a young person is at risk and they are not, the cost is a wasted conversation. If the reverse happens, the cost falls on someone who was already being missed.

What to watch

  • January 2027, when engineers start and the first workstreams are named
  • Whether the tools are described in public, because a project that cannot be explained cannot be scrutinised
  • Whether advisers are given an override, and whether overriding is recorded
  • What happens to the data when the 12 months end

This is the same argument that runs through every public sector AI deployment: the model is rarely the hard part. The hard part is the legal basis, the data quality and who is accountable when the output is wrong. For background on how quickly the underlying capability has moved, our piece on how open-weight models closed the gap explains the technical side, and what a neural processing unit does covers where this work actually runs.

The bottom line

A national AI team is moving into a combined authority for a year to see whether joined-up data can get a vulnerable 17-year-old a conversation with an adviser sooner. That is a reasonable ambition and a hard problem. The test is not whether the model works in a demo, but whether the data is good enough and the accountability is clear enough to use it on real people.