Faster isn’t the same as right: why assistance still needs a human in the language loop

Few corners of the travel insurance world have moved as quickly on artificial intelligence (AI) as language services.

Machine translation now turns a foreign discharge summary around in seconds, and AI telephone interpreting is moving from pilot to procurement. For assistance companies under constant pressure to cut case-handling times and costs, the appeal is obvious.

The question is no longer whether AI belongs in the language workflow – it already does. The question is where it should stop.

Where medical risk emerges

Assistance cases are rarely routine for long. A toxicology result can decide whether a claim is covered. A handwritten note from a rural clinic may be the only record of a medication given. A fit-to-fly assessment can determine whether a repatriation goes ahead – and whether the patient travels in a commercial seat, on a stretcher or by air ambulance.

These are the documents where AI is least reliable – and where errors cost most. Foreign medical records are rarely clean source material: scanned pages, handwriting, local abbreviations and drug brands that exist in only one country.

The danger is not that machine translation looks wrong – it is that it looks right. AI output is fluent and confident, and rarely flags what it is unsure of.

A case manager cannot tell which line was misread until a medical director or claims auditor discovers it the hard way. In a medical context, a translation that cannot be trusted line by line cannot be trusted at all.

The danger is not that machine translation looks wrong – it is that it looks right

The consequences are tangible: treatment authorised on an incomplete history, a patient flown before they are fit, a claim declined in error, or days added to an overseas hospital stay while information is chased and re-checked.

Accountability matters too. Under UK GDPR, and with the EU AI Act’s obligations phasing in, insurers and assistance providers need to know who is answerable for a decision built on translated information. ‘The algorithm’ is not a satisfactory answer for a regulator, or for a policyholder.

Human in the loop, by design

At Lexxika, we use AI throughout our processes – and we are clear about where human expertise remains essential. The model that works is not human versus machine, but a defined division of labour: AI for speed and scale, specialist linguists for clinical accuracy, nuance and sign-off.

At Lexxika, we use AI throughout our processes – and we are clear about where human expertise remains essential

In practice, that means risk-based workflows with anything that informs a medical, coverage or repatriation decision being reviewed by a translator with medical and insurance expertise.

Keeping people at the centre

Assistance has always been a people business. Behind every file is a traveller in an unfamiliar hospital, often frightened and far from home. AI can make the industry faster, but trust is built by getting it right.

The providers that thrive will be those that use technology to give their people more time for the moments that matter – not fewer people for those moments.