AI and the new travel risk intelligence cycle

AI and the new travel risk intelligence cycle

AI is transforming travel risk management – from accelerating intelligence gathering and analysis to enabling more personalised traveller communications. Chloe Fox speaks with industry experts about the complex challenges this raises, including misinformation, verification, and the need for human oversight

Artificial intelligence (AI) is beginning to reshape how travel risk intelligence is gathered, interpreted, and communicated, with clear implications for underwriting, claims handling, traveller engagement, and duty of care.

For insurers, assistance providers, and travel risk management (TRM) firms, it brings the promise of faster insight and more tailored communication, alongside the potential for more responsive operational decision-making. At the same time, concerns are growing around misinformation, deepfakes, and automation bias, as well as the limits of systems that still struggle to apply meaningful context and judgement.

As with many areas of insurance and assistance, the conversation around AI in travel risk management is moving rapidly from theory to implementation. While the technology is evolving fast, many experts argue that its value lies not in autonomous decision-making but rather in helping organisations process information at a scale and speed that would previously have been impossible.

“The area where AI is earning its place is aggregation and speed,” said Graeme Dean, Founder and CEO of Hotspot Cover. “Pulling together news feeds, government advisories, weather events, civil unrest indicators, and doing it in near real time is a real operational improvement on what most of this industry was doing manually until recently.”

Dean said the benefits were already operationally visible: “For us at Hotspot Cover, that translates directly into faster responses to developing situations and more relevant, timely information reaching customers instead of the usual generic advisory that lands 12 hours too late.”

Bruce McIndoe, President and Founder of McIndoe Risk Advisory, also sees AI’s ability to handle large volumes of fast-moving data as its key strength. “The strongest use cases today are multi-source monitoring, entity extraction, language translation, summarisation, relevance scoring, anomaly detection, and targeted alerting,” he said.

“In practice, that means AI is helping teams process large volumes of threat and incident reporting, weather data, advisories, local-language content, transport disruption data, and internal traveller itineraries faster than human analysts could do alone.” Gulnaz Ukassova, Director of Information and Analysis – Middle East at International SOS, said the technology was proving particularly valuable in situations where speed matters.

“It allows teams to process large volumes of data quickly, detect certain trends or patterns earlier, and flag anomalies across sources such as aviation data, weather patterns, health signals, and incident reporting,” she said. “This capability is especially valuable in fast-moving situations where early indicators matter.”

Alex Craxton, TRA Mentor at Travel Risk Academy (TRA), pointed to the growing use of AI for monitoring developing situations: “In many risk intelligence vendors today, they are deploying AI to monitor for new and developing situations, looking at events, and analysing all the different sources for trust, accuracy, and relevance,” he said. “There is also some innovation around more relevant and personalised pre-travel advisories based on the individual, changing circumstances, and their travel itineraries.”

While AI’s ability to gather and sort information is improving rapidly, most experts remain cautious about how far the technology can currently go when it comes to interpretation and judgement.

Ukassova stressed that actionable intelligence still depends heavily on human expertise. “AI can surface signals, but it cannot reliably assess nuance or the local dynamics that may have a significant influence on how a situation will evolve,” she said. “Human analysts remain essential for translating data into meaningful, actionable insights.”

Dean is similarly sceptical of suggestions that AI can independently produce finished intelligence. “Where I’d push back on the hype is contextual judgement,” he said. “AI is genuinely poor at it. It can’t reliably tell the difference between a localised protest that’s over by lunchtime and something that signals deeper instability.

“The intelligence AI surfaces still need a human to interpret it and any organisation presenting AI outputs as finished, actionable intelligence rather than a starting point is, frankly, taking a risk with their clients.”

McIndoe agreed that autonomous judgement remained one of the industry’s most oversold areas. “Generative systems are not yet reliable enough to independently determine severity, intent, likely escalation, or the correct duty-of-care action in ambiguous situations,” he said.

Craxton also touched on the danger of overenthusiasm around agentic AI: “I think general hype around AI and agentic AI, more in particular on how it can automate activities, is distracting people from the fact they still need humans in the loop to check, authorise, and be sceptical,” he said.

That need for oversight is becoming increasingly important as misinformation, manipulated content, and synthetic media become more sophisticated, particularly as the volume of AI-generated material increases across the wider information ecosystem.

Several experts warn that the growth of AI-generated content is already making it harder for organisations to distinguish signal from noise.

“We hear from many organisations that they are receiving too much risk information, and it is hard or impossible to process it all within their team, given the pressures and time constraints,” said Craxton. “Many larger global security operations centres (GSOCs) analyse multiple intelligence sources and complain that it is becoming too noisy.”

Dean told ITIJ that the wider information landscape was becoming increasingly cluttered. “More information has never automatically meant better information, and right now AI is making it easier than ever for people who have no business contributing to this conversation to do exactly that,” he said. “You see it everywhere, LinkedIn especially. It all looks polished, it all sounds authoritative, and a lot of it is just noise dressed up as expertise.”

The rise of deepfakes and synthetic media is adding another layer of complexity. “A fabricated video of unrest in a destination can spike social media signals and trigger automated risk escalations that have absolutely no basis in what’s happening on the ground,” Dean said. “We’ve seen it.”

For many organisations, this is reinforcing the importance of structured validation frameworks. “Validation is the most critical element of using AI responsibly in risk intelligence,” said Ukassova. “Organisations are starting to use AI within a structured human-in-the-loop framework, where outputs are reviewed, cross-checked and contextualised by experienced analysts before being shared or acted upon.

“In an era of misinformation and synthetic content, disciplined verification and analytical oversight are non-negotiable.”

McIndoe said leading intelligence teams were increasingly relying on layered validation approaches rather than single-model outputs.

“They validate AI-generated risk data through layered source controls: source tiering, provenance checks, cross-source corroboration, geolocation and timestamp verification, human analyst review for high-impact decisions, and ongoing model performance monitoring,” he said.

He believes the industry is gradually moving away from treating social media as evidence. “The key operational change is that leading intelligence teams no longer treat social media as ‘evidence’; they treat it as a lead.”

McIndoe argued that the sector was now shifting towards what he describes as “AI-assisted evidence chains”, rather than AI-generated answers. “That is the more durable approach that will consistently produce better outcomes.”

The implications of better data gathering and analysis are also beginning to influence underwriting and pricing, particularly as insurers look for more granular ways to assess traveller exposure and operational risk.

Bex Deadman, Co-Founder of the TRA, said AI was helping organisations access and utilise information that previously sat unused. “There is a plethora of information out there that we are currently not harnessing, in the way that an AI engine can,” she said. “All these things combined set forth the stage for value-based pricing on evidence-based risk mitigation.”

Dean believes AI is allowing insurers to move away from broad underwriting categories that often fail to reflect real-world exposure: “Traditionally, underwriting has lumped travellers into large buckets – destination categories, age bands, trip types – and the result is cover that often doesn’t reflect the actual risk of what someone is doing or where they’re going,” he said. “AI is changing that by enabling genuinely bespoke underwriting decisions based on real-world, real-time risk.”

He pointed to the possibility of increasingly dynamic itinerary-based pricing. “We’re working towards a model where a client can upload their full itinerary – every destination, every activity, every date – and receive a quote in real time based on exactly what they’re doing and where they’re going.”

McIndoe believes the near-term impact is more likely to be improved segmentation and operational efficiency rather than fully dynamic pricing.

“I would think that the near-term effect is not fully dynamic retail pricing on every itinerary, but better risk differentiation and faster underwriting operations,” he said. “AI is making travel underwriting faster, more granular, and more operationally efficient, but not exempt from long-standing actuarial, conduct, and fairness requirements.”

Traveller communication is another area where AI is rapidly changing operational models. Rather than issuing blanket alerts to broad traveller groups, organisations are increasingly using AI to deliver targeted messaging based on location, itinerary, and traveller profile.

“Sending a targeted, timely alert to someone who is actually in or heading to an affected area rather than blasting a generic warning to everyone on your books is a much better experience for the customer,” said Dean. “People notice when something is relevant to them. They also notice when it isn’t.”

Martijn van der Voort, TRA Mentor at the TRA, believes relevance is central to whether communications are effective. “When a system factors in where someone is, what their itinerary looks like, and their individual risk profile, communication shifts from generic broadcast to something the traveller pays attention to,” he said.

Craxton said some organisations were already exploring broader employee engagement models built around AI-driven communication. “We are starting to see these as employee engagement solutions rather than just travel,” he explained. “They work to build trust with employees from day one of employment.”

“AI should be improving real-time traveller communication in four concrete ways: localisation, personalisation, channel orchestration, and timing,” said McIndoe. “It can translate content rapidly, tailor messages to itinerary and proximity, decide whether to use app push, SMS, email, or voice, and reduce latency between incident detection and traveller notification.”

Yet experts also warn that poorly configured systems risked overwhelming travellers with unnecessary alerts.

“Excessive alerts or poorly prioritised messaging can lead to alert fatigue, causing travellers to disengage,” said Ukassova. “Effective communication relies on relevance, clarity, timeliness, and, to some extent, restraint.”

Dean said desensitisation risks remained under-discussed: “If your system is flagging everything, people stop reading anything,” he said. “Bombard someone with precautionary notifications and you’re not keeping them safe; you’re just stressing them out.”

McIndoe told ITIJ that the real benchmark should be whether communication reduces cognitive burden rather than adding to it. “The benchmark is not how many alerts were sent; it is whether the right traveller received the right message at the right time with the right action,” he said.

Alongside the operational opportunities, experts also point to growing ethical and governance concerns as organisations hand more responsibility to AI-assisted systems.

Van der Voort said hallucinations and automation bias remained among the most serious risks.

“A confident-looking but wrong safety assessment or misread entry requirement is not an edge case; it is a foreseeable failure with direct consequences for the traveller and real legal exposure for organisations,” he said.

He also highlighted concerns around consent and data governance as organisations increasingly aggregate location, behavioural, and health information.

“These platforms aggregate location, health, and behavioural information at scale and the consent frameworks most corporate travel programmes operate under predate all of it.”

Craxton believes questions around compliance and data sovereignty will become increasingly important: “Which AI solution are the various platform vendors using? Is this within a customer’s own compliance? Where is that data hosted? What is being used to train them?

“When you think in context [with] General Data Protection Regulation (GDPR), customers need to know their data is not being shared or mixed with other customer data.”

Ukassova said responsible organisations were increasingly focused on governance frameworks and clearly defined escalation processes. “Mitigation requires strong governance: clearly defined use cases, regular reviews, bias testing, and explicit escalation paths to human decision-makers,” she said.

Dean believes AI is also changing the competitive landscape by lowering the barrier to appearing credible. “Smaller operators without the teams, infrastructure, or genuine expertise can now generate polished risk alerts, authoritative-sounding intelligence briefings, and confident LinkedIn commentary at the push of a button,” he said. “The result is a growing noise layer of AI-generated content that looks credible on the surface but has no meaningful operational capability sitting behind it.”

For buyers, he argued, accountability and operational capability will become increasingly important differentiators.

“Trusted brands who consistently produce verified, well-contextualised intelligence are going to become increasingly valuable,” Dean said.

McIndoe agreed that insurers and corporate buyers needed to become more demanding when assessing AI governance. “A prudent platform buyer should not ask whether a vendor ‘uses AI responsibly’; they should ask how the vendor governs, validates, audits, overrides, and recovers when AI is wrong,” he said.

For now, most experts agree on one point: AI is becoming an increasingly valuable operational tool across travel risk management, but not an autonomous replacement for human judgement.

Chloe Fox is an Editorial Assistant for Voyageur Group, joining in 2024. She writes for ITIJ and AirMed&Rescue, covering a range of topics including international travel and health insurance, medical assistance provision, and air medical transportation. Chloe holds a BA (Hons) in English and an MA in English Literature from the University of Bristol.