TTI Autumn Conference 2025

Sponsored by Inspiretec

The Future is Agentic: AI and Travel Tech Reimagined

A seismic shift in travel looms. Agentic commerce is poised to redefine not just how we book, but why and when we travel. AI agents will soon handle every step of a travel booking, from sourcing optimal itineraries to flagging unexpected delays in real time. This evolution won’t simply streamline operations; it will demand new standards to ensure our AI advisors abide by ethical data usage and clear consent, strategic partnerships between technology and finance players, and human oversight. 

For travel and tech leaders, the challenge is twofold: to build seamless, secure AI experiences and to champion the human insight that gives them purpose. A packed room of delegates at the TTI Autumn Conference 2025 (#AI4Travel), staged at Holiday Inn London – Regent’s Park on 17 September, and TTI members watching the livestream were eager to learn how from five expert speakers…

The Future of Tech at loveholidays

Mike Jones, Chief Technology Officer at loveholidays, emphasised the importance of laying firm foundations for Agentic AI. He showed screenshots of their Claude desktop is connected via an MCP (Model Context Protocol) to all the trading and operational data loveholidays has. It analyses what's available to answer the question, writes the queries, corrects itself when it's wrong and summarizes the results.

‘The possibilities are huge for this,’ said Jones. ‘And decisions we made when I started at loveholidays in 2019 helped us get here. Decisions that seemed completely unrelated to AI — because frankly, AI wasn't even on our radar yet.’

First, the company invested heavily in technology: ‘Our contact centre ran on ancient hardware, with CRT monitors covered in laminated passwords stuck to them. Broken technology kills growing businesses. So we had to rip them out and start again.’

Jones’s team created a platform that could adapt quickly, and they became obsessed with measuring to prove it was working (‘We believed if you can't measure it, you can't improve it’).

Choosing the right people was also key: ‘We deliberately hired people who see possibilities, not limitations. People with curiosity and problem-solving ability, not just technical skills. That helped us kick on with AI.’

Looking back, Jones can identify three phases of AI at loveholidays. The first was in 2020–21 when the Covid pandemic struck. The call centres were hit with a year’s worth of enquiries in one month. His team created ‘Sandy’ focused on one task – giving customers their money back. It worked and led to the development of much more sophisticated Sandy of today which can deal with 500 different tasks, speaks three languages, and resolves 50% of chat enquiries and 23% of phone enquires. ‘We get old people saying I can’t believe I’m being helped by a robot,’ smiled Jones. ‘Our contact centre costs have been reduced by 20% and invested in other areas.’

Phase two (2022–24) was about optimising the customer experience: ‘Our team manually managed 2,000 hotels. The business needed 50,000-plus in three languages. AI stepped into help. ‘We now process 10 trillion packages daily, 1.8 quintillion last year – more than grains of sand on all beaches, to help customers find that one perfect holiday. At this scale, humans alone can't cope.’

Jones showed a sample page of AI-generated content (text and images) for a hotel: ‘When we launched in Germany, the translation cost was £250k. But with AI, we recreate all hotel content across five countries for a few hundred pounds.’

Phase three (2024–present day) has seen AI layered across everything. Six years of foundations converged with the ability to query across all their data… but, plot twist! Jones revealed that AI-generated numbers he’d shown on a previous slide had all been a hallucination! ‘Claude seemed so sure, but it was wrong – that’s pretty scary!’ He also mentioned how Sandy can sometimes randomly spout nonsense, for instance, randomly inserting the word ‘wanji’ (‘yes, please’ in Ugandan) into chats!

This led to a change of approach to a more considered AI rollout using sophisticated workflow management, understanding that expertise in AI means knowing not just what it can do, but also what it can’t.

Jones sees 2025 as an inflection point between the old world dying and a new one struggling to be born. Broken systems, gut-feel decisions and maintenance workers are being replaced by robust platforms, measured outcomes and builders working in partnership with AI. 

‘AI is only going to get better and the future is humans and AI each doing what they do best,’ he argued. ‘We’ve got the systems in place, the people in place and the workflows in place to hand over to Agentic, if and when it becomes a reality. We accidentally built a great foundation for AI. Now we're intentionally building the perfect partnership with it.’

Digital Darwinism in Travel: Evolve Fast or Get Cancelled

Ian MacArthur, Global Chief Strategy Officer at Remarkable, started by getting audience members to hold their hands up like they were on a rollercoaster. He then read out four statements – if one wasn’t true for your business, you had to put one hand down.

• My company is digitally mature.

• I can deliver real-time personalisation (across every channel).

• My martech stack doesn’t slow me down. 

• My team isn’t drowning in content chaos.

Only a minority still had a hand or two raised by the end, indicating a realistic attitude not prevalent across the travel industry. According to MacArthur, ‘85% ‘claim digital maturity, but 88% lack a single customer view – that is a serious contradiction. ‘You’re not digitally mature just because you have a mobile app. If you can’t adapt in real time, you’re not transforming – you’re treading water.’

He talked about the ‘escape economy’, worth around $10 trillion last year, which has grown significantly since Covid as more people seek to escape their day-to-day lives. 

‘All incredibly buoyant news for everybody here, but if you don’t have that digital intent are you really capturing that demand?’ he asked. ‘Welcome to “Digital Darwinism”, survival of the digitally fittest.’ 

And to survive, you need to meet the expectations of travellers, who evolve faster than brands. This means effortless booking, check-in and updates, delivering what they want in real time, plus authenticity, sustainability and personalisation. This personalisation can be anything from on-screen messages at the airport gate that the overhead bins on the plane are full to huge billboards above the road from the destination airport (MacArthur showed one such successful campaign by Remarkable client Atlantis in Dubai, where Mastercard offered 25% resort-side discounts).

MacArthur identified three main killers in travel customer experience. First, content chaos: ‘48% can’t keep up,’ he said. ‘One of the biggest areas we work in currently is content governance. Accelerations in AI which are really positive turn some organisations into the Wild West.’ 

Second, technical debt: ‘Only 7% have zero debt or really fast upgrades. As Mike talked about, organisations that are on legacy stacks will never be able to embrace AI.’

Lastly, personalisation paralysis: ‘81% can’t dynamically personalise, and sometimes it’s just because they make it more complicated than it is. You can go really small with personalisation and that can have quite a big effect. As Mike showed, you can generate on-brand content through AI, so there is nothing really holding you back.’

The reason for that of course is the sudden proliferation of GenAI. He pointed to the incredible growth of ChatGPT from zero to one million users in a day, and 100 million users in two months.

‘We talk about digital maturity and four years ago the number or organisations across all sectors in the UK that were digitally mature was about 27%,’ he said. ‘Month on month, watch that number get smaller. The technology is outstripping the organisations and everyone needs to keep up because the customer is charging ahead.

He reported that Remarkable’s customers have seen their organic search fall by 50% in the past six months. This is because ‘AI overviews’ remove the need to actually go to their websites. ‘We’re doing lots of work so that when a Large Language Model (LLM) is looking for info that touches on your business, that your brand is as high up that as possible.’

MacArthur reeled off a list of direct impacts AI is having on travel. For example, over three-quarters of travellers say that AI improves their planning, and 41% of businesses say they now have the budget in place to do this properly. Over 90% of airlines have a really active Customer Data Platform (CDPs) to personalise customer experiences (‘that’s really positive’). ‘If you don’t evolve, somebody’s already done it,’ he concluded.

MacArthur was joined by Remarkable co-founder Paul Stephen and Mike Jones for an audience Q&A before the coffee break. This included a question from Justin Morshead from Travel Ledger about using AI for pricing.

‘For us AI enables the number of data points to affect the pricing,’ MacArthur replied. ‘Traditionally, with dynamic pricing you might have, at best, the time of year, the inventory that had already sold, an event in the calendar that will draw people to the area. AI enables you to have hundreds of data points. The technology can help you go wild with it.’

‘We have dynamic pricing with our own pricing engines,’ said Jones. ‘We’ve got our own Machine Learning models that feed that pricing in real time. About 4,000 requests go through it per second and each is processed according to a series of rules we’ve set, balancing margin versus conversions. So, for us it’s pre-made Machine Learning models that are integrated into our systems.’

Stephen added: ‘What we’re seeing on flights is that if you ask ChatGPT what the best time fly or best value flight is on a particular date, Google is going to tell you. So you don’t even have to go through to an aggregator, and the choices people are making are moving further up the funnel.’

Does Agentic AI have a Place in Corporate Travel?

The prize for the speaker with the grandest job title went to Matthew Newton, CWT’s Chief Architect & VP Enterprise Architecture, Technology Strategy & Planning, no less. Newton opened the second session with his insights on the potential role AI has to play in corporate travel. 

‘Agentic AI is not the answer to every question,’ he began. ‘In fact, it’s the answer to very few questions at the moment. It has magical qualities, but those are not enough to be determined as a silver bullet. But it has given us a platform to imagine a better future in a way that blockchain simply didn’t.’

While preparing for a meeting abroad, without you lifting a finger, your AI agent will review your calendar, find an itinerary that balances cost and convenience, checks it fits company policy, prebook a car to the airport and flag potential issues (like a rail strike) and adjust accordingly. That’s the promise of Agentic AI anyway… ‘The reality is that Agentic AI is not going to be one entity that does everything for you, but a series of small bots, small services that link together to look after what you are doing,’ he said.

Quite a few of the elements are already in place. For example, autonomous disruption planning (i.e. monitoring prices and changing the booking to get a better price), smart itinerary planning and seamless expense automation (building expense reports as the traveller spends). ‘The trick is how we link these things together.’

Newton likened Agentic to a mirror because ‘it can only do what it is told to do and use the information it has access to – it just stays in its lane. You have to ask whether your company’s ecosystem has the building blocks in place to take this forward.’

These building blocks include policy clarity, data and systems that talk to each other and have something meaningful to say, and governance and oversight.

Newton said he’s been asked more about AI than anything else in his career, ‘because everybody is using it. And everyone is interested, excited or scared about what it has to offer.’ He talked about the complexities of the corporate travel ecosystem (involving suppliers, Travel Management Companies (TMCs), governments and regulators, etc) and the difficulty of adding Agentic AI into the mix. ‘It’s a very busy picture, so we need to think about how we join all of these things up,’ he said.

Over 90% of travel managers have implemented AI in some form – ChatGPT, machine learning or similar – but half of them say they have implementation issues, where systems aren’t joined up or functions within organisations aren’t joined up about what they want to do. 

The customer demand is there, though. Newton quoted a recent Amadeus survey of 6,000 travellers and half of them said they’d be quite happy for AI to make their booking for them. However, a recent Financial Times study showed the success rate of ChatGPT-4 booking a complicated travel itinerary was very low – just one in every 160 booking attempts. ‘It will improve, but the technology cannot yet achieve what we’re able to do as humans, by some distance,’ he said.

To catch up with demand, CWT is starting with use cases that offer quick wins – for example, price optimisation, where a journey may be rebooked at a different price when there is enough time to inform the traveller of the change of itinerary.

‘With AI, we need to get governance involved at a much earlier stage,’ he said. ‘In my experience, the governance is not just internal, it’s making sure the client’s posture on AI is being adhered to. Some want the ability not to use AI on particular bookings. Setting those guardrails is very important.

He also emphasized the need for common standards: ‘Organisations like IATA are talking about this, and it will come in time. Not exciting, but important.’

AI costs have fallen dramatically to a fraction of what they were just two years ago, the barriers to entry are much lower (‘It is a democratically available technology’). But the return on investment is a mixed picture, currently. Newton noted that the savings from GenAI Copilot implementations have netted around eight minutes of time savings per day, not a great productivity gain. But AI is giving customer service benefits i.e. chatbots do give you an answer rather than leave you hanging on the phone.

‘Agentic AI is augmenting,’ he said. ‘It is going to assist our colleagues; it’s not going to replace the human touch. We need people to be cross-checking AI because It hallucinates. The speed gives people confidence. It’s a bit like the SAS, whose basic tenet is that if you look like you belong and act confidently, people will just let you get on. The problem with that analogy is that the SAS is about causing chaos. Not necessarily what we want!

‘Agentic AI will have an impact in the corporate travel space, he summarised. ‘It will have an impact in every space. But it’s going to take more than just one or two people. If you want to go fast, go alone, but if you want to go far, you have to go together. The trick with Agentic AI is how many people we can get on board to go on the journey.’

The Role of AI in the Tours and Attractions Industry

Dr Douglas McIlwraith is the Director of Data Science & Machine Learning at Viator, part of the TripAdvisor group. Viator is the world’s largest online marketplace for tours, activities and attractions, offering over 300,000 travel experiences on its website. 

McIlwraith reiterated the growth of the escape economy, with statistics showing that over 78% of millennials would spend money on experiences over physical goods. Bookings are expected to grow in value from $270bn in 2024 to $365bn in 2028.

‘The experiences industry is growing fast, especially online, but there is room for more growth, he said. ‘At around 20%, online penetration in Experiences remains approximately half of other travel sectors like hotels. We’re on a mission to create an experiences marketplace that’s wildly compelling to travellers, essential to operators, and simple for all.’

Viator Machine Learning is aligned to this mission in the three main ways. One, sorting which products are presented to users and in what order. Second, extracting and presenting accurate product information to create traveler confidence. Third, enabling travellers to search against a broad set of specific needs and wants, either in the app or on desktop. 

‘But wait, isn’t the future supposed to be Agentic?’ he said. ‘Can’t Agentic AI book my holiday for me? Maybe it can, but booking a holiday is a messy process. Agentic is not quite there yet for people to trust it.’

McIlwraith believes marketplaces, such as Viator, will remain as relevant as ever, but the challenge is move from being a supply aggregator to a knowledge aggregator. Large Language Models (LLMs) play a pivotal role in this transformation. 

He talked about how Viator are leveraging LLMs to enhance product discovery, and overcoming the complications involved in that. For example, if they have a combo package with both a city highlights and boat tour, depending on the way the supplier has structured the tour, the LLM model could interpret it as city highlights or boat tour. So Viator asked their supply and content team to define what the labels are for a bunch of products.

‘We created the prompts and asked ChatGPT to see what happens,’ he recalls. ‘It was really time-consuming and not scaleable. So, instead, we tried taking the product description and running it through an automatic prompt engineering framework, where we created a load of prompts, then evaluated them, improved them, and repeated until our output labels are good. This resulted in a more scaleable solution.

‘If we task our supply and content team with providing great labels, and our Machine Learning and engineering teams with making great processes using to get those labels using those tools, then everyone wins.’

McIlwraith then shifted gears to talk about reviews. He showed three five-star reviews and posed the question: ‘Are these reviews the same value to you?’ The first just said ‘It’s a great tour’, while the third added detail: ‘It’s a great tour! Our guide, Michael is knowledgeable and friendly. Food is must-try!’

‘The review content tells us the “why” behind the rating,’ he said. ‘Relevancy, recency and content are the most important aspects of a review. So we gave our Machine Learning team a challenge: “How do we decide which is the most important reviews?”’ 

They did it with Aspect Based Sentiment Analysis using LLMs. The LLM extracts all aspects of the review (in the example above, overall tour experience and specific comments about the guide and food) to create a sort score that ranks a review’s importance. 

‘LLMs allow us to do this super-quickly and easily,’ said McIlwraith. ‘It also allows us to provide an overview of reviews (like on Amazon), display and filter themes and show product highlights, helping customers to make decisions more quickly.

‘Whether a traveller is booking themselves, using a travel agent or, in future, using Agentic – an automated agent, we’re continuing down this path of creating great experiences for our customers. 

Viator likes Agentic AI’s application in customer service, because there are repetitive requests (for instance, verifying customer refunds) and well-defined rewards. The rationale for requests can be open-ended and can involve task planning – ‘Agentic AI is really good for this,’ said McIlwraith.

‘LLMs have really revolutionised our work, but how we use them is really important,’ McIlwraith concluded. ‘Setting them up in businesses is really hard. Getting people to do the right tasks is really hard. We need to focus on providing a great experience for our traveller, and Agentic has its place.’

Agentic AI – Into the Future

Conference moderator Paul Richer introduced the final speaker Andy Owen Jones, Co-founder & Managing Director, SMARTSEER, as ‘one of the industry’s pioneers’. He humbly disagreed, but admitted he had worked with Machine Learning for a longer time than most.

Jones company has just been bought by GIATA, one of the main sources of verified hotel data in the world and the provider of TTI codes. 

‘This is an interesting acquisition,’ he said, ‘because it positions both companies in line for where we see the market going. One of the areas I expect to see Agentic AI used is to create reviews of experiences that people haven’t been on. Very easy to do, but verification is going to be really important.’

Jones showed a screenshot of the AI overview he got when he Googled ‘TTI Conference 2025’. It referred to a TTI (Teaching, Translation and Interpreting) conference in Poland, as well as the one we’re attending. ‘You have to read down, because the first thing you get is not necessarily what you want.’

He harked back to 2005 when he was working on Machine Learning at Amadeus at a time when Sabre were taking market share from them. There was a big internal argument within Amadeus about dynamic packaging. One view was that we must connect flight data to hotel data and there must be live availability. ‘They kept saying, this is all about accuracy,’ he recalled. ‘We said, “No, no, no, this a speed game.” … Google redefined search and speed is the most important thing. But Google won search because they got speed, comprehensiveness and relevance right, and it’s easy to use.’

In Jones’ view, this allowed travel agencies to ‘outsource’ comprehensiveness and relevance to Google and compete on speed. ‘When we [Amadeus] came to the UK market in 2006, 2007, it was entirely defined by live dynamic packaging. The defining battleground was around speed, and loveholidays was always going to be the winner in the UK, because they were much faster’

But Large Language Models have changed the equation – temporarily, at least. “When you use a chatbot, and it says, “Taking a little longer to think”, you don’t think, “Oh, that’s rubbish” because we believe the more it’s thinking the better it’s going to be! LLMs are relatively slow, but the improvement of speed is coming.’

Jones then pondered what problems to address with LLMs, because it’s more suited to some tasks than others. For instance, he said, ‘LLMs are not particularly good right now at being trained on dynamic data, Machine Learning is a much better tool.’ 

And at the moment, Agentic AI can’t be trusted to provide end-to-end itinerary generation: ‘If there is room for hallucination at any point of your trip, you’re not going to trust any of it. But it will improve,’ he said.

‘The more people use ChatGPT and Copilot at home, the more demand to bring it onto your site,’ he added. ‘It’s an inevitable transition, but the question is when and how to organise yourself for it.’

He explained why travel is hard for GPTs – for example, they are not specifically designed for sorting, ranking or filtering large data sets and they are designed for probable outcomes, not precision. ‘However, they are amazing as an interface, they can analyse data very quickly and they can be very good for relevance. They are changing the way search and discovery is happening, whether we like it or not, so we need to be ready to interface with them.’

Jones recommended a continuously evolving user interface, moving towards chat, but not neglecting the power of visual and feature comparison, because people need to see and compare things before making a decision. ‘I don’t think we’ll see the end of Viator and loveholidays, because you still need a full marketplace stack to make things easily searchable,’ he said.

‘Disruption is already happening for search and discovery,’ he said in summary. ‘You can no longer outsource relevance to Google, because companies are going to start bringing that in house or find partners than can help them to do that. If you are a tour operator, I think the power of ATOL will become much more important. Why not shout about your biggest asset that we protect people in the event of disruption. And loads of disruption can start to be handled by Agents.

‘Unique content, powerful brand and great user stories will count more and more highly, and I would encourage you to look for verified stories, not random ones generated by some kind of Agent. You still need to master your brand, know what sets you apart and be good at what you do.’

The three speakers from the second session joined together for a Q&A before the close of the conference. Among the queries from the audience, Jonathan Wall of Xeinadin travel accountants, asked whether Agentic AI will make it harder for small or new operators to compete.

‘The simplistic answer is yes,’ replied Newton. ‘As Agentic AI is to become more prevalent, the rules and regulations around how it will be used are going to tighten up. If a new operator is going to be responsible and have longevity, they are going to have to operate in a responsible way. I think that will be the hurdle rather than just getting started.’

‘I spoke to a smaller operator last week who’d lost about 60% of his Google traffic since the end of last year,’ said Jones. ‘He was getting quite a lot of queries through ChatGPT and those are converting 10 times as much, but he hasn’t got enough to make up for what he’s lost through Google, and he doesn’t know how to replace that lost traffic. To be honest he was quite scared.’

‘If you are provider of a niche service, then maybe you’ll see more traffic,’ McIlwraith said. ‘But the traffic is still going to be directed to the big players who are trusted, have all the information on their site and strong reservations systems. We’ll have to see who wins in this battle.’