Tourist destinations have always planned campaigns, events, infrastructure and services on the basis of forecasts. Today they can do something more dynamic: observe what is happening, test a response on a limited scale, measure its effects and adjust the model before extending it.
The word sandbox belongs primarily to the fields of technological and regulatory innovation. In tourism, it is often more appropriate to speak of living labs, pilot programmes, tourism data platforms, testbeds, model demonstration projects or destination intelligence systems. These are different tools, but they share a similar logic: testing a solution in the real world before turning it into a stable policy or product.
This approach also changes the meaning of tourism marketing. Philip Kotler argued that the task of marketing is to manage the level, timing and characteristics of demand. Demand may be absent, latent, irregular, full or excessive, and each condition requires a different response.
In tourism, this means that the same destination can have too many visitors in one place and too few just a few kilometres away. It may suffer from extreme concentration at certain times while having unused capacity during the rest of the day. It may need demarketing in congested areas, synchromarketing to distribute flows more effectively and, at the same time, developmental or stimulational marketing towards territories and products that are still little known. Tourism literature does indeed use demarketing and synchromarketing as tools for addressing excessive demand and temporal or spatial imbalances respectively.
Seeing a Destination While It Changes
The first element of these models is the ability to observe.
Arrivals and overnight stays remain essential indicators, but they mainly describe what has already happened. New destination intelligence platforms can add mobile phone data, bookings, payment cards, transport, ticketing, online searches and other territorial signals.
Mobile data can, with appropriate statistical methodologies and aggregation, be used to estimate visitor presence, origin and distribution. Their use requires caution: they do not automatically represent the absolute number of tourists and must distinguish between visitors, commuters, transit movements and other forms of mobility.
The Korea Tourism Data Lab makes this precaution particularly clear. The Korea Tourism Organization platform uses telecommunications data to analyse foreign visitors by area and origin, but warns that figures generated from big data are primarily intended for trend analysis and should not be interpreted directly as absolute counts.
This is an important point: an intelligent destination does not simply accumulate data. It must understand what those data can actually demonstrate.
South Korea: When Data Discovers a Destination Before Marketing Does
South Korea represents one of the most advanced cases of tourism intelligence applied to territory.
The Korea Tourism Data Lab combines data relating to visitors, tourism consumption, geographical areas, international markets and different forms of tourism. The platform also offers analyses based on telecommunications data to reconstruct the territorial distribution of foreign visitors.
In July 2026, the Seoul AI Foundation used mobile phone foot traffic data and tourism reviews to analyse foreign presence in 16 locations across the capital. Myeong-dong was the most visited area, while Naksan Park, Achasan and Gwanaksan showed significant increases in international presence.
The value of data also lies in what it can reveal before a marketing strategy formalises it.
If international visitors begin spontaneously frequenting hiking routes, neighbourhoods or attractions that had previously remained marginal, a DMO can decide to study their potential, improve services and accessibility or avoid accelerating their growth too rapidly.
Here, developmental marketing finds a possible territorial application: recognising emerging demand and building a product capable of accompanying it.
However, it is important to distinguish fact from interpretation. The Seoul analysis documents visitor behaviour; it does not prove that every area identified is automatically transformed into a new tourism product. It is precisely this transition — from data to decision — that represents the most complex part of destination management.
Hakone: Testing Technologies Starting from Real Problems
Japan offers an even more explicit model of experimentation.
The Japan Tourism Agency has been conducting model demonstration projects dedicated to Tourism DX for several years. In 2025, it selected 25 experimental projects in three categories: regional regeneration models, generative AI applications and open data. In 2026, it also selected another 12 projects dedicated, among other things, to regional visitor dispersion, overtourism and increasing tourism expenditure.
The concept of a sandbox is particularly close to this reality.
The government officially refers to them as model demonstration projects, projects designed specifically to test models that may subsequently be replicated.
Hakone shows how this logic can be applied at the level of an individual destination.
The area receives approximately 20 million visitors a year despite having a population of around 11,000 residents. The local DMO has progressively developed digital tools to address very concrete problems, including transport congestion, staff shortages and difficulties encountered by international visitors.
One of the solutions is a digital map displaying information on transport congestion and local activities in order to encourage better circulation throughout the area. In January 2026, Hakone also launched an experimental AI-assisted restaurant booking service, created in response to a problem identified directly in the field: many foreign visitors did not dine at their accommodation facilities and had difficulty finding information or making reservations in multiple languages.
The value of the Hakone case lies in the process:
problem → data → hypothesis → solution → test → evaluation.
This is a very concrete form of tourism prototyping.
From Kotler’s perspective, Hakone combines at least two approaches. Synchromarketing can help distribute people more effectively across space and time; developmental marketing creates services capable of transforming existing demand into greater satisfaction and spending within the territory.
Singapore: Learn, Test, Scale
Singapore probably represents the most structured model.
The Singapore Tourism Board developed Tcube, officially organised around the Learn – Test – Scale framework.
The Test phase is described directly by STB as the transformation of real industry problems into technological pilots; the Scale phase is intended to extend cases that demonstrate useful results.
In 2026, STB and the Chinese platform Mafengwo tested multilingual AI robotic guides for one month at Sentosa and the Mandai Wildlife Reserve. The pilot, from 18 April to 17 May, offered storytelling and assistance in English and Mandarin. Sentosa itself is described by the partners as a testbed for new ideas.
Tcube, however, goes beyond the individual experiment. The programme connects tourism businesses, industry stakeholders and technology companies to develop pilots that can subsequently be scaled across the industry. STB also operates the Singapore Tourism Analytics Network, which provides datasets and visualisations useful for operators’ decision-making.
Singapore therefore demonstrates the transition that many destinations still struggle to achieve: moving a project from testing to scale.
The Problem Comes After the Test
Technology is often the simplest part.
An experiment may demonstrate that moving some visitors twenty kilometres away would reduce congestion in an iconic city. It may show that changing a timetable, proposing an alternative itinerary or promoting a nearby town would improve the distribution of demand.
Applying that conclusion, however, requires governance.
A tourist rarely perceives the administrative boundaries they cross. For them, a territory may be a single destination made up of cities, villages and landscapes connected by a thirty-minute train journey. For public administrations, those same places may belong to different municipalities, provinces, regions, DMOs and budgets.
Here one of the contradictions of overtourism emerges.
Many cities declare that they want to reduce visitor concentration, but genuinely promoting nearby destinations also means accepting that part of the expenditure, overnight stays, taxation and visibility will be distributed elsewhere.
The model would be relatively simple: demarketing in the congested location and stimulational marketing in the nearby destination, connected by a strategy of territorial synchromarketing.
Politically, it is much more complex.
Each administration inevitably tends to measure its own results. Redistribution, by contrast, requires a shared metric: not how many visitors each municipality has gained, but whether the entire territorial system has improved the quality of the experience, length of stay, expenditure and sustainability.
The Countermarketing of Competition
Countermarketing can also enter, in a highly attenuated form, into competition between destinations.
A city can position itself implicitly against a congested competitor through messages focused on greater tranquillity, authenticity, space, contact with the local community or more relaxed visiting times.
Technically, this use is closer to competitive positioning or selective marketing than to Kotler’s classic countermarketing, which was conceived to counter demand considered undesirable. Tourism literature does in fact distinguish countermarketing, demarketing and synchromarketing as responses to different states of demand.
In destination language, however, the principle remains useful: a competitor’s weakness can become the element against which a territory constructs its alternative positioning.
Who Has the Vision?
The real question, therefore, concerns who governs the process.
The DMO understands markets and visitors. The administration has regulatory tools. Universities can build methodologies and verify results. Operators transform ideas into services. Technology partners make visible phenomena that were previously difficult to observe. Tour operators and DMCs can verify whether a product works in origin markets.
The decisive part consists in bringing these skills together under a common vision.
The three cases show different levels of the same process.
Seoul measures and discovers emerging behaviours.
Hakone starts from problems and builds testable solutions.
Singapore institutionally organises the Learn–Test–Scale transition.
They are different models, and none of the three should be forced into a single definition of “tourism sandbox”.
Together, however, they show how a destination can increasingly function as a system capable of learning.
Data help it to see. Tests make it possible to experiment. Marketing offers different strategies according to demand. Governance determines whether what has been learned can become policy.
And this is precisely where the most difficult frontier lies: a destination can learn quickly from visitors; learning to share objectives and benefits with surrounding territories takes much longer.
One Destination, Many Different Forms of Demand
Experimentation takes on an additional meaning when connected to the states of demand described by Philip Kotler. A destination rarely presents a uniform condition. On the same day, it may have an overcrowded attraction, a little-known neighbourhood, a museum with available capacity and a nearby destination struggling to attract visitors who are only a few kilometres away.
Territorial marketing can therefore perform different functions simultaneously. The objective is not necessarily to increase arrivals, but to understand what type of demand exists in each place and choose the most appropriate response.
Very weak or non-existent demand — Stimulational marketing
Promote new places, periods or products.
Existing interest but an incomplete product — Developmental marketing
Transform latent demand into a tourism experience.
Demand poorly distributed in time or space — Synchromarketing
Redistribute visitors towards other times, seasons or territories.
Declining demand — Remarketing
Reposition a destination or renew its tourism product.
Balanced demand — Maintenance marketing
Preserve quality, capacity and satisfaction.
Demand exceeding capacity — Demarketing
Reduce, slow down or redistribute pressure.
Need to differentiate from competing destinations — Competitive positioning / lighter forms of countermarketing
Highlight tranquillity, space, authenticity or accessibility compared with congested competitors.
This matrix makes clear why experimental models are particularly useful. A city could apply demarketing in an iconic location while simultaneously using synchromarketing towards another area and stimulational marketing towards a nearby city. The visitor sees a single travel territory; public strategies, by contrast, still too often stop at administrative boundaries.
The most difficult step lies precisely in turning this interpretation into a shared policy. If relieving pressure on one destination means directing some visitors towards a neighbouring municipality, the distribution of expenditure, hotel stays and economic benefits also changes. A technically simple solution can therefore become administratively complex.
From Programmed Marketing to Dynamic Marketing
The next frontier is even more interesting: moving from strategies established months in advance to models capable of observing, forecasting and correcting demand much more frequently.
It is possible to imagine a destination in which mobile phone data, transactions, bookings, online searches, weather, event calendars, transport and attraction capacity continuously feed a predictive model. Artificial intelligence could estimate what is likely to happen in the following hours or days and suggest different interventions: promoting an area with available capacity, modifying a digital campaign, proposing alternative itineraries, providing information on congestion or intervening on pricing and bookings.
Such a structure follows a cycle:
detect → forecast → intervene → observe the response → correct → forecast again.
Academic research is rapidly moving towards this model.
A study published in 2026 in Information Technology & Tourism by researchers from Eurecat, Universitat Politècnica de Catalunya and Universitat Rovira i Virgili developed daily tourism-demand forecasts by combining card transactions, Google Trends, calendar data, weather and other variables. The model, also based on a Temporal Fusion Transformer, produces forecasts up to thirty days ahead for individual locations and even distinguishes the main origin markets. Barcelona, Girona, Salou and Lloret de Mar were included in the tests.
The research is significant because it shows the transition from traditional monthly national forecasting to a daily, local reading segmented by nationality. The authors themselves note that the literature has now reached, in some cases, hourly forecasts and even five-minute intervals.
The nature of the data is also changing. In 2026, another study proposed a framework integrating heterogeneous sources and machine learning for tourism-demand forecasting, precisely because weather, seasonality, policies and events can produce sudden variations that traditional models struggle to capture.
The most recent research is going even further. A paper published in 2026 proposes TDF-Agents, a multi-agent system based on Large Language Models in which AI agents collaborate in variable selection, data interpretation and dynamic demand forecasting. It remains a research project rather than an ordinary system used by a DMO, but it clearly indicates the direction in which these models are moving.
Another study published in Tourism Management uses multimodal data — online reviews, web searches, calendar data, weather and historical arrivals — to forecast demand in five destinations including Macau, Hawaii and Singapore. The results show how combining different sources can improve forecasting compared with relying on a single historical time series.
Research into movement networks is also particularly interesting. A 2025 study applied to Tibet uses big data on visitor trajectories and Markov chains to predict both the most likely next destination and, over the longer term, how visitors are likely to distribute themselves among different territorial nodes.
This evolution brings destination marketing close to a concept we might define as dynamic destination management. The predictive model should not autonomously determine tourism policy; it can, however, provide administrations with information that previously arrived weeks or months later: if we do not intervene, where are visitors likely to concentrate?
Sources
Japan Tourism Agency, Tourism DX Model Demonstration Projects, 2025–2026.
JNTO, Hakone DMO Case Study, 2026.
Singapore Tourism Board, Tcube – Learn, Test, Scale.
Singapore Tourism Board, AI Robodog Pilot at Sentosa and Mandai, 2026.
Korea Tourism Organization, Korea Tourism Data Lab.
Seoul Metropolitan Government / Seoul AI Foundation, analysis of tourism flows, 2026.
Eurostat, studies on the use of mobile positioning data in tourism statistics.
Kotler, P.; Levy, S.J., Demarketing, Yes, Demarketing, Harvard Business Review, 1971.
Kotler, P., Marketing Management.
By Daniele Di Stefano
