📈 economics

A Market Expansion Strategic Model in the Tourism Industry: A Case Study in a Portuguese Travel Tech Startup

This study proposes a machine learning-based strategic model for a Portuguese travel tech startup's market expansion, demonstrating that a weighted K-Means approach effectively identifies high-potential European cities like Zagreb and Warsaw while offering a broadly applicable methodology for the tourism industry.

Original authors: Guilherme Casal, Diogo Correia, Eugénio Rocha

Published 2026-08-31

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Guilherme Casal, Diogo Correia, Eugénio Rocha

Original paper licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

In the bustling world of modern travel, the difference between a thriving business and a failed venture often comes down to a single, difficult question: where should we go next? For companies that build technology to help people move around, finding the right city is not just about guessing which place looks nice on a map. It requires understanding a complex web of factors, from how easy it is to get from an airport to a hotel, to how much money the average worker earns compared to the cost of rent. This is the realm of data analysis, a field where researchers use computers to find hidden patterns in vast amounts of information. By grouping similar places together based on their characteristics, analysts can see which cities share the same potential for success. This approach moves beyond simple intuition, offering a way to make strategic decisions with a clearer view of the risks and rewards involved in expanding into new territories.

A team of researchers working with a Portuguese travel technology startup set out to solve this exact problem. They wanted to help the company decide which European cities were the best candidates for opening new services. Instead of relying on gut feelings or simple lists of popular destinations, they built a model to sort forty-five cities into groups based on how well they fit the company's specific needs. The team tested three different ways of organizing this data. The first method was a straightforward computer algorithm that grouped cities based purely on the numbers provided. The second method first simplified the data by removing unnecessary details, then applied the same grouping algorithm. The third and most innovative method brought human experience directly into the process. Before the computer started sorting, a group of five company experts—people who had spent years managing operations and business development—told the researchers what a "good" group of cities should look like. They used their knowledge to define the ideal mix of characteristics for a successful market.

The researchers then ran a massive search to find the specific combination of factors that would make the computer's groups match the experts' vision. They tested nearly 280,000 different ways of weighing the importance of each variable, such as flight traffic, the number of hotel listings, traffic congestion, and the cost of living relative to local salaries. The goal was to find a set of weights that would cause the computer to naturally sort the cities in the same way the experts had predicted. When they compared the results, the method that included the experts' input proved to be the clear winner. The other two methods, which relied solely on the computer's own logic, produced groups that were less distinct and often mixed together cities that were very different from one another. For instance, a purely mathematical approach might have grouped a high-cost capital city with a more affordable one simply because they shared a single trait, ignoring the broader economic reality that would make one a poor choice for expansion.

The winning model, which combined the experts' guidance with the computer's sorting power, created five clear groups of cities. One of these groups stood out as the most promising for immediate expansion. It contained cities that offered a powerful combination: high volumes of tourists, relatively low costs for doing business, and strong logistical connections. The analysis pointed specifically to cities like Zagreb and Warsaw as top priorities. These places shared the same strategic profile as the startup's existing successful markets, such as Lisbon, Athens, and Budapest. The model showed that these cities had the right balance of flight activity, accommodation supply, and economic feasibility to support the company's growth. In contrast, the model also identified groups of cities that were less suitable, such as those with extremely high living costs or poor transport links, effectively ruling them out for the next phase of expansion.

This study highlights a crucial lesson for the future of business strategy: the most powerful tools are often those that blend human insight with machine precision. While the computer could process thousands of data points in seconds, it needed the experts to tell it what those points actually meant in the real world. Without that human guidance, the computer's best mathematical solution still missed the mark, producing results that looked good on paper but did not align with the realities of the market. The researchers found that the best way to predict success was not to let the data speak for itself, but to let the data speak in a language that the business leaders understood. By validating the computer's groups against the experience of people who had already navigated these challenges, the team created a roadmap that was both statistically sound and practically useful.

The findings offer a clear path forward for the startup, suggesting that their next steps should focus on the cities identified in the high-potential group. However, the researchers are careful to note that this is a starting point, not a final guarantee. The model relies on historical data and current conditions, which can change with new economic shifts or global events. To be truly effective, the strategy will need to be tested in the field, with real-world results confirming the predictions. The study also suggests that this method could be adapted for other companies in the tourism sector, provided they bring their own experts into the loop to define what success looks like for their specific business. Ultimately, the work demonstrates that when technology and human experience work together, they can turn a chaotic sea of possibilities into a clear, navigable course for growth.

Technical Summary: A Market Expansion Strategic Model in the Tourism Industry

Problem Statement
The research addresses the critical challenge faced by tourism-oriented startups: the lack of effective, data-driven methodologies to identify and prioritize new markets for expansion. While many startups succeed in their initial markets, the transition to new cities often relies on intuition or basic comparisons, leading to sub-optimal decisions and inefficient investments. The complexity of the tourism sector—shaped by global trends, seasonal preferences, infrastructure, and macroeconomic fluctuations—requires a robust model to quantify the influence of interconnected variables (e.g., operational costs, accessibility, tourist volume) on potential market success. Existing literature often utilizes clustering for operational optimization or visitor profiling but frequently lacks the integration of domain expert knowledge required for strategic market prioritization.

Methodology
The study proposes a structured, six-step framework that combines quantitative machine learning with qualitative expert validation to support strategic expansion decisions. The methodology was applied to a Portuguese travel tech startup, analyzing 45 European cities (10 existing "supervised" markets and 35 potential "unsupervised" targets).

Data Collection & Preprocessing: The dataset included seven key variables derived from public and proprietary sources:
- Flight Activity: Composite of international flights arriving before 11 a.m. and departing after 3 p.m.
- Outbound Tourism: A proxy for demand and city attractiveness.
- Economic Feasibility: A composite metric of Average Annual Salary multiplied by Cost of Living.
- Logistics: Airport-to-center distance and average traffic time for 10km.
- Market Maturity: Number of short-term rental listings.
- Consumer Interest: Google Ads Cost Per Click (CPC) for the keyword "Luggage Storage."
  Data underwent exploratory analysis, normalization, and correlation checks to address skewness and multicollinearity.
Modeling Approaches: Three distinct clustering strategies were implemented and compared:
- Approach 1 (Standard K-Means): Applied directly to standardized data. The optimal number of clusters ( $K=5$ ) was selected based on the Silhouette Score.
- Approach 2 (PCA + K-Means): Principal Component Analysis (PCA) was used to reduce dimensionality (retaining 4 components explaining 94% of variance) before applying K-Means.
- Approach 3 (Expert-Weighted K-Means): This novel hybrid approach utilized a focus group of five company experts to pre-define the expected clustering of the 10 existing markets. A combinatorial grid search ( $6^7$ combinations) was conducted to find variable weightings that allowed the K-Means algorithm to reproduce these expert-defined groupings.
- Approach 4 (Automated Optimization): A variant using Optuna (Tree-structured Parzen Estimators) to automatically optimize weights was tested but ultimately rejected due to a lack of strategic interpretability.
Evaluation: Models were evaluated using the Silhouette Score for cluster cohesion and separation, as well as qualitative assessment by the expert focus group regarding the strategic relevance and interpretability of the resulting city clusters.

Key Results

Performance Metrics: Approach 3 (Expert-Weighted K-Means) yielded the highest Silhouette Score of 0.485, significantly outperforming Approach 1 (0.2742) and Approach 2 (0.3011). Approach 4 achieved a score of 0.38 but failed the qualitative validation.
Strategic Alignment: While Approach 2 improved technical metrics over the standard K-Means, its clusters were often heterogeneous (e.g., grouping Paris and Madrid together), limiting strategic utility. Approach 3 successfully grouped cities into semantically interpretable clusters that aligned with operational reality.
Variable Importance: The expert-weighted model revealed that Total Flights (Non-domestic), Average Annual Salary × Cost of Living, and Traffic were the most influential variables in defining high-potential clusters.
Expansion Recommendations: Based on Approach 3, Cluster 4 was identified as the highest priority for expansion, characterized by high tourism volume, relatively low living costs, and logistic feasibility. Cities such as Zagreb and Warsaw were specifically identified as high-potential targets, grouping them with the startup's existing successful markets (Athens, Budapest, Lisbon).

Significance and Claims
The paper claims to present a novel methodology that bridges the gap between unsupervised machine learning and practical business strategy. Its primary contribution is the demonstration that integrating domain expert knowledge into the clustering process (via variable weighting) yields superior results compared to purely data-driven or dimensionality-reduction approaches.

The study asserts that:

Hybrid Methodology is Superior: Combining quantitative clustering with qualitative focus group input creates models that are not only statistically robust but also strategically actionable.
Interpretability Over Automation: Automated optimization (Approach 4) can produce mathematically decent scores but may ignore critical business variables (e.g., assigning zero weight to tourist volume), reinforcing the need for a "human-in-the-loop" approach.
Scalability: The proposed framework is designed to be flexible and replicable, allowing other tourism startups to adapt the model to their specific contexts, data availability, and business models to reduce uncertainty in market expansion.

The authors maintain a modest stance, acknowledging limitations regarding data quality, the specific geographic scope of the study, and the reliance on historical data, noting that future work should include empirical field validation and the incorporation of time-series data to capture evolving market trends.

Technical Summary: A Market Expansion Strategic Model in the Tourism Industry

More like this