Decomposing hospitality distribution channel market share

Hospitality distribution

The adoption of contemporary ICT and the Internet has revolutionized various facets of life, initiating a new marketing landscape and reshaping interactions among market participants (Aamir et al., 2025; Amin et al., 2021). Carr (2000) believed that ecommerce would increase the mediation of all industries rather than shorten the distribution chain or shift it towards direct distribution. In the particular case of hospitality distribution, the emergence of the Internet resulted in a new paradigm (Buhalis and Law, 2008; O’Connor, 1999; O’Connor and Frew, 2004; Xu et al., 2014). Some scientists considered that these changes would lead to disintermediation, which was later acknowledged not to be so, and the term reintermediation started to be used instead (Kracht and Wang, 2010). The term disintermediation referred to the disappearance or significant reduction of traditional intermediaries (global distribution systems - GDS - and travel agencies) in order to increase direct sales through the Internet (Law et al., 2015; O’Connor and Frew, 2002; Thakran and Verma, 2013). The term reintermediation is used to refer to the process in which traditional intermediaries that have been disintermediated first are reasserting their intermediary role (Granados et al., 2008). Other authors refer to reintermediation as the adoption of new intermediaries or the emergence of new intermediation functions, resulting in a longer value chain (Flecha et al., 2017). Finally, others consider that the process of reintermediation coincides or is referred to as SoLoMo (Social, Location, and Mobile, see Thakran and Verma, 2013). The emergence of these new intermediaries rendered hospitality distribution even more complex (Beritelli and Schegg, 2016; Chen, 2024; Lei et al., 2019). Such complexity came not only from a multichannel distribution system from the perspective of the supplier firm (Garcia et al., 2022) but also from new business models that increased competition between hoteliers and suppliers (Chang et al., 2019; Iazzi et al., 2017).

Development of hospitality distribution channels

Distribution channels play an important role in hotel business strategies, profitability, and customer supply and demand (Deyá-Tortella et al., 2022; Kracht and Wang, 2010) and constitute one of the core activities of hospitality revenue management (David-Negre et al., 2018; Guillet and Mohammed, 2015). With the expansion of the World Wide Web during the 1990s, hoteliers started allocating resources to develop both direct and third-party online distribution channels (Amaro and Duarte, 2013; O’Connor, 2001, 2003; Sahay, 2007; Toh et al., 2011) due to the creation and transformation of distribution channels (Berne et al., 2012; Myung et al., 2009).

The multi-channel strategy is a common practice in hospitality distribution as it increases market exposure and potential sales (Beritelli and Schegg, 2016; Cazaubiel et al., 2020; Mei, 2014). It can be even more beneficial for small hotels as they have lower popularity, capital, scale, and market exposure (Ennis et al., 2023; Toh et al., 2011); but determining the right portfolio of distribution channels has been a challenge for both hoteliers and researchers (Beritelli and Schegg, 2016; Kracht and Wang, 2010; O’Connor and Frew, 2002; Poór et al., 2019). Each distribution channel contributes to the hotel channel mix and runs simultaneously and competes with other channels (Tan and Dwyer, 2014).

Hospitality distribution utilizes two primary categories of channels: direct booking channels, which involve no intermediaries and, consequently, no commission; and indirect booking channels, where intermediaries play a role, leading to fees (Poór et al., 2019; Šimunić, 2021). Since the emergence of the Internet, offline and online sales are feasible, whether the channel is direct or indirect (Poór et al., 2019). Each channel has different associated distribution costs and provides different benefits (Dolasinski et al., 2019). Transaction costs are normally lower in direct than in indirect channels (Helsel and Cullen, 2005). In the pre-Internet era, intermediaries, particularly travel agencies, received a standard commission, typically around 10%. However, with the widespread adoption of the Internet, online intermediaries (OTAs) did not accept such low commission due to their substantial investments in infrastructure, brand development, and marketing (Flecha et al., 2016). Traditional channels such as call centers and GDS have significantly lower costs than OTAs (Law et al., 2015). The brand web channel is “10 to 15 times cheaper than the OTA channel and 4 to 10 times cheaper than the GDS” (Beritelli and Schegg, 2016: p. 72) but proprietary websites do not have the same marketing power and hotel exposure as the much more expensive OTAs (Dolasinski et al., 2019; Poór et al., 2019).

The rapid evolution of the distribution paradigm with the appearance of the new intermediaries has created tensions with hoteliers regarding the configuration of the offer, mainly due to the bargaining power of the former (Chen and Chen, 2024; Wang, 2025). This situation leads to distribution practices and the negotiation capacity of hoteliers not being as efficient as they could be or as they used to be (McLeod et al., 2018). OTAs have a very strong influence on hotel reservations compared to the other players in distribution (Andriotis and Paraskevaidis, 2021; Guo et al., 2024; Ivanov et al., 2015; Wong and Nasir, 2022). Some of the big OTAs forced hoteliers to use price parity (Xue et al., 2020). The imposition by a platform of price parity clauses prevents suppliers from independently determining prices across various distribution channels. European governments have been investigating OTA platform parity clauses (Cazaubiel et al., 2020; Jiang and Erdem, 2018), and some countries such as France, Germany and Austria have prohibited them (Ennis et al., 2023).

Due to the evolution of technology and the changes in the hospitality distribution ecosystem, the management of the different booking channels has become fundamental for the success of any hotel (Dolasinski et al., 2019). Understanding distribution channels is crucial for the success of revenue management strategies and effective distribution, enabling hoteliers to avoid dependency on a single channel, optimize channel exposure, and increase direct market share (Ibrahim et al., 2022).

Distribution channel market share

Market share of distribution channels has been explicitly studied previously by D-EDGE Hospitality Solutions (2020); and Schegg (2018, 2020), understood as the portion of total sales by distribution channel within a specific period and geographical area. D-EDGE Hospitality Solutions (2020) measures the distribution market share in revenue evolution over almost 4 years in 13 countries. The data used in the study comes from the hospitality management systems of almost 4,000 hotels, but this white paper excludes offline channels and GDSs while Schegg (2018, 2020) measure the market share in the evolution of room-nights over 5 years in more than 20 countries. In Schegg’s studies, data is extracted from a questionnaire rather than from the hospitality systems. Both authors seek to determine the market share evolution (one in terms of revenue and the other in terms of room-nights) of the studied channels in their investigation. Neither breaks down market share into revenue, room-nights, and bookings nor considers the impact of length of stay and price.

Emphasizing the previously mentioned fact that there is limited research containing hard data from the hospitality systems, we have identified other studies where such data exists but has not been examined in terms of market share and evolutionary trends or broken down into distribution channels. We note six articles where distribution channel percentages are present but not analyzed through time; Bigne et al. (2021) have hotel data from 27 consecutive months but analyze only the aggregated data and focus on the advance booking window across the different channels. Chalupa and Petricek (2020) use data from only one hotel over two consecutive years. The authors only study three online channels and they focus on the average daily rate and booking window. Lei et al. (2019) Study 1 year of channel share data and focus on the channel mix to enhance revenue per available room. Masiero et al. (2015) and Masiero and Law (2016) also study only 1 year of data and take only some online channels into consideration. Poór et al. (2019) study only 1 year of data in one hotel and the channel breakdown is generically defined as indirect versus direct and online versus offline.

There are two articles in which channel evolution data can be found, but they do not delve into studying channel trends in hospitality distribution and market share breakdown; Cazaubiel et al. (2020) include 4 years of distribution data but study substitution only among online channels, comparing the years when Expedia was available in Oslo. Deyá-Tortella et al. (2022) study almost four consecutive years of almost 300 hotels in the Balearic Islands but only look into the intermediated channels.

Case study data

The data used in this case study comes from a global hotel chain comprising over 800 hotels worldwide. This includes nearly 100 properties in Africa across around 20 countries, over 70 properties in the Americas (excluding the U.S.) across at least 17 countries, over 250 hotels in the Asia-Pacific region in at least 20 countries, and approximately 400 properties in Europe spanning at least 40 countries. The chain has over seven brands that target a diverse range of traveler preferences and budgets. Its portfolio primarily includes airport, city, resort, and convention hotels, with city hotels contributing significantly to the core business. The average size of the hotels is between 150 and 200 rooms.

Data collection was carried out through the company’s hospitality systems, in particular the property management system and the channel manager, in November 2021. This interconnected system processes and stores aggregated data from the company’s global distribution including room availability, reservations, check-in and payments, by channel. The case study utilizes yearly aggregated data spanning from 2013 to 2020. The subject company uses five distinct distribution channels (Table 1).

OPEN IN VIEWER

Table 1.

Distribution channels in the chain case study.

Channel	Examples
Hotel direct reservations (direct)	Reservations made directly with a hotel, such as walk-in reservations, group reservations, direct calls to the hotel, etc.
Brand web reservations (brand web)	Reservations through the chain’s website
Voice reservations (voice)	Reservations through the chain’s call center
GDS reservations (GDS)	Reservations made by the traditional travel agencies using Sabre, Amadeus or Travelport systems
Reservations though Internet intermediaries (Internet)	Reservations made on third-party online channels like Expedia or Booking.com, including OTAs, bedbanks, and other third-party platforms

Direct channels encompass the hotel direct, brand web, and voice channels, whereas indirect channels include the Internet and GDS. Hotel direct, and voice are categorized as offline channels, while brand web, GDS, and Internet are considered online channels.

The market share data obtained from the chain are percentages by booking numbers, by room-nights, and by revenue. All other data were considered confidential and were not provided to the authors. Booking-number market share indicates the percentage of reservations made through each distribution channel, while room-nights share represents the percentage of occupied rooms throughout the year by channel. Lastly, revenue share reflects the percentage of income generated by each channel.

Even if no direct data on prices and length of stay were made available, the methodology in this article makes it possible to study how length of stay and price contribute to market share. These categories are computed as follows: the contribution of length of stay is determined by the ratio of room-night market share to reservation market share, while the contribution of price is established by the ratio of revenue market share to room-night market share. Further specifics are outlined in the subsections below.

Market share as CoDa

CoDa analysis is a widely used statistical methodology for analyzing data that provides information about the relative importance of different parts of a whole, often with a fixed sum. Its origin can be traced back to Aitchison’s influential work on chemical and geological compositions in the 1980s (1982, 1986), where only the proportion of each component (a.k.a. part) matters, and the absolute amounts are irrelevant and reflect only the size of the sample (Buccianti et al., 2006). Over the past four decades (Coenders et al., 2023; Greenacre et al., 2023), CoDa analysis has become a standardized toolkit for conducting statistical analyses that deal with the relative importance of magnitudes in STEM (science, technology, engineering and mathematics) and more recently in the social sciences (Martinez-Garcia et al., 2023); including tourism (Coenders and Ferrer-Rosell, 2020; Ferrer-Rosell et al., 2022) and marketing (Ferrer-Rosell et al., 2021).

CoDa has been defined in a number of ways. All definitions share some common traits. It is generally considered as an array of D non-negative (in most definitions strictly positive) numbers called parts, in our case, these represent the distribution channels. Furthermore, researchers’ interest lies in the relative importance of the parts.

Because of the focus on relative importance, the composition x is usually divided by the sum of all its parts through the so-called closure C operator. The closed compositions have unit sum.

By definition, market share is compositional, as it contains relative information, and is always closed, as it has a fixed sum set at 100% or, alternatively, 1 (Morais et al., 2018b). For this reason, CoDa methods are being increasingly used for market share analysis (Arata and Onozaki, 2017; Dargel and Thomas-Agnan, 2024; Dyhrberg et al., 2025; Grifoll et al., 2018, 2019; Huang et al., 2022; Kunce, 2023; Morais et al., 2018a, 2018b; Morais and Thomas-Agnan, 2021). In this article, each composition includes the market share of all distribution channels for a given year, summing to 1.

Classical statistical techniques are inappropriate for CoDa analysis because the data is constrained by positiveness and unit sum, rendering standard statistical tools like mean, variance, correlation, and distance meaningless when applied to CoDa (Egozcue and Pawlowsky-Glahn, 2019; Pawlowsky-Glahn et al., 2015). Just to give two examples: the fact that one part can only increase if some other(s) decrease(s) produces spurious negative correlations; the classical distance between 1% and 2% appears to be the same as between 10% and 11% in spite of the fact that the second figure doubles the fist in relative terms while the fourth figure is only off the third by a factor 1.1.

To address these limitations, the most commonly used CoDa methods involve expressing an original compositional vector with D parts as logarithms of ratios between these parts (Aitchison, 1982, 1986). These log-ratios are unconstrained and thus have the potential to satisfy the assumptions of classical statistical methods (Filzmoser et al., 2018; Greenacre, 2018).

One of the simplest transformations is the so-called centered log-ratio (clr, Aitchison, 1983). A clr can be understood as a row-centered log transformation. For the jth part the clr is computed as:

Logarithms are a well-known transformation to treat differences as relative (in a log scale, the distance between 1% and 2% market share is the same as between 10% and 20%). Row centering (subtracting the row average) is a well-known transformation to focus on differences within rows (the so-called shape) while dropping information on the differences between the rows’ sizes (Greenacre, 2017). The clr combines the two transformations and it makes data relative in two meanings of the word. It expresses them in a relative (logarithmic) scale, and in relative terms with respect to the remaining values of the same row. In this article, each row represents a year.

A compelling advantage of CoDa analysis is that after transforming the original composition into clr, classical statistical techniques that are applicable to unconstrained data can be used in the regular manner (Coenders and Ferrer-Rosell, 2020; Ferrer-Rosell et al., 2021).

The clr transformation can be fruitfully combined with many multivariate descriptive statistical methods, such as principal component analysis or cluster analysis (Aitchison, 1983). The variability in a compositional data set is defined as the sum of the variances for all clr-transformed parts clr(x) _j .

Biplots

In combination with principal component analysis of clr (Aitchison, 1983), the biplot is frequently used to visualize CoDa matrices with various rows and columns (Aitchison and Greenacre, 2002). In particular, the biplot has proven itself as a powerful tool to explore market share dynamics (Arata and Onozaki, 2017; Grifoll et al., 2019). In this article, rows are years and columns distribution channels, i.e., the parts in the market share composition. Form biplots are best to represent the evolution of the rows and are interpreted as follows (Pawlowsky-Glahn et al., 2015).

In form biplots, distribution channels appear as rays emanating from the origin of the coordinate system, and the origin itself represents the average market share for all years. Individual years are displayed as points. When points are located close together in the biplot, it indicates that the corresponding years have similar market share distributions across all channels, enabling visual clustering of those years. Points near the origin correspond to years whose market share composition is close to the average market share for the whole period. Additionally, projecting the points over a ray makes it possible to order the years from lower to higher clr of the channel represented in the numerator by the particular ray, and thus visually trace relevant dynamics that approximate the evolution of the channel’s market share, in a logarithmic, i.e., relative scale, and row-centered. Unlike standard (i.e., non-compositional) biplots, angles between rays play no role in the interpretation.

The compositional biplot can be understood as the most accurate representation of a compositional data table in two dimensions. Said accuracy is measured by the percentage of the total clr variance explained by the first two principal components.

Market share by reservations, room-nights and revenue can each be represented in its own form biplot. The treatment of price and length of stay is dealt with in the next subsection.

Perturbation difference

Two operations on compositions can be used to characterize their products and ratios in such a way that these products and ratios are still compositions and can be submitted to clr transformations and biplots (Arata and Onozaki, 2017; Egozcue and Pawlowsky-Glahn, 2011, 2019).

For any two closed compositions x = C [ x 1 , … , x D ] and p = C [ p 1 , … , p D ] , we define the perturbation operation and denote it by ⊕ as the closure of the componentwise product of two compositions (Aitchison, 1986):

Perturbing composition x with composition p produces the composition X = C [ X 1 , … , X D ] .

We also define the perturbation difference operator represented by ⊖ as the closure of the componentwise ratio between two compositions (Aitchison, 1992):

The composition by which to perturb the composition x in order to produce the composition X is the composition p.

If composition X = C [ X 1 , … , X D ] contains market share expressed as revenue proportions for each distribution channel 1,…,D, and x = C [ x 1 , … , x D ] contains market share expressed as room-night proportions, then the ratio between both p = C [ X 1 x 1 , … , X D x D ] is the contribution of the average price per room night to the market share. Thus, X can be decomposed into X = p ⊕ x .

In the same vein, if x contains market share expressed as room-night proportions, and x' contains market share expressed as proportions of reservation counts, then the ratio of both l = C [ x 1 x 1 ′ , … , x D x D ′ ] is the contribution of room-nights per reservation (i.e., length of stay) to the market share, so that x can be decomposed as x = l ⊕ x ′ .

Thus, the perturbation and perturbation difference operators make it possible to decompose market share expressed as revenue proportions (X) into market share by reservation counts (x'); the contribution of the average length of stay (l); and the contribution of the average price (p) as:

It is not essential that compositions are closed in order to apply these operators, but they were all closed in our case study. Actual reservation counts, revenue values, lengths of stay, and so on were not available.

Perturbation differences are compositions and can be clr transformed. In terms of the clr transformation, the decomposition of market share into reservation proportions and the contribution of prices and length of stay is even more visible as it is expressed with the ordinary sum operator as:

Each of the compositions X, x', l, and p can be represented in biplots representing the aforementioned contributions to market share.

All analyses were carried out with CoDaPack2.03.06 (Comas-Cufí and Thió-Henestrosa, 2011, freely available at https://ima.udg.edu/codapack/. See Ferrer-Rosell et al. (2022) for a gentle introduction to the CoDa methodology and the CoDaPack software. The entire methodological process has been summarized in Figure 1.

OPEN IN VIEWER

Figure 1.

Methodology overview.

Channel market share by reservation numbers (x’)

The analysis of channel market share by reservation numbers reveals notable shifts over time. The total variance of the composition x’ is 0.2909, with the two-dimensional form biplot in Figure 2 explaining 98.7% of this variance, representing an extremely high accuracy of the biplot.

OPEN IN VIEWER

Figure 2.

Form biplot of market share by reservations.

Making orthogonal projections of the years on the directions defined by the channels shows distinct patterns in the evolution of reservation numbers. In the early years of the study (2013–2015), traditional channels such as voice, hotel direct, and GDS retained a more prominent share of reservations compared to later years (clr transformed). Specifically, 2013 stands out as a year in which these channels had relatively the highest reservation counts, reflecting the lingering dominance of traditional booking methods at the time. The distribution of market share for 2016 is close to the mean of all years for all channels (the 2016 point is located at the origin of the rays).

However, the period from 2017 to 2020 reveals a significant shift toward the Internet channel. Years 2017, 2018, and 2019 cluster closely together, and indicate a consistent trend in their projections on the Internet ray. The year 2020 marks a peak in the Internet relative market share. If we follow the timeline and project all points on the direction defined by the Internet ray, the relative importance of the Internet channel in terms of reservation counts has been ever increasing. 2020 has the lowest relative importance of both the direct and GLS channels, while the voice channel had its darkest hour in 2018.

Channel market share by room-nights (x)

The biplot for room-night market share (Figure 3) presents a very similar overall trend to that observed for reservation numbers. The total clr variance is 0.2434, with 98.2% of this variance explained by the biplot. This consistency between reservation numbers and room-nights suggests that the length of stay has had a negligible impact on market share variations over the study period. In our particular case study, the corresponding length-of-stay biplot (l) will be omitted from the analysis.

OPEN IN VIEWER

Figure 3.

Form biplot of market share by room-nights.

The negligible contribution of length of stay to channel market share is further confirmed by the low total clr variance of the corresponding length-of-stay composition l at 0.0074. This indicates that channel preferences for booking durations remained relatively stable across time and channels. Therefore, the key drivers of market share evolution are more tied to changes in booking volume and pricing strategies, rather than variations in length of stay.

Channel market share by revenue (X)

When examining market share by revenue, significant differences emerge compared to reservation counts and room-nights. The biplot for revenue market share (Figure 4) has the highest total clr variance of 0.6300, with 93.5% of this variance explained. This indicates that revenue distribution across channels exhibits more variation over time than the other metrics.

OPEN IN VIEWER

Figure 4.

Form biplot of market share by revenue.

The year 2013 stands out as unique, with the voice channel generating the highest revenue market share over the study period, in relative terms. This reflects the historical significance of traditional booking channels in generating high-value bookings. However, a shift begins to appear from 2014 onward, with the Internet channel progressively increasing its revenue contribution.

Interestingly, the biplot reveals a division in the timeline. From 2014 to 2017, GDS and brand web channels play the most prominent role over the study period, but their relative importance diminishes in subsequent years. From 2018 to 2020, the hotel direct channel emerges as a more significant contributor to revenue and to generating high-value bookings. The consistent performance of the hotel direct channel during these later years suggests that it has become a critical component of a profitable distribution strategy, particularly for higher-priced bookings. This evolution in revenue market share highlights the strategic importance of direct channels for hoteliers aiming to maximize profitability. While online channels have gained in reservation counts, direct channels increase their relative revenue performance, underscoring the need to balance volume with revenue optimization.

Contribution of price to the channel market share (p)

To further understand the differences in revenue contribution across channels, we examine the biplot of price contribution to market share (Figure 5). This biplot is constructed using the perturbation difference between revenue and room-night market share p = ( X ⊖ x ), with a total variance of 0.4549, of which 99.7% is explained by the biplot. As opposed to length of stay, price does have a relevant contribution to market share.

OPEN IN VIEWER

Figure 5.

Form biplot of the contribution of price to market share.

The biplot reveals two distinct periods in terms of price contribution. From 2013 to 2017, traditional channels such as GDS, brand web, voice, and Internet exhibit higher relative prices. These channels appear on the left-hand side of the biplot, indicating that they commanded higher price points during this period.

However, from 2018 onward, a shift occurs. The hotel direct channel emerges as the dominant contributor to price-based market share, as evidenced by its position on the right-hand side of the biplot for the years 2018, 2019, and 2020.

This suggests that hoteliers have increasingly relied on direct bookings to capture higher-value customers. The evolving role of the hotel direct channel in driving price-based market share highlights a critical strategic insight: while digital channels are essential for driving booking volume, direct channels remain pivotal for maximizing revenue through higher prices. This trend indicates that hoteliers should not only focus on expanding their digital footprint but also invest in strengthening their direct booking capabilities to capture premium business.