Dynamic Pricing in the Hair &

Beauty Industry

Case Timma

Bachelor’s Thesis
Lari Mykrä
4.1.2018
Business
Technology

Approved in the Department of Information and Service Economy .1.2018 and

awarded the grade
 Aalto University, P.O. BOX 11000, 00076 AALTO
www.aalto.fi
Abstract of bachelor’s thesis

Author Lari Mykrä

Title of thesis Dynamic Pricing in the Hair & Beauty Industry – Case Timma
Degree Bachelor of Science (Econ.)
Degree programme Business Technology
Thesis advisor(s) Niina Mallat
Year of approval 2018 Number of pages 20 Language English

Abstract
Dynamic pricing has become increasingly popular especially on electronic marketplaces with the
help of the rise of modern computer systems. Essentially a form of price discrimination and a
means of revenue management and maximization, dynamic pricing has been most widely studied
and used in the air travel and hospitality industries. In the very core, dynamic pricing is based on
being able to accurately estimate demand for a certain product or service. Electronic marketplaces
make demand forecasting easier for businesses in many cases, so dynamic pricing becomes
possible for a variety of industries.
In my thesis I have studied different factors that affect the probability of sale of a certain time slot
in a marketplace selling appointments for hair & beauty services. The research data consists of 62
137 available time slots during the month of November 2017. The data was analyzed in R and a
Generalized Linear Model (GLM) with a binomial variance function and a logit link function was
fitted to study regressions. My research proves that time of day, proximity to center, consumer
ratings and weekday all have a statistically significant effect on whether a particular time slot is
sold or not.
The results concerning the effect of consumer ratings and proximity to center (location) are in
alignment with previous research from the hospitality industry. The effect of time of day or
weekday on the likelihood of sale had not been researched before in a relevant industry.
It seems that to develop a deeper understanding of the factors affecting likelihood of sale and
thus demand in the hair & beauty industry more variables would have to be taken into account.
This seems like a feasible subject for further research. Also, a similar study from a different service-
based industry would be interesting.

Keywords dynamic pricing, hair & beauty, electronic marketplace

1 Introduction ......................................................................................................................... 1
1.1 Background and Scope of the Study ........................................................................................... 1
1.2 Research Problem and Questions .............................................................................................. 2
1.3 Outline of the Research ............................................................................................................. 2

2 Literature Review ................................................................................................................. 3

2.1 Dynamic Pricing as a Concept .................................................................................................... 3
2.1.1 Overview of Dynamic Pricing ........................................................................................................ 3
2.1.2 Use Cases of Dynamic Pricing ....................................................................................................... 4
2.1.3 Customer Perceptions on Dynamic Pricing ................................................................................... 5
2.2 Differences between Electronic Marketplaces ............................................................................ 6
2.3 Limitations of Dynamic Pricing on Electronic Marketplaces ........................................................ 7

3 Methodology ....................................................................................................................... 8
3.1 Case Selection and Introduction ................................................................................................ 8
3.2 Data Collection.......................................................................................................................... 8
3.3 Research Method ...................................................................................................................... 8

4 Case Study: Timma – Marketplace for Hair & Beauty ............................................................ 9

4.1 Structuring the Data .................................................................................................................. 9
4.2 Formulating the Hypotheses .................................................................................................... 10
4.3 Testing the Hypotheses ........................................................................................................... 11

5 Discussion .......................................................................................................................... 16
5.1 Implications to Research ......................................................................................................... 17
5.2 Implications to Practice ........................................................................................................... 17
5.3 Limitations and Future Research .............................................................................................. 18

References ............................................................................................................................... 19
1 Introduction

Electronic marketplaces that gather sellers and willing buyers of a particular product or
service in the same virtual location have been trending massively over the course of the
past decade. Examples from various industries such as Hotels.com, Supersaver,
Trivago, eBay, Airbnb, Foodora or Venuu make it easier for sellers and buyers to meet,
make pricing more transparent and remove unnecessary friction from the transaction
processes. These kinds of platforms almost always live off commissions from
transactions made via the platform, so it is in the interest of the companies running these
platforms to streamline every aspect of their service to maximize revenues of the
supply-side participants.

On the other hand, every participating seller (company or individual) is likely looking to
maximize their own sales. In many cases the answer to doing this is dynamic pricing, a
form of price discrimination. In essence, dynamic pricing is the concept of
automatically changing the price of a product/service based on estimated demand at
different price points to maximize profits.

The concept of dynamic pricing first became popular in the airline industry in the 1970s
along with the deregulation of the industry. Another major industry benefiting from the
use of dynamic pricing is the hospitality industry. New applications of dynamic pricing
are found constantly in different industries as technology and algorithms making
dynamic pricing possible become available more easily and with a cheaper price tag.

1.1 Background and Scope of the Study

To develop a functioning dynamic pricing model or algorithm requires deep

understanding of a certain market and the factors affecting demand in the market. A
case study from the hair & beauty industry (hair and beauty salons) will shed light on if
it could be a line of business suitable for dynamic pricing of services.

The salon industry consists of the hair, beauty and massage sectors. In Finland the value
of the salon industry is estimated to be roughly a billion euros yearly whereas in Europe
the estimated value is 100 billion euros yearly. This entails products and services sold
through salons. Still, surprisingly little research has been made concerning the industry.

1
Although the salon industry as a whole is quite large based on revenues, individual
salons are small (on average 1.8 people work in a single salon) and there are few chains.
In other words, the market is very fragmented as the estimated number of companies
operating in the space in Finland is around 20 000. It is noted, however, that in many
cases two or more companies or sole proprietors operate in the same salon.

1.2 Research Problem and Questions

The main purpose of this research is to review the theory on the subject of dynamic
pricing and to find out which factors would have to be taken into account when
developing a model for dynamic pricing of hair and beauty services.

The aim of this study is to familiarize the reader with the concept of dynamic pricing
and its most common applications in electronic commerce. Also, a case study of a
marketplace selling hairdressing & beauty appointments is conducted. I aim to answer
the following questions:

- Do certain factors (location of salon, time of day, consumer reviews) affect the
probability of sale of a certain time available?

- Would dynamic pricing be feasible for these services on the marketplace?

It is not, however, within the scope of this study to formulate a dynamic pricing model
or algorithm for these services.

1.3 Outline of the Research

In the literature review chapter principles and theory behind dynamic pricing are
discussed. Different angles and views of dynamic pricing are evaluated, such as which
factors make a particular service or product especially suitable to being priced
dynamically. Attention is also given to how consumers perceive the use of dynamic
pricing and whether it affects their view on a marketplace or brand.

In the third chapter the case company Timma is presented. Also, the research method
and research design are discussed in more detail to familiarize the reader with how the
data collection and research of the case company was performed.

2
A case study will then be conducted to identify different factors that distinguish various
service providers from one another and to test whether the effect of these factors on the
probability of sale on the Timma –platform is statistically significant.

Finally, findings are discussed, as well as implications to practice and research. Also
suggestions for further research are made.

2 Literature Review

2.1 Dynamic Pricing as a Concept

As defined by Pigou (1932), there are three degrees of price discrimination. First-degree
price discrimination occurs when a company charges the maximum possible price for
each unit consumed and thus is able to capture all available consumer surplus. In reality
this rarely happens. Second-degree price discrimination happens when a different price
is charged for different quantities of a product or service consumed. This is quite
common e.g. when giving bulk discounts on a product. Third-degree price
discrimination means charging different prices from different consumer groups.
Examples include pricing movie tickets differently for children, adults and seniors or
giving student discounts from software purchased online. Dynamic pricing is a form of
third-degree price discrimination (Krugmann, 2000), where the price can be changed
automatically based on remaining inventory and estimated future demand (Elmaghraby
& Keskinocak, 2003). (Pigou, 1932)

2.1.1 Overview of Dynamic Pricing

Dynamic pricing as a means of price discrimination has gained popularity with the help
of the rise of modern computer systems. While no single researcher or group of
researchers are recognized as inventing the concept of dynamic pricing, in their 1963
article Kincaid & Darling (1963) were among the first ones to study an inventory
pricing problem, where the optimal solution was to use dynamic pricing to maximize
expected total revenues. Furthermore, Lovelock (1984) argues that perishable goods are
especially suitable to be dynamically priced. Since services are perishable by nature,
dynamic pricing would be an optimal solution for service pricing. Of course, as
Lovelock (1984) further explains, the demand curve and price elasticity of a
3
service must be known in order for dynamic pricing to be optimal. Jayaraman & Baker
(2003) take Lovelock’s findings a bit further and argue that perishable services such as
hotel rooms and air travel are optimal for dynamic pricing for five reasons:

1. Not possible to store or keep inventory

2. Possible to class and price differentially (e.g. different rates for different
consumer segments)

3. Centralized order processing eases estimating supply and demand

4. Value differential between incremental value and incremental cost is high

5. Changes in demand can be sudden and temporary

Today, applications of dynamic pricing can be found from various industries, but it is
most used and researched in the airline and hospitality industries.

In essence, dynamic pricing means being able to determine the price of a product or
service based on estimated demand to maximize revenue or profits (Elmaghraby &
Keskinocak, 2003). If the demand of a product or service is known or can be reliably
estimated, prices can be set in a way that maximizes estimated profits. Alternatively,
Kannan & Kopalle (2001) determine dynamic pricing as a “pricing strategy in which
prices change over time, across consumers or across product/service bundles”.
Regardless of how it’s determined, to reach the optimal solution for any dynamic
pricing problem the demand curve for the product/service in question must be known.
Besbes & Zeevi (2009) argue that this rarely is the case in real life and develop a model
for solving dynamic pricing problems without knowing the demand function.

2.1.2 Use Cases of Dynamic Pricing

Burger & Fuchs (2005) studied dynamic pricing in the context of the airline industry.
They found that “a dynamic pricing strategy has a neutral or a positive effect on
revenues, depending on competitors’ revenue management strategy”. Traditionally
airlines have tackled the price discrimination problem with seat inventory control,
booking-limit calculations and e.g. minimum stay requirements. Burger & Fuchs (2005)
found that dynamic pricing works especially well against these traditional revenue
management methods. However, if two (or more) competing airlines would apply
4
dynamic pricing they would dynamically start to undercut each others’ prices which
would in turn likely result in having a neutral total effect on demand. (Burger & Fuchs,
2005)

Abrate et al. (2012) studied the use of dynamic pricing in the context of the European
hotel industry. Their research consisted of data from some 1000 hotels in eight capital
cities in Europe. According to their findings, over 90% of prices changed during their
test period suggesting that at least some form of dynamic pricing was performed by a
vast majority of the hotels studied. Furthermore, in almost half of the cases the prices
observed changed by more than 10%, meaning that savvy consumers could get
significant savings. It is noted, however, that price fluctuations are smaller than in the
airline industry. It is also shown that whether prices go up or down when getting closer
to the date of the booking depends on a variety of factors such as star-rating, the
weekday for which the room is booked and the overall demand situation in the city in
question. (Abrate et al., 2012)

Dynamic pricing also has its applications in more traditional industries, such as retail.
Jayaraman & Baker (2003) argue that “dynamic pricing is best suited for products that
are clearly specified or widely understood, that are either perishable (consumables) or
time-sensitive (hotel rooms) or have a depreciating value (e.g. computer components,
automotive parts)”. Elmaghraby & Keskinocak (2003) studied dynamic pricing in the
context of products where inventory restrictions apply. They argue, that e.g. fashion
goods, which arguably fit into the category of having depreciating value, are in many
cases suitable to be priced dynamically. The life cycle of fashion items is usually short,
thus making inventory replenishments hard or impossible for retailers, so they can in a
way be considered perishables similar to e.g. hotel rooms. (Elmaghraby & Keskinocak,
2003)

2.1.3 Customer Perceptions on Dynamic Pricing

Dynamic pricing has been proven to be effective a means for revenue management on
the supply side, but how do consumers feel about ever-changing prices? Kannan &
Kopalle (2001) argue, that the trustworthiness of a vendor is negatively affected if
consumers realize they have ended up paying more because of dynamic pricing. The
key to retaining customers’ trust is making dynamic pricing structures transparent. An
5
example of this is giving discounts to loyal customers or frequent shoppers, since most
customers understand the logic behind this and accept it. (Kannan & Kopalle, 2001)

Haws & Bearden (2006) continue on the subject, confirming in their study that it is
indeed the differences in pricing between customers that result in the greatest
perceptions of unfairness. Also, price changes within very short periods of time are
viewed as more unfair than changes during a longer period. If the change in price has
occurred over a period of one month, fairness perceptions aren’t affected anymore.
(Haws & Bearden, 2006)

2.2 Differences between Electronic Marketplaces

The concept of homogeneous versus heterogeneous supply marketplace was coined by

Golden (2017). While the publication is not scientific, the classification very well
illustrates key differences between electronic marketplaces. Golden (2017) categorizes
electronic marketplaces based on homogeneity or heterogeneity of supply.

Homogeneity of supply means that a certain supplier’s product or service is not easily
distinguishable from the product or service of another. This is the case for example with
Uber. In heterogeneous marketplaces on the other hand, such as Airbnb, the supply is
unique and distinguishable from each other. Moreover, Golden (2017) argues, that
heterogeneous supply marketplaces are search marketplaces, where one of the greatest
transaction barriers for the consumer is deciding the right supply, whereas
homogeneous supply marketplaces are matching marketplaces that mostly have to
worry about making the transaction as smooth as possible for the consumer. (Golden,
2017)

As previously discussed, dynamic pricing is best suitable for goods or services that are
perishable, time-sensitive or have a depreciating value (Jayaraman & Baker, 2003).
However, as Elmaghraby & Keskinocak (2003) argue, if demand can be reliably
estimated, dynamic pricing can be effectively used to maximize profits. This, in turn,
means that dynamic pricing could be suitable and effective for any electronic
marketplace or a company selling its goods or services via an electronic marketplace if
only demand can be forecast on a reliable enough level.

6
Generally dynamic pricing in its automated form is better suitable for marketplaces
where the size of the companies providing the supply is large (e.g. flights) than for
marketplaces where the supply is mostly provided by small companies (e.g. Airbnb),
since creating a dynamic pricing algorithm requires a deep understanding of demand
data (Jayaraman & Baker, 2003) and is quite likely very expensive. This could,
however, be solved by Airbnb and its likes offering “Dynamic Pricing as a Service”,
where the marketplace provider would offer a dynamic pricing algorithm for the supply
to use.

2.3 Limitations of Dynamic Pricing on Electronic Marketplaces

While dynamic pricing seems to be widely accepted as a suitable means for maximizing
revenues on electronic marketplaces, some limitations should be considered. A
downward spiraling price war situation, an escalation of the airline example from
chapter 2.1.2 can be explained as Bertrand’s paradox (1903). If a competitor reduces
prices a company is tempted to reduce its own prices even more. This will result in both
companies ending up on the lowest set price level. Because of this, lower limits for
prices need to be set in dynamic pricing algorithms if they are competing against each
other.

Also, as mentioned before, dynamic pricing is essentially based on being able to

estimate the future demand of a product or service. For a company like Uber operating a
homogenous supply marketplace where it completely controls the supply this is
reasonably easy. Therefore it makes sense for Uber to use their surge pricing. A
company running a heterogeneous supply marketplace, like Airbnb, might have an
equally good estimate of future demand. However, they are not allowed to dynamically
price vacation homes listed on the site without the permission of the homeowners.
Airbnb could offer dynamic pricing for venues listed on the marketplace, but the
algorithm would have to be created by Airbnb since it is not feasible for most venue
providers to formulate a dynamic pricing algorithm themselves, as mentioned before.

7
3 Methodology

3.1 Case Selection and Introduction

Normally appointments in the hair, beauty and massage industries are booked on the
phone or via online booking on the business’ own website. The Timma-marketplace
provides and additional booking and marketing channel, where professionals working in
the industries can market their available times for up to five following days. These times
are at a high risk for being left empty and thus producing zero revenues. Timma aims to
solve this inefficiency by gathering the times and customers looking for last-minute
times to the same platform. Additionally, Timma makes it easier for consumers to
browse through businesses near their location and compare prices, reviews and available
times. When a consumer makes a booking, she is forced to pay in advance to minimize
the amounts of no-shows. Consumers are often willing to do this, since most of the
times sold on the platform are discounted.

3.2 Data Collection

Case data was gathered from Timma Oy’s Timma.fi –marketplace sales in Helsinki over
the period of one full month, November 2017. According to the company there is some
seasonal variation in the sales within the month, so it is important to study the data from
a whole month’s period. This seasonal variation is caused by e.g. paydays in Finland
usually being the 15th and 30th days of the month.

A data set of 62 137 total available time slots was studied. The amount of appointments
sold during the month was 6 308. It is important to note, however, that an available time
slot can be for example four hours long and a sold appointment is usually between half
an hour and two and a half hours long. This means that it is not possible to estimate the
percentage of sold appointments from all available time slots only with this data.

3.3 Research Method

Generalized linear model (GLM) was used to analyze the regressions between different
factors and their effect on the likelihood of sale of a certain time slot. It is important to

8
note, however, that GLMs reveal the correlations between different variables but
correlation does not necessarily imply causality.

Generalized linear models were introduced by Nelder & Wedderburn (1972). GLMs
make it possible to model non-normal data when usual assumptions used in regression
models are not satisfied. Generalized linear models consist of a positive variance
function and a monotonic link function. The variance function describes the varying of
the response as a function of the mean while the link function transforms the mean to a
scale, making the model linear. (Nelder & Wedderburn, 1972)

4 Case Study: Timma – Marketplace for Hair & Beauty

4.1 Structuring the Data

The data obtained from the company was divided into two spreadsheets, available times
and sold times. The column headers of available times and sold times are shown in
Figures 1 and 2 respectively. The meanings of the headers are shown in Figure 3.

start end customerName rating createdTime postalCode

Figure 1. Headers of available times table.

start end customerName rating purchaseTime postalCode

Figure 2. Headers of sold times table.

Header Meaning

start starting time of sold or available slot

end ending time of sold or available slot

customerName name of the salon where the time is available or sold

rating customer rating of the salon, 1 to 5 stars

9
createdTime time when available slot was created

purchaseTime time when the purchase was made

postalCode postal code of the salon

Figure 3. Meanings of headers from available and sold times tables.

Based on this information, two additional columns were created to help analyze the data
better. Same three columns were added into both data sets. Headers and meanings are
shown in Figure 4.

Header Meaning

timeOfDay morning, day, afternoon or evening

proximityToCenter downtown, closeby or outside

Figure 4. Meanings of headers of new columns added.

Time of day was divided into morning, day, afternoon and evening based on the starting
time of the available or sold time slot. 06:00 – 09:45 is morning, 10:00 – 13:45 is day,
14:00 – 16:45 is afternoon and 17:00 – 21:00 is evening.

Proximity to center means (quite self-explanatorily) the closeness of a salon to Helsinki

city center. Proximity to center was divided into three categories, downtown, closeby
and outside. Basically the most densely populated areas such as Kamppi, Punavuori and
Kallio are categorized as downtown, district centers such as Vuosaari or Lauttasaari are
categorized as closeby and the remaining areas, such as Paloheinä or Etelä-Haaga are
categorized as outside.

4.2 Formulating the Hypotheses

It would seem logical that a salon in a densely populated area would be likely to sell
more than a salon in the outskirts of the city. Also, a salon with a high customer rating
should sell more than a salon with a lower rating, since ratings are a form of online

10
word of mouth and generally better word of mouth leads to more sales (Chevalier &
Mayzlin, 2006). Most people are working during the day, so an available time in the
afternoon or evening could perhaps be preferable for many.

Ögut & Tas (2012) proved that in the context of the hotel industry, a higher customer
rating significantly increases online sales. The effect of location on hotel selection has
been researched by Rivers et al. (1991) and Ananth et al. (1992). Both found that
convenience of location is among the most significant factors affecting hotel selection.

There isn’t, however, sufficient research on the effect of time of day on sales from
relevant industries. This might be because time of day is not a relevant factor for most
service industries excluding the hair, beauty & restaurant industries.

Based on these findings two hypotheses were formulated:

𝐻1: 𝐻𝑖𝑔ℎ𝑒𝑟 𝑟𝑎𝑡𝑖𝑛𝑔 𝑐𝑜𝑟𝑟𝑒𝑙𝑎𝑡𝑒𝑠 𝑡𝑜 ℎ𝑖𝑔ℎ𝑒𝑟 𝑝𝑟𝑜𝑏𝑎𝑏𝑖𝑙𝑖𝑡𝑦 𝑜𝑓 𝑠𝑎𝑙𝑒

𝐻2: 𝑃𝑟𝑜𝑥𝑖𝑚𝑖𝑡𝑦 𝑡𝑜 𝑐𝑒𝑛𝑡𝑒𝑟 𝑐𝑜𝑟𝑟𝑒𝑙𝑎𝑡𝑒𝑠 𝑡𝑜 ℎ𝑖𝑔ℎ𝑒𝑟 𝑝𝑟𝑜𝑏𝑎𝑏𝑖𝑙𝑖𝑡𝑦 𝑜𝑓 𝑠𝑎𝑙𝑒

In the absence of relevant research concerning time of day no hypothesis on its effect
can be made. The correlation of time of day with the probability of sale is however also
studied.

4.3 Testing the Hypotheses

The hypotheses were tested using R, an open source programming language for
statistical computing. The data from available times and sold times were first combined
in a new table to form the basis of the regression analysis. A dependent variable y was
added to the table. Depending on whether a time slot is sold or not, y gets a value of 1
or 0, 1 meaning sold and 0 meaning not sold.

Two separate GLM models were tested. In the first model only rating, time of day and
proximity to center were taken into account as explanatory variables. In the second
model also weekday was added as an explanatory variable.

A binomial variance function with a logit link function was used in the model.

11
 1. 𝑦 ~ 𝑟𝑎𝑡𝑖𝑛𝑔 + 𝑡𝑖𝑚𝑒 𝑜𝑓 𝑑𝑎𝑦 + 𝑝𝑟𝑜𝑥𝑖𝑚𝑖𝑡𝑦 𝑡𝑜 𝑐𝑒𝑛𝑡𝑒𝑟

2. 𝑦 ∼ 𝑟𝑎𝑡𝑖𝑛𝑔 + 𝑡𝑖𝑚𝑒 𝑜𝑓 𝑑𝑎𝑦 + 𝑝𝑟𝑜𝑥𝑖𝑚𝑖𝑡𝑦 𝑡𝑜 𝑐𝑒𝑛𝑡𝑒𝑟 + 𝑤𝑒𝑒𝑘𝑑𝑎𝑦

This kind of analysis reveals which factors have the most effect on whether a time slot
is sold or not and if the effect is statistically significant.

After the GLM models were tested, ANOVA with chi-squared test was analyzed to
study if the different factors actually helped to explain the dependent variable y.

The results were quite well aligned with the hypotheses, as can be seen from Picture 1.
Proximity to center is the single most important factor affecting the probability of sale.
This feels pretty logical, since downtown areas have the highest population density and
in addition the greatest amount of workplaces. Also, the greatest amount of shopping
possibilities, restaurants, bars, cafes and other attractions are also found in downtown
areas. In other words, people are going there anyways, so it is suitable to also book e.g.
a haircut from the same area. This is also in alignment with the findings of Rivers et al.
(1991) and Ananth et al. (1992). It seems that as is the case with hotels, location is also
a major factor affecting customers’ choice of salon.

The effect of rating is not unambiguous. While having a rating of 4, 4.5 or 5 stars
produced significantly better results than e.g. a rating of 2.5 stars, a rating of 4.5 stars
yielded the best results. This calls for further research, since e.g. the number of ratings
given wasn’t considered in the model and might have an effect on the probability of sale
also.

Time of day also had a significant effect on the probability of sale. Times starting in the
afternoon and in the evening had twice the probability of being sold than times during
the day and four times the probability of times starting in the morning. Results from
GLM model 1 are shown in Picture 1.

12
Picture 1. The effect of different variables on the probability of sale, GLM model 1.

When the second model was tested it was found that the weekday actually has the
biggest effect on the probability of sale. Sunday and Saturday have the highest
probability of sale, two times as high as Wednesday. There might be two reasons for
this. First of all, many if not most salons are closed or at least open a very limited time
on weekends. This means that the amount of times available on weekends is quite low.
Also, as can be seen from Picture 2, the weekdays are in complete reverse order in terms
of likelihood of sale. This suggests that generally people don’t like to use services
provided by salons as much early in the week than they do when the weekend is getting
near. Table 1 further illustrates the results from GLM model 2.

13
Picture 2. The effect of different variables on the probability of sale, GLM model 2.

14
Table 1. Results from GLM model 2 as a table.

ANOVA with Chi-squared was also analyzed to find out if the addition of each variable
improves the model significantly, i.e. if the variables help to explain the dependent
variable significantly. As can be seen from Table 2, the residual deviance is initially
42095. The residual deviance decreases with the addition of each variable with a
significant p-value, so it can be concluded that the variables (time of day, rating,
proximity to center and weekday) each help to explain the dependent variable – whether
the time is sold or not.

15
Table 2. ANOVA table, GLM model 2.

5 Discussion
My research proves that proximity to center, time of day and rating all have a
statistically significant correlation with the probability of sale. Additionally, weekday,
which was originally not meant to be included as an explanatory variable had the
largest effect on the probability of sale. The findings concerning proximity to center
(location) and rating are in alignment with previous literature. (Rivers et al., 1991;
Ananth et al. 1992; Ögut & Tas., 2012) The effect of rating does not, however, seem to
be linear. In fact, a rating of 4.5 out of 5 stars seems to yield best results. There might
be multiple reasons for this. Perhaps a rating of 4 stars or more is considered “good
enough” by consumers and after that other factors affect the purchasing decision more
than the rating. Also, written reviews were not taken into consideration as an
explanatory variable. A salon with a 4-star rating might have more impressive written
reviews than a salon with a 5-star rating. Furthermore, the number of reviews was also
not taken into account in the model. It might be that a salon with a 4,5-star rating and
huge total number of ratings is considered more appealing than a salon with a 5-star
rating and only a small number of reviews.

In the case study different factors affecting the likelihood of sale are studied and tested.
However, as discussed in the literature review section, dynamic pricing is essentially
based on being able to estimate the future demand of the service/product as accurately
as possible. While my research cannot be used as a base for fabricating a dynamic
pricing model for hair and beauty services in the context of the Timma-marketplace, it
16
provides valuable insight on which factors are the most important ones affecting the
probability of sale of a single time slot. It can be argued that this information is crucial
when estimating demand for these services – when likelihood of sale is high, so is also
demand. Of course, to develop a well-functioning demand-estimation model more
factors would have to be taken into account.

5.1 Implications to Research

No previous research about the sales of hair and beauty appointments via an electronic
marketplace exists. My research suggests that location of the salon, time of day,
consumer ratings and weekday all have a statistically significant effect on the
probability of sale of an available appointment. This seems to be in alignment with
previous research, although previous research from the hospitality industry only
includes two of these factors, location and consumer ratings (Ögut & Tas, 2012; Rivers
et al., 1991; Ananth et al., 1992).

My findings prove that in addition to previous studies from the hospitality industry,
weekday and time of day affect the sales of hair and beauty appointments significantly.
These might also be interesting factors to consider when studying the sales of other
services, such as hotels, restaurants or golf tee times to name a few.

5.2 Implications to Practice

My research reveals that weekday as a factor has the biggest effect on the probability of
sale. Available times later in the week have a significantly higher likelihood of being
sold than times early in the week. Moreover, times on Saturday and Sunday are sold
twice as likely than times on Monday. Timma could use this information to help its
salon customers sell more on the platform e.g. by encouraging salons to work longer
hours on weekends. Also, as times in the afternoon and evening are have a significantly
higher probability of sale than day times, salons should, if vacant, be closed during the
day and remain open until later in the evening.

The demand for times early in the week seems to be lower than later on in the week.
Salons might be able to sell more times for e.g. Monday and Tuesday by lowering prices
specifically for those days. Similarly, times in the morning don’t sell as easily as times
in the evening or afternoon. Thus, lowering prices for times in the morning seems

17
sensible.

As discussed in the literature review, it is probably not feasible for individual salons or
even salon chains to start formulating an algorithm to automate dynamic pricing.
However, it seems feasible for Timma to start offering “Dynamic Pricing as a Service”
for its salon customers. This could e.g. be priced as a flat monthly fee or value-based, if
the increases in revenues for the salons could be reliably measured. The formulation of
a dynamic pricing model or algorithm would require a deeper understanding of the
different factors affecting the demand and remains an issue calling for further research.

5.3 Limitations and Future Research

In the context of this study the effect of four explanatory variables on the probability of
sale of a single time slot were studied. Although it seems that all of these factors affect
the probability of sale and the effect is statistically significant, it is not possible to
accurately forecast sales based on only these variables. To develop a deeper
understanding of how to formulate a model for estimating demand for these services
more variables would have to be taken into account. These include e.g. discount
percentage, absolute price, number of customer reviews and how much in advance the
booking was made. Moreover, it could be feasible to test the model individually for
different services (e.g. haircut, cut & color, eyelash extensions) to develop an
understanding about whether the factors affect the likelihood of sale differently for
different services. Also, factors salons are not in control of, such as weather and date
might also have an effect on the probability of sale.

The data studied includes only one city, Helsinki and one month, November 2017. It
would be interesting to repeat the study with data from a longer time period and from
various cities and different countries to find out whether the results from different
cities and countries are in line with each other or if differences arise. It would be also
interesting to see this study perhaps in a broader form repeated in some other service
industry, such as restaurants to see if the results are in accordance with my study.

18
References

Abrate, G. & Fraquelli, G. & Giampaolo, V. 2012. Dynamic pricing strategies: Evidence
from European hotels. International Journal of Hospitality Management. Vol. 31. P.
160-168

Ananth, M. & DeMicco, F. & More, P. & Howey, R. 1992. Marketplace Lodning Needs
of Mature Travelers. The Cornell Hotel and Restaurant Administration Quarterly. Vol.
33:4. P. 12-24

Bertrand, J. 1889. Calcul des probabilités. Gauthier-Villars. P. 5-6

Besbes, O. & Zeevi, A. 2009. Dynamic Pricing Without Knowing the Demand Function:
Risk Bounds and Near-Optimal Algorithms. Operations Research. Vol. 57:6. P. 1407-
1420

Burger, B. & Fuchs, M. 2005. Dynamic Pricing - A future airline business model.
Journal of Revenue and Pricing Management. Vol. 4:1. P. 39-53

Chevalier, J. & Mayzlin, D. 2006. The Effect of Word of Mouth on Sales: Online Book
Reviews. Journal of Marketing Research. Vol. 43. P. 345-354

Elmaghraby, W. & Keskinocak, P. 2003. Dynamic Pricing in the Presence of Inventory

Considerations: Research Overview, Current Practices, and Future Directions.
Management Science. Vol. 49:10. P. 1287-1309

Golden, J. 2017. Four Questions Every Marketplace Startup Should Be Able to Answer.
[Cited 28 Dec 2017. ]Available at: https://medium.com/@jgolden/four-questions-
every-marketplace-startup-should-be-able-to-answer-defb0590e049

Haws, K. & Bearden, W. 2006. Dynamic Pricing and Consumer Fairness Perceptions.
Journal of Consumer Research. Vol. 33. P. 304-311

Jayaraman, V. & Baker, T. 2003. The Internet as an Enabler for Dynamic Pricing of
Goods. IEEE Transactions on Engineering Management. Vol. 50:4. P. 470-477

Kannan, P.K. & Kopalle, P. 2001. Dynamic Pricing on the Internet: Importance and
Implications for Consumer Behavior. International Journal of Electronic Commerce.
Vol. 5:3. P. 63-83

Kincaid, W.M. & Darling, D.A. 1963. An Inventory Pricing Problem. Journal of
Mathematical Analysis and Applications. Vol. 7. P. 183-208

Krugmann, P. 2000. What price fairness? The New York Times

Lovelock, C. 1984. Strategies for Managing Demand in Capacity-Constrained Service

Organizations. Service Industries Journal. Vol. 4:3. P. 12-30

Nelder, J.A. & Wedderburn, R.W.M. 1972. Generalized linear models. Journal of the
Royal Statistical Society A. Vol. 135, P. 370-384

Pigou, A. 1932. The Economics of Welfare. 4th ed. London: Macmillan

19
Rivers, M. J. & Toh, R. S. & Alaoui, M. 1991. Frequent-stayer programs: The
demographic, behavioural and attitudinal characteristics of hotel steady sleepers.
Journal of Travel Research. Vol. 29:4, P. 41–44

Ögut, H. & Tas, B. 2012. The influence of internet customer reviews on the online sales
and prices in the hotel industry. The Service Industries Journal. Vol. 32:2. P. 197-214

20