Viva Aerobus Analytics Report, Adam Hide 1
Viva Aerobus Analytics Report, Adam Hide
1
Contents
1. Company profile
2. Project Goals
3. Approach
4. Main & Key Findings
5. Overview of Techniques
6. Recommendations
7. Limitations
2
Company Profile
• Low-Cost Mexican Airline founded in 2006
• Part Owned by Ryanair
• “The low cost airline of Mexico” – company slogan
• Serves predominantly domestic flights
• 1 international destination – Heuston, TX
• No frills airline – aimed at minimizing costs, and driving high volume of passengers
• High growth company focused on expansion
3
Project Goals
1. Identify advertising and marketing opportunities for Viva Aerobus
2. Identify opportunities for website design, structure and improvements
3. Explore whether actions need to taken to improve conversion rates on certain devices, OS and browsers
4. Examine opportunities for remarketing campaigns
5. Understand the characteristics of Viva Aerobuses website users
4
ApproachMy approach to this analysis is representative of the dataset I was given of Viva Aerobuses website interaction between Jan 1st 2014 & March 31st 2014. I grouped the data by similar themes and created graphs to highlight some initial insights and then applied some statistical analysis to derive more in depth conclusions.
For each section/theme I provided my main findings and then the key findings supported by the results of the statistical analysis. From these key findings I then discuss the implications for Viva Aerobuses in terms of the business objective for it’s website. In many cases the majority of the insights can be found by studying the graphs. A lot of the data provided was already summarized and therefore applying statistical analysis on minimal observations was not appropriate
As the fundamental purpose of the website is to sell airline tickets I put a particular focus on the ecommerce metrics in the data such as conversions rate, revenue, value per visit etc. I then provide a number of recommend actions for VA to take as a consequence of my findings.
5
Active Periods – Main Findings
0
20000
40000
60000
80000
100000
120000
140000
160000
180000
Visits Jan - March
0
20000
40000
60000
80000
100000
120000Mean Visits By Day
There is a clear cyclical pattern to VA’s website visits. Visits are highest Monday - Thursday while sharp declines are experienced during the weekends from Fridays. Between Mon 6th and Thurs 9th Jan there were particularly high number of visits which is the only exception to the general pattern
6
Visits
Mean 75,553
Standard Error 3,133
Median 79,735
Mode #N/A
Standard Deviation 29,724
Sample Variance 883,528,660
Kurtosis 0
Skewness 0
Range 137,474
Minimum 32,635
Maximum 170,109
Sum 6,799,792
Count 90
Active Periods – Key Findings
Weekday Weekend
Mean 89895.3 40249.7
Variance 505405797.9 48969981.0
Observations 64.0 26.0
Hypothesized Mean Difference 0.0
df 85.0
t Stat 15.9
P(T<=t) one-tail 0.0
t Critical one-tail 1.7
P(T<=t) two-tail 0.0
t Critical two-tail 2.0
Weekday vs Weekend T-test – This clearly shows that there is a statistically significant relationship between website usage and weekdays. This was evident in the pattern observed in the graph but this T-test confirms that pattern. The T critical value of 15.9 illustrates how strong this relationship is and the fact that the weekday mean is more than double the weekends.
Descriptive Statistics for Visits – From the above table it is clear that there is significant variances in visit patterns to VA’s website. The standard deviation is particularly high along with the range. This points towards that different days have significantly higher/lower visits
7
Site Engagement – Main Findings
23%
4%
7%
19%
23%
16%
7%
0%
5%
10%
15%
20%
25%
0
200000
400000
600000
800000
1000000
1200000
1400000
1600000
1800000
0-10 seconds 11-30 seconds 31-60 seconds 61-180 seconds 181-600seconds
601-1800seconds
1801+ seconds
Duration of Visits
Visits % of Total
From the above table it is clear to see that users of VA are spending a long time on VA’s site. The 0-10 seconds would be associated with the bounce rate of the site. Apart from that 46% of users are spending over 2 minutes on the site. This implies that purchasing the tickets or finding desired information takes quite a while for users to complete. This highlights that their may be issues with site usability particularly for users who are spending over 30 minutes on the site (7%).
8
Site Engagement – Main Findings
0
500000
1000000
1500000
2000000
2500000
3000000
3500000
4000000
4500000
0 1 2 3 4 5 6 7 8-14 15-30 31-60 61-120 121-364 365+
Vis
its
Days Since Last Visit
Days Since Last Visit
The above chart highlights that the majority of user to VA’s website a first time visitors (64% of traffic)
9
Site Engagement – Main Findings
0
500000
1000000
1500000
2000000
2500000
3000000
1 2 3 4 5 6 7 8 9-14 15-25 26-50 51-100 101-200 201+
Vis
its
Count of Visits
Count of Visits
Visits
From this chart it is clear to see that apart from single visits, users are returning to the website a significant amount of times. This suggests that users are attracted to the site and are have reasons to revisit the sites. Further analysis into what exact pages users a visiting may provide some actionable insights. For example if a particular destination or flight is being checked multiple times this may signal future demand or opportunities to increase/decrease prices
10
Site Engagement – Key Findings
11
Data Correlation – The above correlation table doesn’t provided any ground-breaking insights. Some of the correlations are very much expected as a lot of the metrics are closely related to each other in the first place. Such as time on site with ecommerce metrics or pages/visit to average duration. Correlations also do not imply causation so utilising insights from this data should be taken with caution
Visits % New Visits New Visits Bounce Rate Pages / Visit
Avg. Visit
Duration Transactions Revenue
Ecommerce
Conversion
Rate
Visits 1
% New Visits -0.43 1
New Visits 0.99 -0.41 1.00
Bounce Rate -0.33 0.86 -0.33 1.00
Pages / Visit 0.54 -0.90 0.53 -0.77 1.00
Avg. Visit Duration 0.52 -0.91 0.53 -0.73 0.96 1.00
Transactions 1.00 -0.42 0.99 -0.32 0.54 0.53 1.00
Revenue 1.00 -0.42 0.99 -0.32 0.54 0.53 1.00 1.00
Ecommerce Conversion
Rate 0.61 -0.72 0.61 -0.49 0.91 0.89 0.62 0.62 1.00
Language – Main Findings
84%
0% 16%
Visits by Language
Spanish
French
English
Clearly the majority of VA’s website users speak Spanish, but VA should consider that 16% are English speakers so they should ensure that all pages of site have both Spanish and English options.
12
Language – Key FindingsSpanish Non-Spanish
Mean 4.1% 3.6%
Variance 0.00058916 0.000402404
Observations 4 6
Pooled Variance 0.00047244
Hypothesized Mean Difference 0
df 8
t Stat 0.37964444
P(T<=t) one-tail 0.35704453
t Critical one-tail 1.85954804
P(T<=t) two-tail 0.71408905
t Critical two-tail 2.30600414
T-Test Spanish vs Non Spanish – The result of the above t-test show that language seemingly does not impact on conversion rates. Non-Spanish users have a slightly lower mean conversion rate but this is not statistically significant at the 5% level
13
Country
95%
5%
0%
Revenue by Country
Mexico
United States
Other
From this graph it is clear to see that the majority of VA’s customers are based in Mexico and that the US is the only secondary market of any significance for VA. The remaining companies combined represented less than 1% of revenue and visits
Mexico United States
Per Visit Value 136.40 111.98
Ecommerce Conversion Rate 4.99% 3.70%
0.00%
1.00%
2.00%
3.00%
4.00%
5.00%
6.00%
0.00
20.00
40.00
60.00
80.00
100.00
120.00
140.00
160.00
Val
ue
per
Vis
it
Value Per Customer
This graph also shows that on a per visit basis, Mexican customers are more valuable to VA than US users.
14
Browser & Device Type – Main Findings
desktop tablet mobile
Bounce Rate 19.19% 17.74% 25.76%
Ecommerce Conversion Rate 5.28% 2.54% 0.69%
0.00%
5.00%
10.00%
15.00%
20.00%
25.00%
30.00%
Axi
s Ti
tle
Bounce & Conversion Rate by Device Type
This chart highlights the serious issue VA are having with mobile site users. The conversion rate on mobile is 0.69% which shows that users are not purchasing flights very often with these devices. The bounce rate for mobile is also the highest at 25.76%.
Tablets have the lowest bounce rate but the ecommerce conversion rate (2.5%) does is still less that half of desktop users (5.2%). This implies that the site layout is reasonably attractive to users but there are some issues with converting these visits to sales
Clearly desktop users are much more likely to lead to sales15
Browser & Device Type – Main Findings
desktop tablet mobile
Pages / Visit 7.81 7.31 4.69
Avg. Visit Duration 546.18 420.80 273.12
0.00
100.00
200.00
300.00
400.00
500.00
600.00
0.00
1.00
2.00
3.00
4.00
5.00
6.00
7.00
8.00
9.00
Axi
s Ti
tle
Visit Duration & Pages Per Visit By Device Type
Again this chart highlights the issues with mobile. Users are spending far less time on and viewing less pages on VA’s website versus tablet and desktop.
This implies that users are finding it more difficult to complete purchases and/or find the information they need through VA’s mobile site which is evident from low pages per visit and low visit duration
Desktop and tablets perform reasonably well here which suggests that the sites layout may be a similar experience for users
16
Browser & Device Type – Main Findings
0.00%
1.00%
2.00%
3.00%
4.00%
5.00%
6.00%
0
500000
1000000
1500000
2000000
2500000
3000000
3500000
Browser Type
Visits Ecommerce Conversion Rate
90%
7%3%
Traffic
desktop mobile tablet
Overall VA’s website performs equally well across different desktop browsers with none having a significantly higher conversion rate. The issue again is where mobile device browsers are concerned in particular Android devices.
The traffic pie chart shows how visits are distributed by device type. Desktop accounts for the majority of traffics and which naturally corresponds with revenue. Conversions rates as I have mentioned are also much higher for desktop users.
17
Device Type – Key Findings
Factors Influencing Mobile Conversion Rate – from the above regression table it is clear to see that bounce rate and pages per visit have a significant impact on conversion rates. Obviously there is an element of colinearity here as pages/visit will influence bounce rates. Interestingly new visits don’t have a statistically significant relationship at 5% level of confidence but it does come very close. VA should identify the reasons for high bounce rate levels and aim to reduce it as doing so will increase their conversion rate. The adjusted r square of this model is 0.44 which is shows that this model has a relatively high predicting power.
SUMMARY OUTPUT
Regression Statistics
Multiple R 0.671848
R Square 0.451379
Adjusted R Square 0.442422
Standard Error 0.007384
Observations 250
ANOVA
df SS MS F Significance F
Regression 4 0.010991 0.002748 50.3935727 6.48E-31
Residual 245 0.013359 5.45E-05
Total 249 0.02435
CoefficientsStandard Error t Stat P-value Lower 95%Upper 95%Lower 95.0%Upper 95.0%
Intercept -0.02829 0.004643 -6.09404 0.00 -0.03744 -0.01915 -0.03744 -0.01915
% New Visits 0.008179 0.004673 1.750183 0.08 -0.00103 0.017383 -0.00103 0.017383
Bounce Rate 0.017376 0.005869 2.960466 0.00 0.005815 0.028936 0.005815 0.028936
Pages / Visit 0.006101 0.000696 8.762597 0.00 0.004729 0.007472 0.004729 0.007472
Avg. Visit Duration -8E-06 7.86E-06 -1.02239 0.31 -2.4E-05 7.45E-06 -2.4E-05 7.45E-06
18
New vs Returning Customers – Key Findings
Returning Visitor New Visitor
Per Visit Value 152.95 100.36
Ecommerce Conversion Rate 5.61% 3.58%
0.00%
1.00%
2.00%
3.00%
4.00%
5.00%
6.00%
0.00
20.00
40.00
60.00
80.00
100.00
120.00
140.00
160.00
180.00V
alu
e p
er V
isit
$
Value of Visitors
This chart shows the value of a returning website visitor versus a new visitor. Clearly the returning visitor is much more valuable to VA as they are more likely to lead to sales which is also represented in the significantly higher conversion rate.This indicates that remarketing campaigns should be a major focus for VA digital advertising campaigns. It may also signal that first time users may be finding it difficult to purchase flights as they may be unfamiliar with the layout. So VA should ensure that their homepage is clear and easy to navigate 19
Overview of Techniques
20
Descriptive StatisticsThe below table summarizes the key points of how descriptive statistics are used to present and describe data. To perform these calculations I used the descriptive statistics function through the data analysis tool pack in Microsoft excel.
21
Correlation
• Measures the linear relationship between two variables
• Does not imply causation – does not highlight the underlying reason as to why something is correlated
• Measures the degree at which two variables are associated with each other
• To perform these calculations I used the correlation function using the data analysis tools in excel
22
T-Tests
• Used to highlight significant differences between two sets of data
• In my case I used independent T-Tests rather than paired sample t-tests as I only had data from one period therefore the hypothesised mean difference was 0
• Tested at the 95% level of confidence so any P Values above 0.05 were considered not significant
23
Linear Regression
• Tested at the 95% level of confidence so any P Values above 0.05 were considered not significant
• Examines the relationship between a dependent variable Y to a number of independent variables X1, X2, X3….
• Seeks to provide a predictive equation in the form
Where b denotes the coefficient of each variable which is provided when a regression is run through the Data Analysis pack on Microsoft excel in my case
• The adjusted r-squared value provided by regression models is a measure of its predictive effectiveness or the % of variances in the data that can be accurately explained
24
Recommendations
1. Mobile Site - Viva Aerobus should consider redesigning or improving their mobile site to make it more friendly and easier for customers to purchase tickets. Mobile has by far the lowest conversion rate and the highest bounce rate. Even though mobile (smartphones) only accounts for 7% of traffic this is clearly a sales opportunity for VA. To determine what factors are causing these issues VA should perform mobile usability tests to identify areas to improve on and check if the site functions correctly
2. Remarketing – VA should increase their digital advertising spend on remarketing campaigns to target users who have previously visited the site. These users have higher conversion rates and higher average spends than new visitors. This does not mean they should not attempt to gain new visitors but it does offer a significant strategy for increasing revenue.
25
Recommendations
3. Website Activity – Monday to Thursday represent the busiest times for VA’s website. VA should use this information to ensure that bandwidth capacity in increased for these periods. If any major site maintenance needs to be conducted I would recommend that this takes place on either a Sunday or Saturday to minimise potential negative impacts.
4. Customer Funnel – VA should examine whether it’s purchase process is too long. Users are spending a long time on it’s site and have to view a high number of pages. VA should identify whether certain pages are causing users to abandon their purchase and remove unnecessary pages which waste users time.
26
Recommendations
5. Advertising Budget - VA should allocate its marketing budget based on revenue and conversion rate. 95% of it’s revenue coms from Mexico so the marketing budget should reflect this. Mexican users also are more likely to spend money if they visit the site than US users. International vistors represent a tiny fraction of VA’s business therefore I would not recommend high advertising spend for these customers.
6. Search & Display Campaigns – should be focused on desktop users as this represents 90% of VA’s web traffic and is also attributed to higher conversion rates. These campaigns would result in a higher ROI than tablet or mobile.
27
Limitations
1. Data Period – the dataset I had was for one period (Jan – March) so the results are limited in the fact that I could not compare different periods to note any significant improvements.
2. Website Actions – the dataset does not provided info as to actions taken by VA on its site such as web designs improvements, sales, promotions, advertising campaigns, referral data , ect. All these actions could have a major effect on the metrics and would influence the findings of this project
3. Metrics – Most of the metrics provided were functions of other metrics, and would all have elements of co-linearity. This reduced the number of tests I could do in terms of regressions for example. Transactions and revenue metrics were all linked.
4. I did not provide an analysis of the time of day when users are most active. This is because the data I was given did not have timestamps. The visits where given as an hour index from 1-2159 which does not highlight what time of day the visits were recorded
5. Dataset – I was limited in fact that in certain circumstances there were not enough observations to perform some statistical tests. So I found that graphing certain key data points provided more relevant insights.
28