The Relevance of Using Google Analytics
Utilizing statistical tools to perform quantitative marketing analysis is a sound strategy for making data-driven decisions. In a rapidly evolving marketplace, companies must often innovate and adapt to remain competitive and resilient (Knobloch, 2021; Sarter & Bailey, 2023). However, decisions based solely on experience or biases unrelated to data can destroy organizational growth. For this reason, data analytics is an integral part of a company’s routine.
GA is a quantitative analytics tool that primarily collects and processes large amounts of user data. With skillful data analytics, companies can significantly increase professional and market performance (Wamba et al., 2020). The characteristics presented on the platform include, but are not limited to, user demographics, geographic location, device types, and behavior patterns.
Companies need to be clear about the goal and identify which variables may be relevant to it (Figure 1). GA will do the rest, collect the necessary data for the specified period, produce a report, and provide a foundation for further statistical analysis that involves a deeper look at marketing data (Semerádová et al., 2020). In contrast, if GA is used without a clear understanding of the goals and selects data that turn out to be uninformative in the context of overall corporate issues, the tool may prove useless or even detrimental. For this reason, it is critical to first outline the purpose, identify the variables, and collect and process marketing information responsibly.
Conceptualization: Google Merchandise Store
This case study chose Google Merchandise Store (“Store”) as the company of interest. Store is an online e-commerce platform that sells branded merchandise owned by Google (Figure 2). The platform offers a wide selection of printed products, enabling customers to leverage the brand identity. One of the motivations for choosing this company for marketing analysis is that GA provides free access to store data as part of a demo account, which helps to explore the capabilities of the analytics service further with a real-world example.
For a substantive analysis of Store, Hamel’s WAMM is a valuable professional tool for assessing a company’s current maturity and weaknesses for further improvement (Hamel, 2009). Given the nature of the store, namely its representation in the virtual media field and the extensive use of the brand for GA advertising purposes, it is not unlikely that the company will be at the advanced or leading levels. Thus, Web analytics in brand operational processes may be widespread. Nevertheless, there are opportunities to improve Store marketing performance.
One of the critical marketing goals of any company, including Store, is conversion. Conversion should be understood as a valuable metric that describes a user action on the brand’s website (SEO DG, 2020). For example, if a user clicked a third-party link to the Store’s official website, registered, added items to the cart, and paid for them, this would be considered a positive customer action for increasing conversions. Higher conversion rates align with the company’s market development and customer loyalty, indicating it is something the brand should strive for (Bag et al., 2022). Conversion is a complex metric that simultaneously includes several internal metrics: purchases, registrations, and engagement. A detailed study of individual characteristics is part of a thorough marketing analysis.
Marketing Goals
The Goals for Store included statistics on purchases, platform registrations, and user engagement, corresponding to the amount of time spent exploring the website in the foreground. A detailed examination of these goals allows for tracking the effectiveness of the store’s marketing strategies and identifying weaknesses.
Appendix A shows a decrease in the percentage of completed purchases (Goal 1) from the total number of users who initiated the process, for Americas and unspecified regions, to 8.99% and 12.80%, respectively, indicating users from those countries had a worse path through the sales funnel for Store. For the number of pages viewed per session (Goal 2), there was a decrease in the percentage of users completing this goal (≥10) in the Americas (3.17%), Asia (2.77%), Africa (7.72%), and for unspecified locations (29.78%) compared to 2021. This may indicate that users from these continents are visiting pages per session.
In the context of higher BRs for these regions, this indicates a negative signal: users were not finding what they were looking for and are leaving the site. Notably, across all regions (except unspecified regions), the percentage of registered users (Goal 3) increased, indicating they are actively engaged in the funnel. This is especially interesting in studying user types (Appendix B), which shows that new and returning users registered, with the former’s share being about 3-4 times higher.
For the age distribution, all younger audiences (18 to 44) showed a decrease in the percentage of completed purchases (0.74% to 5.60%), with only the 25- to 34-year-old cohort showing a 2.36% drop in the percentage of engaged users (Appendix C). Neither group showed a decrease in the percentage of registered users. Thus, younger adults may be more sensitive to in-store shopping than older adults, although all groups are willing to create an account and may consider a purchase shortly. Men were 1.24% worse engaged in completing orders than in 2021 (Appendix D); however, both genders show a cheerful pattern on the other metrics.
As Table 1 shows, there was also a 10.54% drop in overall conversion rates for Store compared to last year and a 0.86% decrease in transactions. Nevertheless, the average purchase price rose, thereby increasing Store profits by almost a third. Interestingly, the overall purchase rate fell by 0.64%. This data could indicate a decrease in the number of users making multiple purchases, a shift toward more affluent customers, or a general price increase. Thus, the Store is experiencing a shortage of quality visitors, who may refuse to buy due to changing consumer habits or stronger competition. This indicates a significant conversion problem.
Table 1. Comparison of profit and conversion rates
Thus, the main goal of the analysis is to understand how to improve the store’s conversion rates.
Marketing Analysis
Marketing data analytics involves conducting a quantitative analysis using SPSS. The fundamental goal of such analysis is to identify patterns that can be used in the context of CPO (Zimmermann & Auinger, 2022). Analyzing user behavior and identifying conversion barriers enables you to adjust the funnel to increase conversion rates and reduce BRs. The full year 2022 was chosen for the analysis because it allows a large sample to be collected, and the analysis is not split into seasons but rather focuses on broad patterns.
Data Preparation
Before quantitative data can be processed, it must be prepared. The collection procedure involved selecting variables of interest as dimensions and metrics, as shown in Figure 3. The motivation for selecting these variables was the weaknesses found in the previous step, and the overall goal of the analysis was to determine deep patterns and causal relationships between conversion factors and user demographics. The generated sample size was 198 lines, but 121 records for the variables Avg. Order Value and Transactions: zero values were observed. Nevertheless, they were not removed because the other variables had meaningful entries.
The data preparation procedure involved examining the nature of the variables to allow for analysis. The demographic variables were nominal, and the remaining variables of interest (metrics) were measured on an interval-ratio scale and presented as numerical or percentage values. In other words, recoding was only required for categorical data with specific numeric label values assigned (Anavir, 2019). There were no missing values because GA created a table with only meaningful metrics and ignored any omissions.
Descriptive Analysis
Primary to the quantitative analysis is descriptive analysis, which aims to determine the general patterns of the sample. Users (n = 198) were evenly distributed across Europe, the Americas, Africa, and Australia, as shown in Figure 4. The numbers of males and females were almost identical: 49.50% (n = 98) females and 50.50% (n = 100) males. The age distribution also shows a balance across age cohorts, with only the oldest generation (≥65) slightly underrepresented. In other words, the sample was demographically representative, with diverse geographic, gender, and age groups represented in roughly equal proportions.
Table 2. Age distribution of the sample
For the ten numeric variables of interest, the results of the descriptive statistics are shown in Table 3. For example, the average cost per completed order was $61.29 (SD = 136.10), the average duration per session was 187.02 (SD = 72.85) seconds, and the average number of transactions was 66.80 (SD = 306.15) — some distributions were more scattered than others (Anderson & Courage, 2022). Regarding conversion goals, the average completion rate for Goal 1 (completed purchase) was 0.67 (SD = 1.57), largely because 121 participants (61.1%) did not make purchases. The average percentage of those users who were engaged and therefore visited more than ten pages was 7.26% (SD = 4.40%). The average percentage of registered users was 3.90 (SD = 2.33). The average number of users for 2022 who started entering the conversion funnel but abandoned it, for example, by not making a purchase or exiting the website, was 276.08 (SD = 893.11), and the average BRs after viewing only one page was 51.04% (SD = 7.78%), indicating a high exit rate of visitors from the funnel and a need for optimization (Wang et al., 2021; Zimmermann & Auinger, 2021).
Regression Analysis
To establish causal relationships and determine the effects of explanatory variables on factors of interest, regression analysis is appropriate. This strategy shows the strength of the influence of one variable on another and allows us to determine its significance. Only five multiple regression models (Appendix E) were statistically significant (p <.05).
Gender had a significant effect on average session duration, while geographic location affected user engagement rate, the average number of pages per session, and BRs. Age significantly affected the number of sessions: older users required fewer sessions to explore the website. This could be a positive signal if such users made more intensive purchases or a negative signal if they left the Store and never returned.
Correlation Matrix
A correlation matrix helps to determine the degree of influence of one variable on another. A correlation (Figure 5) identifies the strength and direction of the relationship between two variables but cannot be used to infer causality (Hayes, 2022). Significant correlations (p <.05) were found for the relationship of the geographic attribute of the sample with the proportion of users involved (r = -.242), the proportion completing a purchase (r = -.188), the number of pages per session (r = -.233), and BR (r =.206), and thus only for BR was a positive relationship with region observed. For age, correlations were found with user registration rate (r = -.151), sessions (r = -.213), and funnel conversions (r = -.174), indicating that younger users registered less, required fewer sessions, but performed better in funnel conversions. For gender, only the relationship with average session duration (r = -.202) was confirmed.
A significant correlation was confirmed between the independent variables. The shares of completed purchases, engaged users, and registrations increase, along with average session duration and order value (Figure 5). The desire to increase the time a user spends in Store will effectively impact marketing results (Wilkins, 2022). The negative relationship between average session duration and average order value in BR supports this. On the other hand, a practical goal is to reduce AFs, so attention should be paid to the positive relationship between missed market opportunities, average session duration, and the average number of sessions, since an increase in one parameter tends to increase the others.
T-Tests and ANOVA
Once it was clear that each demographic was related to the variable of interest (Appendix E, Figure 5), a parametric test should be used to assess differences in means. T-tests and ANOVAs are tools that allow us to assess the significance of differences between two (t-test) or more (ANOVA) groups and thus identify patterns of user behavior (LS, 2020a; LS, 2022b; Izakova, 2021). Because ≥ 2 categories represent the explanatory factors in the sample, it is appropriate to conduct a one-way ANOVA for age and geographic attributes and a t-test of independent samples for gender. Appendix F and Figure 5 show that men, on average, used less time (172.46 seconds) per session than women (201.88 seconds). All differences in the geographic distribution of users across variables of interest were significant.
As seen in Table 4, North America (including the Caribbean) was the leader in terms of average session duration, shares of engaged users, completed purchases, registrations, transactions, and sessions visited, including pages per session, but was also the leader in terms of AFs. Given Google’s widespread brand recognition in this region, such local user activity does not seem surprising (Graham & Elias, 2021; Spanos, 2021; Bavent, 2022). Northern Europe and Central America were moderate leaders, showing strong performance across average session duration, user engagement share, purchases and registrations completed, average order value, and transactions. However, these figures were generally slightly lower than for North America. The conversion was worst in Central Asia: users performed worst in average order value, transactions, and the percentage of purchases completed.
Taxonomic Analysis
Segmenting the existing sample into clusters is the goal of k-means cluster analysis to identify statistically distinct user segments. A vital issue in conducting such an analysis is determining the number of clusters since k-means cluster analysis cannot identify this (Pourahmad et al., 2020). Instead, TwoStep cluster analysis (Schwarz’s Bayesian Criterion) can be used to identify the optimal number of clusters: Figure 6 shows that this number is two. Of all variables, only geographic distribution was not taken into account because a large number of value categories complicates the clustering procedure.
Appendix G shows the cluster analysis results performed for the two segments, with the ANOVA test demonstrating that the variables Avg. Order Value, Registrations (Goal 3), and Gender did not contribute significantly to cluster formation (p >.05) and were removed from consideration for this reason. Appendix H shows the results of the updated cluster analysis for the two segments.
Demographically, the clusters differed in age: the first was centered on younger users, and the second included an older audience. The first cluster (younger) tended to spend more time per session, had a higher engagement rate, and completed transactions more often while having a lower BR; however, the first cluster had higher AFs. Overall, the younger segment has higher conversion engagement, but attention needs to be paid to both clusters, especially given the roughly 40 times lower number of AFs among the second segment.
Recommendations
Based on the results, several recommendations should be made, including comparing crucial conversion metrics between 2021 and 2022 and conducting a sample analysis of 2022 users. The recommendations translated into Store operational practices are expected to benefit the brand’s development and optimize performance (Hamel, 2009). Development refers to increasing conversion metrics and providing measures that allow audiences to move through the funnel more effectively.
Providing Social Proof
Since conversion engagement is generally not low, but BRs and AFs are high, the Store may provide social proof. The availability of such evidence has been proven to increase purchase rates and, therefore, positively impact brand conversion (Moore & Lafreniere, 2020; Kirwin, 2021). Specific steps could be to use the reviews section under each product or to post a placeholder with some customer reviews. A simpler option is implementing a rating based on customer reviews so that new users can see which products are more popular.
Introduction of Personalized Products
The store has seen a high percentage of registered visitors of different ages: this can be leveraged. The store can recommend personalized product selections based on the products viewed or added to the cart. Personalization is one of the essential attributes for increasing conversions, so this strategy seems appropriate (Desai, 2019). The store will be able to increase transactions and completion rates and make the user experience more comfortable.
Retargeting Campaigns
There is an extremely high number of AFs among high-transaction visitors (segment 1). This happens when a customer puts items in their cart but does not pay for them and does not return because either they changed their mind or a competitive offer turned out to be more profitable (Rodrigues et al., 2019). The Store can use retargeting campaigns to fix this and increase conversion rates. This includes a search engine, email, or social media ads reminding customers of abandoned products and possibly offering some discount or loyalty points for paying for selected products.
Increase Session Duration
The results show a positive relationship between session duration as one variable and conversion metrics (Goals #1, #2, and #3) as a second variable. This means that the Store should make efforts to increase session duration. This could be creating interactive and visually appealing content, posting multiple picks, including personalized picks, or creating discount systems that start in “a few minutes” to entice the visitor to stay longer (Wang et al., 2021; Miller & Hosanagar, 2020). The proposed recommendations are expected to increase Store conversion rates.
References
Anavir. (2019). Recoding categorical variable for multiple regression. Stock Exchange.
Anderson, S. & Courage, A. (2022). Standard deviation vs. variance: What’s the difference? Investopedia.
Bag, S., Srivastava, G., Bashir, M. M. A., Kumari, S., Giannakis, M., & Chowdhury, A. H. (2022). Journey of customers in this digital era: Understanding the role of artificial intelligence technologies in user engagement and conversion. Benchmarking: An International Journal, 29(7), 2074-2098.
Bavent, A. (2022). How Google fuels its brand value fire. Kantar.
Desai, D. (2019). An empirical study of website personalization effect on users intention to revisit e-commerce website through cognitive and hedonic experience. Advances in Intelligent Systems and Computing, 2, 1-16.
Google: Official merchandise store. (n.d.).
Graham, M., & Elias, J. (2021). How Google’s $150 billion advertising business works. CNBC.
Hamel, S. (2009). The web analytics maturity model.
Hayes, A. (2022). Correlation: What it means in finance and the formula for calculating it. Investopedia.
Izakova, N., Kapustina, L., & Drevalev, A. (2021). Performance of social media marketing communications of industrial companies.
Kirwin, P. (2021). How revenue can increase through reviews. Forbes.
Knobloch, P. M. (2021). Business model innovation in the film industry: How Nordic film studios adapt to a changing market.
LS. (2020a). One-way ANOVA in SPSS statistics.
LS. (2020b). Independent-samples t-test using SPSS statistics.
Miller, A., & Hosanagar, K. (2020). Personalized discount targeting with causal machine learning.
Moore, S. G., & Lafreniere, K. C. (2020). How online word‐of‐mouth impacts receivers. Consumer Psychology Review, 3(1), 34-59.
Pourahmad, S., Basirat, A., Rahimi, A., & Doostfatemeh, M. (2020). Does determination of initial cluster centroids improve the performance of-means clustering algorithm? Comparison of three hybrid methods by genetic algorithm, minimum spanning tree, and hierarchical clustering in an applied study. Computational and Mathematical Methods in Medicine, 2020, 1-11.
Rodrigues, T., Guerreiro, J., & Loureiro, S. M. C. (2019). How effective are retargeting ads? How effective are retargeting ads.
Sarter, E. K., & Bailey, E. C. (2023). Understanding public services: A contemporary introduction. Policy Press.
Semerádová, T., Weinlich, P., Semerádová, T., & Weinlich, P. (2020). Using Google Analytics to examine the website traffic. In T. Semerádová & P. Weinlich (Eds.), Websitequality and shopping behavior: Quantitative and qualitative evidence (pp. 91-112). Springer.
SEO DG. (2020). What is a conversion? Conversion marketing definition. SEO Digital Group.
Spanos, M. (2021). Brand storytelling in the age of artificial intelligence. Journal of Brand Strategy, 10(1), 6-13.
Wamba, S. F., Dubey, R., Gunasekaran, A., & Akter, S. (2020). The performance effects of big data analytics and supply chain ambidexterity: The moderating effect of environmental dynamism. International Journal of Production Economics, 222, 1-14.
Wang, X., Li, Y., Cai, Z., & Liu, H. (2021). Beauty matters: Reducing bounce rate by aesthetics of experience product portal page. Industrial Management & Data Systems, 121(8), 1848-1870.
Wilkins, J. (2022). How to increase dwell time & keep visitors on your website longer. QuayClick CM.
Zimmermann, R., & Auinger, A. (2022). Developing a conversion rate optimization framework for digital retailers — case study. Journal of Marketing Analytics, 1-11.
Appendixes
Appendix A: A comparison of the geographic distribution of users for the three purposes
Appendix B: Type of user in the context of Goal 3
Appendix C: Age distribution of users for three purposes
Appendix D: Gender distribution of users for the three purposes
Appendix E: Results of multiple regression analyses for variables of interest
Appendix F: Results of t-test and one-way ANOVA
Appendix G: Results of k-means cluster analysis (primary)
Appendix H: Results of k-means cluster analysis (secondary)
