INTRODUCTION

The era of big-data is accompanied by social media, cell phones, the internet of things, e-commerce, and search engines. Thus, data stored in the IT systems is getting doubled rapidly, and this increase has been majorly caused by the massive amount of data created by organisations (Surbakti, Wang, Indulska, & Sadiq, 2020). Previous research highlights that this form of technology may be excessively beneficial for retailers and may aid in profitability; especially during volatile economic conditions, the technology may act as a significant resource and opportunity to fashion retailers (Silva, Hassani, & Madsen, 2020). The use of big-data is recommended for organisational level benefits, especially for banking, retail, insurance, and government sectors (Yadav, Wang, & Kumar, 2013).

The postulate looks valid as retailers evidence a large stream of customer visits daily. And hence need a unified picture of customers’ preferences, especially for making strategic decisions (Butt, Suroor, Hameed, & Mehmood, 2021; Chauhan, Mahajan, & Lohare, 2017). Well-Known brands use the technology to analyse consumer demand patterns and assist companies in assessing adequate stock, price, colours, design, etc. (Oosthuizen, Botha, Robertson, & Montecchi, 2020). Thus, the increase in demand for data-management experts is legitimate, and therefore organisations are striving to increase their budgets to recruit skilful data analysts (Surbakti et al., 2020). Though there are several types of data for, e.g., consumer review, search data, shopping baskets, etc., to perform research extensively and effectively, there is a need to use shopping basket data (Tharwat & Gabel, 2019).

Limited studies may provide insights regarding this beneficial technology concerning Pakistan, and hence the country is experiencing a significant lacking of firms that may incorporate big-data. Still, this lacklustre approach is due to insufficient technology and cultural differences concerning the western world (Shaikh, Sultan, Khaskhelly, & Zehra, 2021). On the other side, analysis of big-data resulted in a correlation of several variables, although correlation does not mean causative relationships and therefore making strategic decisions is not easy (Blackburn, Alexander, Legan, & Klabjan, 2017). However, to assess anomalies in consumption patterns, it is legitimate to consider the use of advanced algorithms for retailers while using big-data analytics (Silva et al., 2020). One must also not ignore the complexity of analysis based on big-data. All the correlations are not significant (Chauhan et al., 2017); hence the skill set of big-data analysts is also much more critical (Surbakti et al., 2020). Thus, this study uses advanced algorithms as a mediator by considering Silva, Khan, and Han (2017), which was not indulged in previous studies. However, without the inclusion of the moderation of skilled data-scientist, the study could not be compelling, as indicated by Surbakti et al. (2020). The study is one of the studies aiming to highlight the significance of big-data, which is termed the future of technology (Katal, Wazid, & Goudar, 2013).

The country’s lack of research and application has been in observation for a long time (Shaikh et al., 2021). Therefore, the significance of the study has several folds as this study will not only increases research work on big-data analytics regarding Pakistan but will also relate the work to the industry, which is perceived as the best for the application of big-data (Dekimpe, 2020). Moreover, the study is also related to the store layout design, one of the most critical sides of retail business and significantly impacts consumer buying behaviour (Aktas & Meng, 2017). Last but not least, to understand the significance of the study, one must consider the significant amount of contribution of the retail sector to the economy of Pakistan (Ahmed, Ullah, & Paracha, 2012). Hence, the study is pervasive and can provide valuable insights to researchers, academicians, and managers better to understand the benefits and implications of the technology. 

LITERATURE REVIEW

RESEARCH MODEL AND HYPOTHESIS

H1 There is a relationship between the use of big-data and the application of advanced algorithms.

H2 There is a relationship between the use of advanced algorithms and the optimisation of assortment strategies in the organised retail sector.

H3 Advanced Algorithms does not mediate the relationship of big-data and optimisation of assortment strategies.

H4 Skilled Data–Scientist does not moderate the relationship of advanced algorithms and optimisation of assortment strategies for organised retail segment.

RESEARCH METHODOLOGY

Research Methodology is a general set of assumptions used to conduct a study in the desired manner (Long, 2014), which interrelates all the parameters to indicate coherence between all the study aspects (Brannick & Roche, 1997). However, the parameters provided by Saunders, Lewis, Thornhill, and Wilson (2009) are found to be much important in formulating a practical set of strategies to conduct research. Therefore, the parameters have also been followed in this study.

STATISTICAL ANALYSIS

According to the rule of thumb, outer loading values for all the indicators must be higher than or equal to 0.708 (Hair et al., 2013) & values lesser than 0.70 must be excluded, significantly when the exclusion aids the overall reliability of the construct (Hair et al., 2013).

However, in the case of exploratory analysis, outer loadings of 0.60 or above are treated as potent enough to be included in the construct (Afthanorhan, 2014). Therefore, in the light of these parameters, all the elements (indicators) used in Table 1 are potent enough to be included in this research.

R-Square (Predictive Accuracy) explains the portion of the variance associated with endogenous variables explained by the structural model (Ringle, Silva, & Bido, 2015). For reflective models, the least acceptable value to reflect the impact of the independent variable is 0.25. At the same time, 0.50 is termed as moderate, and 0.75 or above are treated as substantial for highlighting the impact (Hair et al., 2013). Therefore, in the light of these parameters, all the relations highlighted through Table 2 seems to have a moderate fit and hence appropriate enough to reflect the variance caused by independent variables (Benitez, Henseler, Castillo, & Schuberth, 2020).

Confirmatory Factor Analysis and Outer Loadings

Table 3 indicates Convergent Validity and Construct Reliability (Ringle et al., 2015). AVE measures convergent validity, which measures the extent to which the variable is correlated with the construct (Benitez et al., 2020).

Therefore, the values of 0.5 or above are treated as the confirmation of convergent validity (Ringle et al., 2015). Similar is highlighted by Table 3, where AVE is more than 0.5 in each case, thus ensuring convergent validity. On the other side, table 3 also indicates construct reliability through Cronbach’s Alpha (α), Dillon-Goldstein’s rho and composite reliability. The construct reliability is appropriate enough as all the values reflecting construct reliability are higher than 0.7 (Sijtsma, 2009; Sijtsma, 2009).

Table 3: Construct Reliability and Convergent Validity

	Cronbach’s Alpha	Composite Reliability	Average Variance Extracted (AVE)
Advanced Analysis Algorithms	0.737	0.824	0.542
Assortment Strategies	0.809	0.868	0.568
Big Data	0.745	0.839	0.568
Moderating Effect SDS	1.000	1.000	1.000
Skilled Data Scientists	0.790	0.849	0.618

Table 4: Discriminant Validity (Heterotrait-Monotrait Ratio)

	AAA	AS	BD	BD*SDS
AAA
AS	0.459
BD	0.394	0.489
BD*SDS	0.274	0.136	0.115
SDS	0.343	0.681	0.488	0.122

The purpose of using HTMT as a tool for discriminant validity is to highlight the difference among the variables of the same construct through numeric values (Cheung & Lee, 2010). As Fornell and Larcker (1981), criterion and cross-loadings seem to be less effective measures than the HTMT. HTMT is used to predict collinearity problems in latent construct. Thus, legitimate to declare HTMT as the tool used to assess multi-collinearity (Hamid, Sami, Sidek, & M, 2017). The results must have lower values than 0.85 as values higher than 0.85 indicate lacking discriminant validity (Hair, Sarstedt, Ringle, & Gudergan, 2017). Therefore, according to Table 4, there is no value conflicting with the above mentioned criteria. Therefore, there is no case of multi-collinearity detected in the model.

Table 5: Path Coefficients

	Original Sample (O)	Standard Deviation	T Statistics	P Values
Advanced Analysis Algorithms -> Assortment Strategies	0.219	0.052	4.211	0.000
Big Data -> Advanced Analysis Algorithms	0.322	0.047	6.797	0.000
Big Data -> Assortment Strategies	0.153	0.052	2.952	0.003
Moderating Effect SDS -> Assortment Strategies	-0.025	0.047	0.530	0.596
Skilled Data Scientists -> Assortment Strategies	0.441	0.047	9.397	0.000

Table 5 is used to indicate the impact of latent variables on each other through inferential statistics. This is treated as one of the significant tools for assessing reflective measurement models used in smartPLS. The tool uses t-statistics and p-values to redress. The impact and fulfillment of both of these criterions’ is a must for appropriate analysis. The maximum value to which one may leverage the p-value is 0.05 above that there would be no relationship among the latent variables. Similarly, the minimum t-statistics to assure the relationship is 1.97, and lower than this value, no value may assure the relationship between the latent variables (Hair, Hult, Ringle, & Sarstedt, 2016).

Table 5 also indicated those skilled data scientists are the critical predictor of assortment strategies. However, the moderation of skilled data scientists for skilled data-scientist is not perceived worthy in optimising assortment strategies. This might be caused as the FMCG sector mostly have entire demand, and therefore, the assortment is majorly affected by the use of advanced algorithms rather than the skills of data scientists.

Therefore, in the light of these parameters, it has been indicated that the big-data is perceived as an effective tool for optimising assortment strategies in the organised retail sector of Pakistan. However, big-data cannot be effectively used without appropriate knowledge of advanced algorithms. Similarly, the mediation of advanced algorithms is potent in applying big-data on the assortment strategies of the retail sector(Figure 2).

DISCUSSION

The findings of this study can be validated and justified through the consistency of results with previous study findings. This study is consistent with all the significant literature contributors in big data and its implications to developing countries and the retail sector. The study postulations are valid as the study is found to be coherent with the findings ofDekimpe (2020) as the study was found to be impactful on the retail sector of Pakistan. Moreover, the study signifies the use of big-data on the assortment strategies of the retail sector and therefore signifies the relationship between assortment strategies and employment of big-data as validated by Silva et al. (2020). However, some divergence is found compared to the indication that anomalies exist, and all the correlations are not like each other. Though, the moderation of skilled data-scientists is not found potent for assortment strategies for the organised retail sector of Pakistan. Hence, the findings differ from previous studies which assert that skillful data scientists are essential in the implementation of assortment strategies in the retail sector. Similarly, it is not improper to highlight the study findings which highlighted the high number of data-scientists accompanied by one of the leading US-Based online retailers employed to analyse customer demand patterns through advanced algorithms. However, these changes might be due to the change of culture and a lack of understanding regarding the technology. Besides, this study accentuates the actuality that advanced algorithms are essential to be analysed to use big-data most effectively to retrieve new information (Chauhan et al., 2017).

FUTURE RESEARCH

The study is one of the initial studies which have taken the research intent through incorporating all the significant variables associated with the application of big-data in the retail sector. However, the sample size was much smaller due to a lack of understanding of the technology and its implications. Hence, assortment strategies are the only variable used to understand the impact. However, some other potent variables are included in-store layouts like aisle design, etc. Therefore, further studies may also be conducted in devising a comprehensive model which includes all the potent variables associated with store-layout design. Moreover, the comparative studies on online and organised retail segments may also be fruitful for increasing knowledge and understanding of the technology concerning Pakistan.

ACKNOWLEDGEMENT

The authors acknowledge the help provided by the women wing of the Karachi Chamber of Commerce and Industry (KCCI), who helped link the researchers with the experts in the retail sector industry of Pakistan. The authors also thank all the respondents for their participation in data collection.