Fighting demand uncertainty in e-commerce business
- Marie-Antoinette Mbaye
- 23 oct. 2023
- 6 min de lecture
Most of the managers know how inventory management and forecasting are affecting the supply chain agility. This is more true in dynamic but competitive environment like the e-commerce that requires a dedicated supply chain.
In a context where partners are sharing forecasting and data, partners must be sure of the reliability of the information. Most of the time, we do not know what the parameters and the algorithms are. This leads us to our previous statement: the e-commerce supply chain requires its own method and analyses, not a copy and paste from another situation. Precisely, demand planners should build a forecasting method in e-commerce taking in consideration qualitative methods and branding specificities.
In their research Khodabandehlou, S. and Zivari Rahman, M. (2017)[1] analyzed the shifting perspective of companies. From product orientation, we went to customer satisfaction because understanding the buying pattern prevents lost sales for companies. In doing so, churn techniques which describes the situation where a company customer become attractive to another company, are increasing the cost of retention of the customer. In 2003, Chiang et al. esteemed that companies lose on an annual average of 25% of their customers. In this competitiveness, we understand that products with a high buying rotation are the first ones to be concern by this phenomenon.
Firms should invest in business intelligence and machine learning for Huang et al., (2010); Ivanovic et al., (2011); Rygielski and al., (2002), to reduce customer churn. It will provide insight on:
Customer behaviors that will help predict patterns
Customer trends that will help for the Customer Relationship Management (CRM).
Commercial businesses that will provide decision making patterns (Devi Prasad and Madhavi, 2012; Hung and Wang, 2004; Ngai et al., 2009)
Machine-learning techniques have been effective to predict the customer churn, but still have a divergence of results such as:
Logistic Regression (LR)
Artificial Neural Networks (ANN)
Decision Trees (DT)
Support Vector Machine (SVM)
Among them, the Recency Frequency and Monetary data mining methods (RFM variables) has the best predictions for distinguishing loyal and non-loyal (potential churners) customers (Buckinx and Van den Poel, 2005; Coussement and De Bock, 2013; Dursun and Caber, 2016; Tamaddoni Jahromi et al., 2014; Wei et al., 2010). However, to predict customer loyalty, other variables should be considered. Khodabandehlou, S. and Zivari Rahman, M. (2017), took five factors to develop the RFMITSDP model :
The number of purchased items
The number of returned items
The discount amounts
The distribution times
The number of prizes
Having a customer behavior analysis could help understanding the buying pattern and prevent loss sales. The authors summarized the performances of the methods by needs:
The boosting method is used for the prediction of customer behavior based on data, which increases efficiency and accuracy of prediction.
The ANN method is used for the prediction of customer behavior on the churning status.
The Discriminant analysis method extracts and selects the important and effective variables influencing customer behaviors which predict the classes of the customers.
When discussing online sales problems, it is interesting to have a look at methods implemented by other professionals. In an article published on Demand-Planning.Com[2], Daniel Fitzpatrick recognized that approaching the demand planning for online business with the same tools as the physical retail is not effective. Learning from his experience, Daniel Fitzpatrick explains that the first key to success demand planning in online sales is to really know the customer. In front of a computer, the customer can be located everywhere. Companies have, therefore, to develop tracking tools that can help to assess the customer preferences, motivation or reluctance to buy. As described previously, digital solutions such as CRM or IA can assess online customer data for supply chain purposes.
Those technologies have the advantage to treat an important number of data while recording different customer patterns, but not only. Based on Akinyemi Paul, and Audrey Frances Laing’s researches (2016), the digitalization of the buying process can influence customer behavior. Indeed, if we integrate two technologies with different purposes, like the ERP and the CRM, we could increase customer acquisition, then influence and predict their behavior. If we take for example the customer retention and loyalty, coupling an ERP and a CRM, we will most likely reduce the churn problematic presented by Khodabandehlou, S. and Zivari Rahman, M. (2017). As the retention cost is very high for online business, influencing customer behavior through digital tools decreases forecasting risk.
We saw that companies could adopt strategies to change behavior rather than facing uncertainty. We can take another example with the changing buying behavior in fashion retail. The decrease of the purchasing power and the research of the “best price” by the clients, pushed retailers to increase their number of sales and promotions. For example, in France from two legal sales periods, the country went to four, without considering sales events like The Black Friday or Christmas promotions.
After a couple of years with an increasing number of sales, and consequently, of promotions, French customers are now used to low-price products. This explains why the average buying basket is declining in retail. This concrete example shows that consumer behaviors can be impacted by the political, economic, or social environment, but also by companies’ strategies.
As we are all customers in our personal life, Daniel Fitzpatrick invites the managers to rethink the situations that disrupt forecasting from the consumer perspective. Since extreme customer’s behavior have the power to distort a “normal” behavior in the data, they should be monitored closely by the managers. Managers should, therefore, monitor hoardings (situation “where customers buy large quantities of a product to restrict availability and control pricing”), compulsive customers who are able to buy everything available even if they do not need, or any behavior that have the power to disrupt forecasting.
To depict this topic, we can talk once again about the COVID-19 outbreak. By fear of lacking essential supplies, we saw customers developing hoarding habits during the outbreak in March-April 2020. It then causes a massive shortage in retail shops for essential products like paper toilets or pasta because the demand was superior to the supplying capacities. Forecasting demand during such a crisis was a brain teaser as the habits were decided out of fear and were not rational. This experience shows that monitoring extreme behavior can be helpful. In the case of the COVID-19, extreme consumer behaviors became “the new normal” when customers were responding according to the pandemic progression.
So, demand can be influenced by product characteristics (price, new listing and de-listing), promotions and sales, branding (customers and influencers reviews, and brand visibility), and stock availability (from the brand and its competitors), and the potential crises. When it comes to quantify those elements, it often falls under the product owner spectrum because it can be difficult for a demand planner to assess how a bad review can impact other consumer buying decisions per example.
However, the strength of the online market is that it requires an entire collaboration between internal and external partners. If we supposed that only marketing departments are aware of the e-reputation of one product, we could not anticipate a lift of sales after a social media post or a sales decrease after a bad review on a blog.
The key to success while forecasting for Daniel Fitzpatrick is to focus on the items that bring the most value. If the forecast accuracy is good for 20% of the items that represent 80% of the values of the company, demand planners have more chance to improve their techniques for the 60% products remaining. However, we must counterbalance this point of view. In some contexts, the low sales of items are due to a low service level. Indeed, non-availability and long shipping times can reduce the amount and sales. But, if the supplying aspect is improved, that kind of item might thrive to the fullest of their capabilities.
To conclude this topic, demand planners working on e-commerce products have tools and methods developed for the physical environment. But when it comes to the online world they must innovate and create processes that are resisting the risks, uncertainty and non-rational behavior. The key to forecasting in such a context is to have a deep understanding of all those elements thanks to the collaboration between product owners and supply planners.
Sources
[1] Khodabandehlou S., and Zivari Rahman M., (2017), ‘Comparison of supervised machine learning techniques for customer churn prediction based on analysis of customer behavior’, Journal of Systems and Information Technology, 19(1/2), p.65-93, [online], available at https://doi-org.library.ez.neoma-bs.fr/10.1108/JSIT-10-2016-0061 (accessed 25 July 2020). [2] Fitzpatrick D., (2020), Demand planning for the e-commerce channel, [webpage], available at http://demand-planning.com/2020/06/15/ecommerce-demand-planning/ (accessed 30 July 2020). [3] Jonhson G., June (2017), CRM benefits : influencing customer behavior,[online], available at https://www.toolbox.com/tech/erp/blogs/crm-benefits-influencing-customer-behavior-053117/, accessed the 30/07/2020
Commentaires