Customer Experience: How Can Firms Use AI to Predict Share of Wallet?
Paper accepted at the Frontiers in Service Conference 2024 - Zaki, M. and Witell, L., 2024.
Abstract
Effectively managing customer journeys and experiences is key to a company's performance, influencing key outcomes like brand loyalty and share of wallet (SoW) (Becker & Jaakkola, 2020; Zaki, 2019; Wirtz et al., 2007; Cooil et al., 2007; McColl-Kennedy et al. 2019). Customer experiences are increasingly diverse and differ notably across digital, physical, and social touchpoints in the customer journey (Bolton et al., 2022), significantly influencing SoW (Campo et al., 2021). Despite numerous conceptual CX frameworks, there is scarce knowledge on what factors of the customer experience influence key outcomes such as SoW. Further, can firms use AI to manage CX and predict SoW effectively? Firms face challenges in gathering and analyzing CX feedback due to survey fatigue, challenges in accessing holistic CX data, and inconsistencies across various touchpoints (Zaki et al., 2021). These challenges hinder the development of a comprehensive CX management strategy and adaptability to rapidly evolving consumer markets (Bolton et al., 2018).
This paper contributes to the customer experience management literature through showing how AI can be used to model customer experience and how these models can predict key outcomes. First, we adopt and utilise the Touchpoint, Context, Qualities (TCQ) Framework (De Keyser et al. 2020) to understand the various customer journey touchpoints (digital, social, physical), along with factors such as context and qualities, that can impact brand loyalty measures such as SoW. Second, we used natural language models (NLP) and large language models (LLMs) to extract these factors from a customer feedback dataset of a global retailer, encompassing 13,081 customer responses from January 2021 to December 2022 across the US and UK markets. Our (NLP) models successfully extracted 24,135 features encompassing sentiment, emotion, and topic modeling. Subsequently, we employed several machine learning techniques, including Random Forest, Gradient Boosting, and XGBoost, to predict low and high Share of Wallet (SoW) categories. The XGBoost algorithm outperformed other techniques, achieving an F1 score of 84%. Finally, we use Shapley Additive Explanation (SHAP) analysis to identify the top TCQ factors that impact the share of wallet.
Our findings reveal key insights into what factors of customer experiences have an impact on SoW. The most significant factors influencing the model's predictions are the valence dimensions within the Qualities: positive sentiments (35%), neutral sentiments (32%), and negative sentiments (3%). Market context, particularly in the USA and GBR markets, significantly affects SoW, contributing 7% and 6%, respectively. In terms of touchpoints and journeys, physical store, digital (e.g., app and website), and social (e.g. staff) journeys impact SoW predictions with respective contributions of 6%, 4%, and 4%. Dimensionality factors such as Cognitive factors like ease of shopping and emotional factors such as admiration each play a 3% role in influencing SoW predictions. The model also highlights the varied impacts of sensorial and physical dimensions like stores (3%) and restaurants (2%). These findings highlight strategic implications for firms on leveraging AI to manage and understand factors affecting Share of Wallet.
Authors: Mohamed Zaki and Lars Wittell
