O Nudging Baseado em Inteligência Artificial Pode Promover a Adoção de Produtos Sustentáveis em Plataformas de E-Commerce?

Panwar, Nikita; Shetty, Dasharathraj K.; Shenoy, Sandeep S.

doi:10.1590/1982-7849rac2025250105.en

Tabla de contenido Anterior Actual Siguiente

Theoretical-Empirical Article • Rev. adm. contemp. 29 (06) • 2025 • https://doi.org/10.1590/1982-7849rac2025250105.en linkcopiar

O Nudging Baseado em Inteligência Artificial Pode Promover a Adoção de Produtos Sustentáveis em Plataformas de E-Commerce?

Autoria SCIMAGO INSTITUTIONS RANKINGS

Objetivo: o estudo visa identificar intervenções comportamentais eficazes e modelos de aprendizado de máquina que promovem compras ambientalmente sustentáveis.

Marco teórico: o estudo está fundamentado na economia comportamental e na arquitetura de escolhas, apoiando-se na Nudge Theory (ou Teoria do Nudge) de Thaler e Sunstein, de forma a explicar como elementos sutis de design digital podem servir de orientação para escolhas sem comprometer a autonomia.

Método: foi realizada uma revisão sistemática utilizando o protocolo PRISMA e múltiplas bases de dados acadêmicas. Observa-se o papel da inteligência artificial (IA) no marketing para a sustentabilidade, a efetividade dos modelos e as implicações éticas.

Resultados: técnicas de nudging baseadas em IA, tais como personalização, default green options (ou opções verdes predefinidas), prova social e gamificação, aumentam significativamente o engajamento em comportamentos de consumo sustentável. Modelos de aprendizado de máquina como Random Forests, Deep Neural Networks e Reinforcement Learning têm um papel crucial na otimização do nudging por IA para o consumo sustentável.

Conclusões: o estudo mostra o potencial das técnicas de nudging baseadas em IA podem influenciar o comércio eletrônico sustentável, sobretudo quando empregadas com sistemas transparentes, explicáveis e conscientes de vieses. O estudo contribui para consolidar as principais pesquisas sobre nudge assistido por IA, compara modelos de aprendizagem de máquina voltados a consumidores verdes e destaca os trade-offs entre desempenho e capacidade explicativa, propondo um desenho experimental para avaliar eficácia e ética. Oferece implicações práticas para empresas e formuladores de políticas ao conectar estratégias impulsionadas pela IA aos Objetivos do Desenvolvimento Sustentável 12 e 13 e recomenda a adoção de estruturas éticas de IA para fortalecer a confiança voltados à sustentabilidade em plataformas de comércio eletrônico Pesquisas futuras podem ter como foco experimentos reais, estruturas interdisciplinares e regulações éticas, todas relacionadas à IA, para garantir a adoção responsável dessa inteligência no comércio eletrônico.

Palavras-chave:
inteligência artificial; técnicas de nudging digital; e-commerce; consumo sustentável; Objetivos de Desenvolvimento Sustentável 12 e 13

Model	Best use case in green nudging	Strengths	Weaknesses	Interpret-ability
Random forest (RF)	User classification, eco-purchase prediction, and feature importance-driven nudge design	High predictive performance, robust to noise, moderately interpretable using feature importance, SHAP compatible	Relies on post-hoc tools; slower with very large datasets	Moderate
Support vector machine (SVM)	Classification of eco-responsive users; decision boundary-based nudge personalization	Effective in high-dimensional spaces, kernel flexibility, defines clear decision boundaries	Lower interpretability, sensitive to noise and scale; requires careful tuning for optimal performance	Low to moderate
k-nearest neighbors (k-NN)	Eco-product recommendation via behavioral similarity; clustering for tailored nudges	Intuitive and easy to implement; well-suited for similarity-based personalization; no training required	Limited scalability, sensitive to irrelevant features, reduced efficiency in high-dimensional data	Moderate
XGBoost	High-accuracy behavior prediction and scalable personalization with explainability support	Demonstrates excellent performance on structured data, effectively captures complex feature interactions, and provides fast, precise SHAP-based interpretability	Moderate interpretability without SHAP; risk of overfitting without tuning	Moderate
Reinforcement learning (RL)	Adaptive and personalized nudge optimization through long-term behavior learning	Continuously learns optimal strategies and adapts behavior to personalized, long-term objectives	High data requirements, complex implementation, critical dependency on reward function design	Low
LSTM / RNN	Sequential behavior modeling and session-aware green nudging based on temporal patterns	Captures long-term dependencies and behavioral sequences for time-aware interventions	Data- and compute-intensive; lacks native interpretability; complex model understanding	Low
Decision tree	Transparent, rule-based nudging and interpretable decision-making for sustainable choices	Highly interpretable logic; enables transparent and ethical nudging strategies; easy to deploy	Prone to overfitting; lower accuracy than ensemble models; limited generalization capacity	High

Behavioral psychology principle	AI nudging technique	Applications in e-commerce	Effectiveness
Social norms and peer influence	Social proof nudging	‘X number of users bought this green product today’	Effective but excessive use may cause negative perceptions (Mirbabaie et al., 2023).
Default bias	Pre-selected green options	Auto-selecting eco-friendly delivery options at checkout	Increases sustainable behavior but may create negative perceptions (Michels et al., 2022).
Personalization and habit formation	AI-powered recommendations	Suggesting eco-friendly alternatives based on past purchases	Highly effective in increasing green product adoption (Lo & Lin, 2024).
Loss aversion and reward sensitivity	Gamification and incentive-based nudging	Points for choosing sustainable shipping	Strong motivator for consumers when egoistic rewards are involved (Hollaus & Schantl, 2022).
Information processing and cognitive load reduction	Real-time sustainability information	Carbon footprint label next to product details	Enhances awareness and reduces price sensitivity (Verboom, 2021).

Privacy concern	Description	Best practice	Real-world example
Lack of transparency	AI collects and analyses consumer data without clear disclosures.	Platforms must provide clear explanations of how AI-driven personalization utilizes user data.	Bias in the ‘Buy Box’: Amazon has faced criticism for lacking transparency in how it uses data from third-party sellers and in the way it selects winners for the Buy Box. Limited visibility into these processes raises concerns about fairness and accountability on the platform (European Commission, 2020).
Algorithmic manipulation	AI may prioritize company profits over consumer interests.	Regulatory oversight and transparent guidelines for algorithmic ranking is needed to ensure fairness.	Search and ranking bias: There have been reports and speculation that certain large e-commerce platforms, like Amazon, may have given preferential placement to their own private-label products in search results, which has raised concerns about fairness and competition (Jeffries & Yin, 2021).
Cybersecurity risks	AI-powered recommendations can be targeted in data breaches.	Stronger security protocols and regularly auditing data management practices should be enforced.	Data breach: Shein experienced a significant data breach in 2018, affecting over 39 million users, highlighting vulnerabilities in recommendation-driven data collection (BBC News, 2022).
Consumer trust issues	Hidden AI decision-making reduces confidence in nudging.	Explainable AI (XAI) can improve transparency.	Erosion of trust through delayed disclosure: Shein faced criticism for downplaying a major data breach and failing to notify most affected users. This lack of transparency weakened customer trust (BBC News, 2022).

Bias type	Impact	Solution
Popularity bias	Over-promotes well-known green brands, limiting visibility for small eco-businesses.	Use diversity-enhancing algorithms to support smaller brands.
Data bias	Reinforces past non-sustainable shopping habits.	Re-train AI with ethically curated sustainability data.
Marketing bias	AI over-promotes trendy ‘green’ products rather than truly sustainable options.	Implement fact-checking AI models to validate green claims.
Feedback loop bias	AI keeps showing similar eco-products, reducing diversity in recommendations.	Introduce dynamic AI models that explore new sustainable alternatives.
Demographic bias	AI limits green product recommendations based on income level.	Ensure fair AI models that provide access to green choices for all consumers.

Factor	AI-driven nudging	Traditional marketing
Personalization	Uses real-time data for tailored recommendations.	Relies on broad segmentation and mass marketing.
Efficiency and automation	Automates green choices, reducing decision fatigue.	Requires human intervention for campaign updates.
Scalability and cost-effectiveness	Expands sustainability efforts without additional marketing costs but is often costly to implement and maintain.	Costly due to manual campaign execution. Simpler implementation for small businesses.
Privacy concerns	Requires extensive consumer data collection, raising transparency issues.	Less reliance on personal data, but lower personalization.
Bias and manipulation risks	Can reinforce biases or greenwashing if not properly monitored.	Uses fixed messaging, reducing dynamic bias.
Consumer trust	Needs transparency to maintain trust in AI-generated nudging.	More straightforward but less adaptive to consumer needs.
Technical complexity	Requires advanced infrastructure, skilled talent, and real-time data.	Technologically simple and more accessible.
Adaptability in low-data settings	Less effective without large datasets.	Functional even with limited or no data.

Future research topic	Research focus
Empirical validation of AI nudging	Conduct real-world experiments (e.g., randomized controlled trials) to measure the effectiveness of AI-driven nudges on sustainable consumer behavior.
Bias-aware AI model development	Design and evaluate AI frameworks that detect and mitigate bias in nudging, particularly across demographic groups.
Explainable AI for nudging	Investigate how explainable AI (XAI) affects consumer trust, understanding, and acceptance of sustainability nudges.
Self-nudging mechanisms	Explore user-controlled nudging systems where consumers can personalize or opt-in to the type and intensity of nudges.
Voice-assisted and inclusive interfaces	Examine how voice-based or simplified user interfaces can improve accessibility of AI nudging for non-tech-savvy users.
Income-sensitive AI strategies	Develop models that recommend affordable sustainable options tailored to low-income consumer segments.
Blockchain for sustainability verification	Assess the effectiveness of AI-blockchain integration in verifying green product claims and reducing greenwashing.
AI nudging across cultures	Compare how cultural or regional factors affect the perception and effectiveness of sustainability nudges.
Energy efficiency of AI models	Evaluate and compare the environmental costs (e.g., carbon footprint) of various AI models used for nudging.

	1	2	3	4	5	6	7	8	9	10	11	12
1^st round
2^nd round
3^rd round

location_on

Associação Nacional de Pós-Graduação e Pesquisa em Administração Rua Carneiro Leão, 825,, CEP 87014-010, Maringá, PR/Brasil, Tel. (55 44) 3354-8545 - Maringá - PR - Brazil
E-mail: rac@anpad.org.br

rss_feed Acompañe los números de esta revista en su lector de RSS

[TFN1] Note. RF = Random forest; SVM = Support vector machine; k-NN = k-nearest neighbors; XGBoost = Extreme gradient boosting; RL = Reinforcement learning; LSTM = Long short-term memory; RNN = Recurrent neural network.