Communication Strategies for AI Recommendations

Refined research questions and designed Experiments-based consumer studies

Tools

Overview

Significance of the Study: From '“What” to Recommend to “How” to Recommend

This research targets the optimization of communication strategies to drive business outcomes by shifting the focus from merely “what” to recommend to strategically refining “how” to recommend. By exploring the impact of different linguistic formats in recommendation explanations, the study aims to enhance the persuasiveness of AI recommendations, ultimately influencing decision-making and driving more impactful business results.

Wrote survey questions and created experimental stimuli

Challenges

Initial Checks

Roles

Led consumer behavior research projects focused on AI recommendation systems and robotic agents, managing the entire research process from planning and execution to data analysis and presentation.

Collaboration with mentors was essential in refining methodologies and ensuring research rigor and relevance. The resulting actionable insights and recommendations were tailored for business firms, supporting data-driven decision-making and strategic improvements. The findings were communicated in an accessible manner, making complex statistical results understandable and useful for non-technical audiences.

Design

Responsibilities

Programmed surveys and designed the correct flow and logic

Most current research on recommendation systems prioritizes algorithm accuracy to generate personalized recommendations, focusing on "what" to recommend.

However, challenges like the "cold-start" issue and training on low-quality data can hinder the delivery of tailored experiences. These limitations underscore the need for more robust approaches to consistently provide relevant and personalized customer experiences.

Tailoring explanations for AI recommendations to the characteristics of recipients can enhance personalized experiences and increase customer engagement, ultimately influencing customer behavior.

This research aims to explore how different explanations for AI recommendations influence customers’ responses depending on different situations and the specific psychological mechanism behind these effects.

Methods

  • 20+ rounds of pretests:
    1 x 2 within-subjects experimental design

  • 3 rounds of formal tests:
    2 x 2 between-subjects experimental design

  • Excel

  • SPSS (syntax command language)

  • M-PLUS

Data Analysis

  • Attention check (Bot check)

  • Manipulation check

  • Realism check

  • Descriptive analysis (e.g., Demographics, Mean for measurements…)

Statistical Analysis

  • t-test

  • ANOVA (Analysis of Variance)

  • ANCOVA (Analysis of Covariance)

  • Regression

  • Moderated Mediation model

  • Structural Equation Model (multiple dependent variables)

Impacts

Objective

Tools

Qualtrics - Programmed surveys including adjusting survey flow and using Java coding functions

Conducted statistical data analysis and generated easily processed visualization

Sample

  • Collected primary survey data from an online panel of 1,000+ U.S.-based adults (18+)

Created and customized reports and slides to effectively present to both technical and non-technical audiences.

Measurements

  • Measured 30+ sets of variables, including but not limited to attitudes, future usage intention, perceived relevance, and control variables (e.g., need for uniqueness)

  • Ensured internal consistency across related survey questions.

Output

  • Created customized reports, ranging from brief 3-minute presentations and 1-page summaries to comprehensive 30-minute presentations and 100+ page reports.

  • Shifted the firm's strategic focus from content recommendations to the language used in those recommendations, driving more meaningful consumer engagement.

  • Enhanced consumers' perceived personalization, significantly boosting their attitudes and behavioral intentions through optimized communication strategies.

  • Achieved a 300% reduction in operational costs, delivering substantial savings and improving overall efficiency.

  • How do customers respond to different AI recommendation explanations?

  • Can certain factors change their responses?

  • What are the underlying mechanisms driving these results?

Questions

Survey (primary research)

Experimental Design