AI-supported Tool for Product and Customer Research

This project aims to streamline customer and product research, especially for early-stage product teams that often lack access to large and diverse user samples when exploring new product ideas.

I designed and developed a multimodal AI-powered panel study platform within an intuitive UI that leverages large language models (LLMs), including OpenAI GPT, Gemini, and Llama, to generate configurable synthetic users (personas) for scalable user research. This platform also supports product researchers in curating questionnaires and surveys by providing insights into how users, even at the population level, behave and think.

Background and Goals

We have some key goals in order to guide our user-centered research.

Background

Understanding customer perceptions early in product research and development helps improve user engagement and retention later.
Conducting traditional panel studies involves collecting data through various research activities, such as surveys, interviews, and focus groups, which are resource-intensive and time-consuming.
Recruiting target users with diverse demographics and characteristics can be limited and expensive.

Generative Research Goals:

Research Methods and Process

How effectively can the platform interpret multimodal inputs (e.g., text and images) and generate realistic user feedback?
Which application features enable seamless interactions and support scalable user research?
How closely do synthetic user responses generated by AI match real human responses?

Problem Framing

Reviewed existing research practices and discussed with the marketing team, product researchers, and business units (Safety and Haircare units) to identify research priorities, business impacts, and understand challenges in recruiting diverse participants and conducting large-scale panel studies.

Methods: Literature review and brainstorming sessions.

System & Experience Design

Designed the platform workflows and interface to allow researchers to (1) configure synthetic user personas at specific numbers and characteristics (e.g., age, hair length, race, gender, hair type, etc.), (2) run multimodal research with text and image inputs, and (3) provide both detailed responses and summarized insights to support scalable research analysis.

Tools and methods: Figma, discussions with stakeholders and experts, persona, sketching

Prototype Development

Developed a multimodal AI-powered panel platform using LLMs (GPT, Gemini, Llama) to generate configurable synthetic users and simulate user responses at scale.

Tech: Python, Streamlit, LangChain, LLM APIs, prompt engineering, AI-based summarization methods.

Application Evaluation

Compared AI-generated responses with ground-truth human responses (user-level and product-level) to measure similarity and reliability. Also evaluated the app's ability to generate responses in user studies that involved image analysis.

Findings and Crucuial Insights

Demonstrated that synthetic users can simulate real user feedback
AI-generated text-based responses achieved an average 82% similarity with human responses, suggesting that synthetic panels can support early-stage research when access to real participants is limited.

Enabled diverse research perspectives through configurable personas
Allowing researchers to configure demographic and product-related attributes enabled the simulation of multiple user segments without recruiting participants.

Accelerated insight generation with automated summaries
Providing summarized responses alongside full outputs helped researchers quickly interpret large volumes of synthetic feedback.

Streamlined product and customer research with a multimodal AI-supported tool
Enabling faster and more scalable product research by generating synthetic populations and offering user segmentation to simulate panel studies and evaluate product concepts.

Business Impacts

Accelerated product strategy decisions and user research by providing faster feedback on product concepts and user perceptions through rapid testing with AI-generated synthetic user panels.
Lowered operational costs of panel studies by reducing reliance on recruiting large and diverse participant samples.
R&D team at Procter & Gamble Company

Timeline

May 2024 - August 2024