Synthetic Concept Testing: Revolutionizing Product Development with Generative AI
Synthetic Concept Testing: Revolutionizing Product Development with Generative AI
Dr. Chiranjiv Roy VP and Global Head Data Sciences and Applied AI COE, C5i
Sushant Ajmani Vice PresidentGenerative AI, C5i
Silky Rout Vice President - Customer and Market Insights, C5i
Executive Summary
This paper explores the effectiveness of synthetic concept testing, using generative AI (GenAI) and leveraging Microsoft Azure, in consumer goods industries that need rapid innovation and frequent product launches/updates aligned with current consumer needs & preferences. Using an example of solution deployment for a Beauty & Skincare client, the authors show how adopting GenAI-powered synthetic concept testing allows brands to speed up and scale the testing process, iterate faster, gather highly nuanced consumer insights, and significantly cut operational costs, creating a compelling case for its broad application in Beauty and other
Introduction
Concept testing faces the challenges of high costs, lack of data centralization, sample quality, overdependence on DIY tools, agency dependence, longer timelines for meaningful reporting, and the need for extensive consumer data to validate concepts. These methods often lead to delayed product launches and innovations, directly impacting market competitiveness and growth. There is a need to revamp these methods to meet the fast-paced demands of consumer goods industries.
C5i’s GenAI-powered Synthetic Concept Testing solution revolutionizes market research by generating synthetic consumer response data. Synthetic responses mimic natural consumer behavior & preferences, allowing brands to test concepts in real time and gain actionable insights on multiple concepts faster and close to traditional survey responses. By simulating diverse consumer interactions, the use of OpenAI’s GPT-4 enables us to bypass traditional limitations, offering a scalable and flexible solution for concept testing. This methodology allows for faster iterations of product concepts, providing rich insights into consumer preferences while significantly reducing associated costs and overdependence on sample providers.
This paper also describes C5i’s solution implementation for a renowned Beauty & Skincare brand leveraging Azure’s Data and AI services, which demonstrates its effectiveness. Faced with the task of accelerating product development and enhancing market alignment, C5i adopted synthetic data generation to refine the concept testing approach. This transition decreased the client’s go-to-market time by over 40% and improved the precision of market fit analysis, leading to more targeted and successful product launches.
1. The Need for Agility in Skincare Concept Testing
The global skincare market is evolving rapidly thanks to increased consumer awareness, influencer marketing, advanced technology, and health and wellness trends. Consumers are seeking ‘at-home beauty,’ ‘male grooming’ is picking up, and so is the need for ‘clean beauty,’ ‘hybrid beauty,’ ‘holistic beauty,’ and so on. Marketers are pressured to come up with varied solutions to cater to ‘inclusivity’ and diverse skin tones. There is also the pressure of applying AI and using the power of technology to launch innovative products quickly and stay competitive and relevant in the market. Hence, insights teams, too, must cope with the pressure of testing viable concepts with greater efficiencies.
Traditional early innovation tests, such as idea
screens, concept tests, pack tests, and qualitative studies, have limitations. They are costly and time-consuming, and there is overdependence on traditional agencies or agile sample providers to obtain a niche sample, a specific style of reporting, and actionable data and insights.
Synthetic data offers a transformative alternative for innovation research, enabling an agile and responsive platform. Using advanced algorithms to generate artificial yet realistic consumer data, brands can simulate various consumer reactions to product concepts in real time. This method allows for rapid iteration of ideas, testing multiple concepts simultaneously without the logistical and financial burdens of traditional research. Furthermore, synthetic data generation can fill in gaps where accurate data may be sparse or unavailable, providing a comprehensive view of potential consumer behavior and preferences across diverse demographics. This agility speeds up the research process and enhances the depth and accuracy of insights, facilitating quicker, data-driven decision-making and a smoother transition from concept to product launch.
Synthetic data is artificially generated information that mimics real-world data, offering a cost-effective, scalable, and speedy alternative for generating market insights. This allows researchers to perform multiple simulations and stress tests on product concepts without compromising data quality or consumer privacy.
OpenAI’s GPT-4, an advanced language model, enhances synthetic data creation by generating realistic consumer profiles and responses. Its deep learning capabilities allow it to understand and replicate nuances in human communication, enabling it to
2. Synthetic Data: A Primer
produce varied and realistic datasets. These datasets reflect complex consumer behaviors and preferences, providing valuable insights into a product's performance in the market. However, using synthetic data raises ethical considerations, particularly regarding the accuracy and bias of the generated data. Ensuring privacy and fairness involves rigorous validation against real data sets to detect and correct biases. It’s crucial to transparently communicate the use of synthetic data to stakeholders and rigorously adhere to data protection regulations to maintain trust and integrity in research practices. Balancing innovation with ethical responsibility is key to harnessing the full potential of synthetic data while respecting consumer rights and societal norms.
3. GPT-4 and Synthetic Concept Testing: The Technical Implementation
Building a synthetic concept testing platform on the Microsoft Azure stack involves a series of structured phases. The first is integrating cloud infrastructure to support the scalability and computational demands of GPT-4. Azure provides a robust environment with various services for deploying a high-performance AI application.
The process begins by setting up Azure Blob Storage and Data Lake for secure and scalable storage solutions. These services store vast amounts of raw and processed data, from initial consumer insights to synthesized
responses. Azure storage solutions' accessibility and durability ensure that data management is streamlined and efficient, facilitating easy access and manipulation as needed.
GPT-4 is fine-tuned to generate specific consumer feedback by training it on a curated dataset of fundamental consumer interactions, product reviews, and feedback. This dataset includes various nuances of consumer language and sentiment, which helps create a model to generate realistic and contextually relevant responses. The fine-tuning process involves adjusting the model parameters to reduce error rates and improve the accuracy of the output, ensuring the reactions are as realistic as possible.
The data pipeline is crucial in handling data flow through the system. It involves several Azure AI services, including Azure Machine Learning, to orchestrate the machine learning lifecycle from training to deployment. The pipeline facilitates efficient data processing, running computational tasks in parallel, and managing the model's training with new data iteratively. This setup not only maximizes data processing efficiency but also optimizes the computational resources used, thereby reducing operational costs.
By leveraging these advanced technological resources, our Azure-powered synthetic concept testing platform can accurately simulate and predict consumer behavior, providing invaluable insights that drive product development and marketing strategies more effectively.
4. Case Study: A Global Premium Cosmetic Brand Transforms Concept Testing
Introduction and Challenges:
The ability to quickly and effectively test new skincare concepts is crucial in the highly competitive beauty and cosmetics industry. This brand, a trailblazer in premium skincare innovation, faced challenges with traditional concept testing methods, which were often slow, expensive, and restricted by the size and diversity of consumer panels. This usually delayed product launches and resulted in significant expenses without guaranteeing a broad market representation.
Goals of Implementing a Synthetic Concept Testing Solution:
To overcome these obstacles, the client aimed to implement a synthetic concept testing solution blending Market Research Primary Studies knowledge precisely for Concept Testing and Generative AI (GenAI)/Large Language Models (LLMs)-based Agent-based Retrieval-Augmented Generation (RAG) systems. The primary goals were to:
Reduce the time from concept to market by accelerating the feedback loop.
Decrease the costs associated with traditional market research methods.
Increase the diversity and accuracy of consumer feedback by simulating a wide range of consumer profiles with minimal initial training data from existing studies.
Method and Technical Implementation:
The client decided on the sub-categories of Skincare to be tested: Sunscreen, Moisturizer, Luxury, Hydration, etc.
The sample size per concept, audience groups, demographics, questionnaire flow, and key metrics were then discussed with C5i and finalized. The client shared historical concept test results for the categories used to build a robust model. The implementation of the synthetic concept testing solution was structured around creating a dynamic and responsive AI model that could generate realistic consumer reactions to new product concepts. The LLMs and RAG models were trained on a limited set of high-quality, curated datasets, which included historical market studies, consumer reviews, and demographic information. This allowed the system to extrapolate and generate responses from potential customers across different markets and consumer segments.
The success of this new approach was measured using several key metrics:
Time Savings:
Time taken from concept development to market readiness
Concept Iteration Speed:
Agility based on synthesized feedback
DIY:
Ease of use and need for prerequisite knowledge
Qualitative Feedback:
Cost Reduction:
Costs associated with gathering and processing consumer feedback
Accuracy:
Reliability of insights and accuracy compared to traditional methods
Scalability:
For other category concepts, more audiences, languages, and markets
Feedback from the brand's product development and marketing teams was overwhelmingly positive. The product development team appreciated how the synthetic data allowed them to quickly gauge consumer sentiment and iterate on product designs without waiting for lengthy study results. Marketing professionals noted the advantages of accessing a broader range of consumer feedback, which enabled more targeted and effective marketing strategies.
Impact and Insights:
Our solution helped fetch data for each test concept in all the required categories, meeting the required sample sizes (250 responses per concept) at an overall level and even among the required demographics. It clearly showed the winning idea for each category that could be taken to further stages of development. The winning concept was the one that showed significantly higher scores for the most important metrics, such as intention to purchase, distinctiveness, relevance, etc., compared to other test concepts and scored well among the important audience groups. For example, we recommended going ahead with the ‘glow and shield’ concept in the sunscreen category. To check the accuracy of our algorithm, we helped the client deploy the concept test for sunscreens with a panel provider using actual respondents. Post results, the client could see an 80% similarity between synthetic data scores and the scores coming out of the latter with actual respondents across most measures. This proved our credibility, and the client is now ready to deploy the synthetic concept testing solution for more categories, audiences, markets, etc., and has asked to deploy a full-fledged reporting solution. The synthetic concept testing platform revolutionized how the brand approached product innovation. By integrating advanced AI into its concept testing processes, the brand was able to accelerate product development cycles and dramatically enhance product-market fit. The ability to quickly test and adapt concepts based on comprehensive and diverse consumer feedback led to more successful product launches and a stronger competitive position in the market.
The Beauty brand's early successes showcase the transformative potential of synthetic data in redefining traditional market research paradigms. This approach supports faster and more cost-effective product development and aligns closely with the evolving expectations of a diverse
5. Results
Speed:
Obtained data and insights in less than a minute.
Cost-Effective:
Reduced the cost of concept testing by almost 70%.
Accuracy:
Insights showed up to 80% accuracy when compared with real consumer data.
Truly DIY:
The interface is easy to use, with a (pre-set) menu-driven interface that doesn’t require research knowledge. It also offers automated data analysis and downloadable reports, including significance tests and custom charts.
Scalability:
The solution was easily scalable to multiple concepts, product categories, audiences, languages, and markets.
Flexibility:
The solution allows users to customize questions and reporting formats to meet the needs of different organizational roles, from succinct overviews to detailed analyses.
6. Conclusion:
Like many consumer goods industries, the Beauty & Skincare industry is constantly evolving, driven by shifting consumer preferences, technological advancements, and the relentless pursuit of innovation. In this dynamic landscape, the ability to rapidly and accurately assess the potential of new product concepts is paramount. Traditional concept testing methods, while valuable, often need to catch up in terms of speed, cost-efficiency, and the ability to capture the diverse voices of modern consumers.
Synthetic concept testing, powered by
advanced AI models like GPT-4, offers a transformative solution. By simulating realistic consumer reactions to product concepts, brands can gain invaluable insights into potential market success, all while significantly reducing the time and costs associated with traditional research methods.
The case study presented in this white paper demonstrates the tangible benefits that synthetic concept testing can bring to Beauty and other consumer goods brands, from accelerated product development cycles to enhanced product-market fit.
As the industry embraces digital transformation, synthetic data emerges as a powerful tool for unlocking new levels of agility, efficiency, and consumer-centricity in product innovation.
About Us
C5i Limited focuses on helping organizations drive digital transformation using artificial intelligence ("Al"), advanced analytics and insights. C5i's Al-driven products and solutions and IP-led solutions are supported by industry-specific domain experience and the latest technologies and aim at enabling organizations to solve complex issues relating to their customers, markets and supply chain at speed and scale. C5i combines a multi-disciplinary approach to data integration across structured and unstructured data sources to help businesses grow through informed decision-making.
C5i caters to some of the world's largest enterprises, including many Fortune 500 companies. The company's clients span Technology, Media and Telecom (TMT), Pharma & Lifesciences, CPG, Retail, and other sectors. Course5 Intelligence has been recognized by leading industry analysts like Gartner and Forrester for its Analytics and Al capabilities and proprietary Al-based platforms.