mos/ai/c
Challenge:
In the ceramics industry, much like many creative sectors, the product design process often lacks structure and is driven largely by intuition. This absence of a formal methodology makes it difficult to predict a product’s market potential or assess the effectiveness of design decisions early in the process. The challenge was to bring more precision and data-driven insights into the traditionally subjective realm of tile design. Collaborating with the Spanish Ceramic Tile Manufacturers' Association (ASCER), the project aimed to analyze how tile prototypes are perceived before market launch, leveraging computational tools to align product development with market trends and expert evaluations.
Outcome:
The project successfully introduced a pioneering AI-driven tool that provides manufacturers with clear insights into the positionality and potential success of ceramic tile designs. By analyzing trend data and gathering expert feedback, the tool enables manufacturers to refine their products iteratively, ensuring they are closely aligned with current market needs and preferences. This approach not only optimizes the design process but also fosters a symbiotic relationship between human expertise and computational precision, setting a new standard for innovation in the ceramics industry.
Role: As Project Manager, I led the design discovery process, conducting bilingual user interviews and translating insights into actionable solutions. I also developed the project’s front end and ensured the AI tool addressed industry needs effectively.
-
9 months
-
MaP+S Research Group
Martin Bechthold
Jose Luis Garcia del Castillo
Maggie Chao
DESIGN RESEARCH:
Phase I: Conducted 11 open interviews with key stakeholders in the ceramics industry, including small and large enterprises, glaze makers, and trend analysts. The goal was to gain insights into the decision-making processes and innovation challenges within the industry
Phase II: A system map and user journey were created to outline the industry's design process. Data gaps were identified, and relevant data sources were collected to train the AI model.
INSIGHTS
The design process in the ceramics industry is highly intuition-based, with significant input from sales departments.
Companies often integrate client-initiated designs into broader collections, leveraging external innovation.
Success metrics vary, with some companies focusing on volume, while others prioritize brand impact or niche innovations.
An AI tool that predicts early-stage success could reduce the risk associated with design decisions.
AI for Success: Training Beyond Traditional Metrics
Creating an AI model for a concept as qualitative as success in the tile industry posed unique challenges. Success in this context couldn't be measured by traditional metrics like sales data, as selling volume is not always a direct indicator of a product’s impact or desirability. For instance, certain tiles serve as “hook” products designed to attract attention, even though their sales may not reflect their influence. This nuance required a deep understanding of what we wanted the model to achieve, how to train it accordingly, and what specific questions to ask. Developing a reliable model meant carefully curating a database that went beyond typical metrics, incorporating diverse data points from trend analysis, design evaluations, and expert opinions. This approach—tailoring AI to qualitative insights—can be adapted and applied to future projects where success is multifaceted and difficult to quantify.
SYNTHESIS
The research was synthesized using a multi-factorial measurement of success, where design trends, market appeal, and user desirability were key components. A scoring system was created to evaluate the potential success of tile products, considering various factors such as trendiness, universality, and market demand.
DESIGN SOLUTION
One of the biggest challenges in the first iteration was deciding how, and whether, to display the model's accuracy. Since the model was being trained to predict something as subjective as taste, there were high discrepancies in the data, making it difficult to visually represent confidence in the predictions. This image captures one of our initial attempts at tackling this data visualization problem, as we experimented with different ways to communicate the model's variability and uncertainty.
Feature 1: Sales Volume Prediction: This feature enabled manufacturers to assess how well a product might sell based on data from past trends and current market conditions, helping them decide which products to prioritize for production.
Feature 2: Style and Trend Analysis: A built-in algorithm analyzed the "trendiness" of a product, offering design managers insight into how fashionable or appealing a tile might be to current markets.
Feature 3: Cross-Market Appeal: The tool also included a universality score, which predicted the product’s popularity across different geographic markets, assisting manufacturers in targeting specific regions.
