Project related to full-time master’s degree Fashion and Luxury Business, Major of the Master in International Management, Amsterdam and Milan

My Prompt

FIRST PROMPT (text): Provide an overview of the issues caused by overproduction in the global fashion industry. Describe an innovative technological practice that can be implemented within corporate production processes to minimize its environmental impact, explaining in detail how it operates. The exposition must be technical in order to maintain a scientific approach. Then, provide your idea on how to integrate and improve these processes. Use supporting sources in your discussion.
SECOND PROMPT (visual): Create an image that symbolizes sustainable fashion production using advanced technologies such as machine learning to minimize overproduction. The scene should blend elements of fashion and technology, incorporating eco-friendly materials and practices. Include aspects of nature to emphasize environmental consciousness and the positive impact of reducing waste and pollution.
Create then a second illustration showing the impact of water pollution caused by the textile overproduction in the fashion industry. On the left there should be a polluted, industrial area with heaps of discarded textiles and dark smoke. On the right, a clean river with lush greenery, representing a transition towards a more sustainable future.
Lastly, generate a compelling and visually engaging video that encapsulates the main concepts presented in your written discussion. The video should effectively convey the key points and ideas, using appropriate visuals, animations, and text to enhance understanding and retention.

Ai Tools: ChatGPT4

Smart fashion production: prediciting demand to minimize waste

The fashion market sector is one of the largest and most dynamic industries globally, with an estimated global market value of around 3 trillion dollars and 50 million employees in 2023. However, behind this prosperity lies a worrying reality: fashion is one of the most polluting industries in the world, responsible for serious environmental problems and enormous waste throughout the entire production and distribution chain.
One of the most severe problems in the fashion sector is overproduction. According to a report by the Ellen MacArthur Foundation, around 100 billion garments are produced globally each year, but market demand is lower than the sector’s supply. This excess production leads to enormous amounts of unsold clothing, often ending up in landfills. It is estimated that 30% of the clothes produced annually are never sold and that around 92 million tons of textile waste are generated globally each year. The Ellen MacArthur Foundation estimates that less than 1% of the materials used to produce clothing are recycled into new garments, while 73% of the materials end up in landfills or are incinerated, causing a huge waste of resources.
The fast fashion model adopted by many large clothing chains such as Zara and H&M encourages rapid consumption cycles and frequent purchases. This model is based on new collections continuously introduced, sometimes every week, pushing consumers to buy and dispose of clothes at dizzying rates.

A 2016 McKinsey study highlighted that the average American buys 60% more clothes than 15 years ago but keeps them for half the time. The McKinsey & Company report notes that the number of collections introduced annually by some fast fashion companies has increased from two to 24 in recent decades. Every second, the equivalent of a truckload of textile waste is burned or buried in landfills, according to the Ellen MacArthur Foundation.
The textile sector is also one of the major contributors to water pollution.

Textile production requires enormous amounts of water and generates polluted wastewater. According to the World Bank, the textile industry is responsible for 20% of global industrial water pollution.
Dyeing and finishing textiles use toxic chemicals that are often discharged into watercourses without adequate treatment, contaminating rivers and lakes. Producing a single pair of jeans, for example, requires about 7,500 liters of water, the equivalent of one person’s water consumption for seven years, as reported by the WWF. Textile factory wastewater contains hazardous chemicals such as heavy metals, azo dyes, and solvents, which can cause significant damage to aquatic ecosystems and human health.

The fashion sector is also a significant contributor to global greenhouse gas emissions: it is responsible for about 10% of global carbon emissions, surpassing the emissions produced by international flights and maritime transport combined.
This is mainly due to the production of synthetic fibers like polyester, derived from petroleum, and the energy-intensive processes of production and transport. Polyester production emits about 706 billion kilograms of greenhouse gases annually, according to the Ellen MacArthur Foundation.
Additionally, clothing production requires enormous amounts of natural resources. Cotton cultivation, for example, occupies about 25% of the world’s arable land but uses 16% of global insecticides and 6% of pesticides, contributing to soil degradation and biodiversity loss. Cotton production is highly dependent on water and pesticides. To produce just one kilogram of cotton, up to 10,000 liters of water may be required, according to a WWF report.
So, how can the environmental impact of this economically important yet planet-harming sector be reduced?
With the advent of new technologies, it is possible to address these problems innovatively and effectively. Among these technologies, machine learning and artificial intelligence (AI) systems emerge as powerful tools to improve the accuracy of market demand predictions, drastically reducing waste and making production processes more sustainable.
Machine learning is a branch of artificial intelligence that allows systems to learn from data and improve their performance over time without being explicitly programmed. In the context of fashion, machine learning can be used to analyze vast sets of historical, current, and predictive data to accurately forecast future product demand, enabling companies to adapt production based on actual market needs, avoiding overproduction, and reducing waste. This not only decreases the costs associated with production and unsold inventory but also minimizes the textile waste ending up in landfills.
Moreover, lower production (and therefore production that does not generate an excess market supply) implies optimal use of natural resources; for example, avoiding the overproduction of clothing reduces the demand for raw materials such as cotton and synthetic fibers, consequently decreasing water consumption and greenhouse gas emissions associated with the cultivation and production of these materials.
Machine learning can also help optimize logistics and the supply chain. With more accurate demand forecasts, companies can better plan production and distribution, reducing unnecessary transport and associated carbon emissions. This contributes to creating a greener and more efficient supply chain.

To better understand how machine learning and artificial intelligence tools can be used and implemented by companies to obtain accurate market demand forecasts, let’s see the phases on which these processes applied to corporate production systems are based:

The data collection phase is fundamental for building a machine learning-based demand forecasting model. By using various data sources and collection techniques, companies can obtain a complete and accurate picture of market trends, customer preferences, and inventory dynamics. Once cleaned and prepared, the collected data can be used to train machine learning models that accurately predict future demand, helping to reduce overproduction and improve sustainability:

  • Historical Sales Data: Collecting historical sales data is essential to understand past trends and customer preferences. These data come from corporate sales management systems, such as ERP (Enterprise Resource Planning) software or cash registers (POS), and from monthly and annual sales reports. Factors analyzed include sales volumes, that is to say the number of units sold for each product, which helps identify seasonal trends; these represent recurring patterns showing how sales vary in different seasons or periods of the year, such as the increase in sales of winter clothing during colder months. The presence of promotions and discounts is also considered, i.e., the impact of promotional campaigns and discounted sales on total sales. It is essential to distinguish between regular and promotional sales to avoid distortions in forecasts.
  • Inventory Data: Inventory management is another crucial aspect. Data on current stock levels, i.e., the quantities of products available in the warehouse, depletion rates representing the speeds at which products leave the warehouse, and lead times, i.e., the time needed to restock products from the time of order, come from inventory management systems (WMS) or ERP. These data help respectively avoid overproduction by maintaining optimal inventory levels, predict the necessary replenishment times, and better plan production. For such data collection, automatic identification technologies like barcode and RFID are used to continuously update inventory levels and generate periodic warehouse reports that provide a detailed view of inventory and warehouse movements.
  • Customer Information: Customer information offers a deeper understanding of personal preferences and purchasing behaviors. These data come from CRM systems and e-commerce platforms. Customers are profiled by obtaining demographic information such as age, gender, geographical location, and purchase history. These data help segment the market and predict specific demand for each segment. Additionally, purchase preferences are recorded, identifying preferred products and shopping habits; for example, some customers might prefer online shopping over physical stores. Finally, online browsing behavior data are collected: visited pages, time spent on each page, scroll speed, viewed products, and cart abandonment. The collection is done through web analytics tools to monitor customer behavior on specific websites and feedback and surveys.
  • Social Media Interactions: Social media interactions provide valuable information on consumer perceptions and emerging trends. The main social media platforms that allow data collection are Facebook, Instagram, Twitter, Tiktok, and Linkedin. Conducting sentiment analysis is fundamental: analyzing the emotion in posts, comments, and reviews helps understand how customers perceive products. The popularity of trends is also considered, i.e., identifying trending topics and popular products through mentions and hashtags analysis. Social listening tools like Hootsuite or Brandwatch are used to monitor social media conversations and sentiment analysis through APIs directly provided by social platforms.
  • Online Search Data: Online search data show consumer interest in specific products. Monitoring search frequency and geographical trends helps identify different demand segments, detecting local and regional fluctuations and trends. Google Trends and other SEO optimization tools like SEMrush or Ahrefs are used to monitor search popularity over time and identify peaks of interest by analyzing the most searched keywords.

Feature engineering is the process of transforming raw data into a format that can be more effectively used by machine learning models. This phase is crucial because the quality of the features used directly affects the model’s accuracy and performance.
Each type of data, from historical sales volumes to social media interactions, can be transformed into useful features that help predict market demand and reduce waste.
Historical sales data can be transformed into features that capture seasonal trends, promotional peaks, and long-term sales patterns. For example, a feature could represent average monthly sales or the quarterly sales growth rate. Two techniques are mainly used:

  • Moving Averages: Calculating moving averages to smooth fluctuations and identify trends.
  • Seasonal Indicators: Creating dummy variables to represent the months of the year, thus capturing the seasonal effects on sales.
    If a company, for example, notices that sales increase by 20% every December, a feature could represent this seasonality, improving the model’s ability to predict similar increases in the future.
    Regarding stock levels, they can be transformed into features that indicate the likelihood of depletion or the need for replenishment. This can help prevent both stockouts and excess inventory. A feature could be the ratio between the current stock level and the depletion rate, indicating how long it will be before the product runs out.
    Customer information such as purchase preferences and online browsing behaviors can be transformed into features that predict the likelihood of purchase for each market segment. Analyses include:
  • Market Segmentation: Dividing customers into segments based on similar behaviors, such as purchase frequency, product preferences, and average order value.
  • Online Engagement: Measuring customer interaction with the website or mobile app, such as time spent on each page or the number of repeat visits.
    A feature could thus represent the average time a customer spends browsing products in a specific category, indicating their interest and likelihood of future purchase.
    Social media interactions can be transformed into features that reflect customer sentiment towards products and emerging trends. The techniques used in this context include:
  • Sentiment Analysis: Using natural language processing (NLP) algorithms to analyze the tone and emotion of reviews and comments.
  • Popularity Indicators: Creating features that represent the frequency of product mentions or the use of specific hashtags.
    A feature could be the average sentiment score of comments on a new product, providing insights into its market acceptance.
    Online search data provide direct indications of consumer interest. These data can be transformed into features that predict demand peaks. Investigated aspects include:
  • Search Frequency: Measuring the search volume for specific product terms over time.
  • Geographical Trends: Analyzing searches by region to identify local variations in demand.
    A feature could be the search volume for a specific product in the last four weeks, indicating growing interest and potential future demand.

Selecting and training machine learning algorithms are necessary to create accurate and reliable predictive models. This phase involves choosing the most suitable algorithms, training the models with the data prepared in the feature engineering phase, and validating their performance.
Choosing the algorithm depends on the characteristics of the data and the project’s objectives. There are various types of algorithms, each with its strengths and weaknesses:

  • Linear and Non-Linear Regression: Linear regression is a statistical method used to model the relationship between a continuous dependent variable and one or more independent variables. It is suitable for data with simple linear relationships.
    Specifically, simple linear regression is used when there is only one independent variable, while multiple linear regression is used if there are multiple independent variables. Linear regression can be used, for example, to predict product sales based on price and advertising spending.
  • Decision Trees and Random Forests: Decision trees are graph-like structures that model decisions and their possible consequences, including random outcomes, resource costs, and utility. Random forests are a set of decision trees that improve prediction accuracy by reducing overfitting.
    Specifically, a decision tree represents a single tree that divides data based on predictive variables, while a random forest is a set of decision trees working together to improve the model’s robustness and accuracy.
    Random forests can be used to predict product demand in different geographical regions based on a range of factors such as seasonal trends and demographic data.
  • Neural Networks: Neural networks are computational models inspired by the functioning of the human brain. They are particularly useful for identifying complex patterns in data and are very effective for accurate predictions.
    There are artificial neural networks (ANN) consisting of layers of nodes (neurons) that transform data into non-linear features, and convolutional neural networks (CNN) used mainly for image recognition but adaptable for time series analysis.
    Neural networks are used to analyze historical sales data and identify patterns in the collected data that could influence future demand.
  • Support Vector Machines (SVM): SVMs are supervised learning algorithms used for classification and regression. They are particularly effective in high-dimensional, multivariate spaces and are useful for identifying clear decision margins between classes.
    Techniques include linear SVM, used for linearly separable data, and non-linear SVM, used for data requiring non-linear kernels for separation. SVMs are useful for classifying customers into different market segments based on their purchasing behaviors.
    Once the algorithm is selected, the next step is model training. This involves using historical data to teach the model to predict future demand.
  1. Data Preparation: The collected and transformed data in the feature engineering phase are divided into two sets: one for training and one for testing. A common split is 80% of the data for training and 20% for testing.
  2. Model Training: The model is trained using the training set. During this phase, the algorithm learns the relationships between the features and the output variable (demand). For example, a linear regression algorithm would learn to predict future sales based on variables such as price, advertising spending, and seasonal trends.
  3. Model Optimization: During training, optimization techniques are used to improve model performance, including:
  • Regularization: Technique to prevent overfitting by adding a penalty term to the model.
  • Cross-Validation: Method to evaluate model performance on different test sets, improving the model’s generalizability

After training, the developed predictive model is validated using the test set to verify its accuracy. This step ensures that the model performs well not only on the data it was trained on but also on new, unseen data. Once validated, the model must be continuously monitored to maintain its effectiveness over time.
For proper validation, the dataset is divided into two main parts: a training set and a test set. The training set is used to train the model, while the test set is used to evaluate its performance. Specifically, cross-validation is used, a technique that divides the data into k subsets (folds).
The model is trained on k-1 folds and tested on the remaining fold; this process is repeated k times, with each fold used once as the test set, and the average results provide a robust estimate of the model’s performance. For example, in 10-fold cross-validation, the data are divided into 10 parts. The model is trained on 9 parts and tested on the remaining part, repeating the process 10 times and calculating the average results.
Regarding performance monitoring, real-time data are used to check the model’s predictions and compare them with actual data: if the model predicts that a product will sell 100 units in a week, real-time monitoring will compare these predictions with the actual weekly sales. Feedback from various stakeholders is also collected to identify any discrepancies or issues in the predictions.
Subsequently, the model undergoes periodic retraining, using new data that incorporate new demand trends and market changes, for example, quarterly.
Evaluation metrics are also monitored to train the model over time: if the model’s MAE (mean absolute error) starts to increase, it could be a sign that the model needs to be retrained with new data.

The implementation phase of the machine learning model is where the validated model is put into production to make real market demand predictions. This phase requires close collaboration between data scientists, software engineers, and various corporate stakeholders.
The integrated model is now able to automate various decision-making processes to plan production quantity, optimize inventory levels, and manage promotions; for example, if forecasts indicate an increase in demand for a specific product, the system can automatically increase production and plan targeted promotional campaigns.
As previously mentioned, the implemented model must be fed with real-time data to continuously update forecasts. Continuous monitoring offers several benefits: first, it allows the model to quickly adapt to changes in consumer behavior and the market; secondly, there are positive impacts in terms of accuracy: it maintains high forecast accuracy, reducing waste and overproduction. Finally, automating decision-making processes frees up resources for other strategic activities, promoting greater efficiency.
By integrating the model, companies can accurately predict future demand, reduce waste, optimize inventory management, and improve overall production sustainability.
According to a study published in Management Science, the use of machine learning algorithms has improved demand forecasting by up to 20% compared to traditional methods.
But how could they be further improved to obtain the most accurate market demand estimate possible?
Machine learning and artificial intelligence processes could be applied to currently unconsidered areas to achieve more accurate demand estimates.
Let’s analyze which innovative factors could be considered:

  1. Weather data
    The weather significantly impacts consumers’ clothing choices. Integrating weather data into demand forecasting models could help predict demand peaks for certain types of clothing. For example, during a particularly hot summer, the demand for light and breathable clothing could increase, while in winter, sales of heavy clothing could be higher in anticipation of cold snaps.
    A study by IBM found that weather forecasts can improve sales forecasts by 20-30% for fashion retailers, as weather directly influences consumers’ purchasing decisions.
    Integrating medium- to long-term weather forecasts could help improve production planning.
    Possible technology implementation:
    o Climate Forecasts: Use seasonal and annual climate data provided by organizations like NOAA (National Oceanic and Atmospheric Administration) and WMO (World Meteorological Organization).
    o Climate Models: Collaborate with meteorological research institutes to obtain long-term climate models predicting seasonal trends.
    o Machine Learning: Integrate these data into machine learning models to forecast seasonal and specific clothing demand based on weather conditions.
  2. International events and festivals
    International events, festivals, concerts, and other large gatherings can significantly influence fashion demand. For example, during Coachella or New York Fashion Week, the demand for certain types of clothing and accessories can increase significantly. An analysis by Eventbrite shows how large events and festivals can increase the demand for specific fashion items by up to 40% during peak periods.
    A possible implementation could be done using event monitoring tools to track the dates and locations of major events, identifying their target, scope, and main characteristics to design a targeted and specific production line.
    Integrating such data into predictive models could more accurately define the interested demand segments.
    Detailed event information would also allow fashion companies to create specific marketing projects, for example, using their social media channels and online advertisements to promote specific collections that respond to the themes and trends of upcoming events.
    Event data could also inspire product innovation: if the data show that attendees of specific events prefer a certain type of clothing, companies could develop new product lines that meet these needs.
  3. Forecasting tv series and movie trends
    Fashion trends often emerge from the popularity of TV series and movies. Anticipating trends based on upcoming series and film releases can help prepare for increased demand for clothing and accessories inspired by these media.
    An analysis by Parrot Analytics demonstrated how TV series and movies significantly influence fashion trends, with demand increases corresponding to the release of new seasons or successful movies.
    Obtaining a preview of series and movies estimated to have a significant impact on the audience (based on the marketing campaigns associated with them, distribution channels, influential actors and actresses, and the target audience) would allow predicting the scope of such impact and consequently the trends that could be generated.
    Additionally, monitoring the release dates of new TV series and movies along with related search trends would allow the application of trend prediction models to forecast the influence of such releases on fashion demand.
  4. Mobility Models and Geospatial Data
    Mobility data, such as traffic flows and people’s movement patterns, can provide insights into geographical areas with higher probabilities of selling specific items, identifying potential demand increases. Geospatial data could help optimize product distribution in physical stores.
    A Google report on COVID-19 mobility data showed how changes in mobility patterns directly influence purchasing habits and sales in physical stores. Mobility data were collected by Google through users’ mobile device location services, providing anonymous and aggregated information on people’s movements. These data demonstrated that movement restrictions and lockdown policies had a direct impact on consumers’ purchasing habits.
    Collecting mobility data could support fashion companies’ production planning based on changes recorded in movement patterns; for example, during periods of increased mobility to parks and recreational areas, there could be higher demand for sports and casual clothing.
    Additionally, it would allow optimized inventory management: imagining a decrease in visits to physical stores, companies could increase inventory for online sales while reducing that of physical stores.
    Companies could use such data to launch targeted marketing campaigns; for example, if data showed an increase in transit station traffic, campaigns could be oriented towards commuters, promoting comfortable clothing or travel accessories.
  5. Macroeconomic data analysis
    Macroeconomic data, such as GDP variations, unemployment rates, and economic policies like specific consumer subsidies, influence consumer spending behavior. Incorporating these data into predictive models can improve demand forecasting during periods of economic change.
    Companies could use economic indicator databases (such as those of the International Monetary Fund) to collect relevant macroeconomic data, including these variables in machine learning predictive processes to more precisely plan productions and seize potential economic changes that will influence demand.
    Integrating these innovative factors into demand forecasting models could enable fashion companies to achieve significantly more advanced precision, improving the operational efficiency of their production processes.

In conclusion, integrating new factors like those listed into machine learning mechanisms allows companies to quickly adapt to market dynamics, optimizing production based on real demand. This approach not only minimizes waste and associated costs but also promotes more efficient use of natural resources, contributing to a more sustainable future for the fashion industry.
Reducing the amount of unnecessary clothing produced also decreases the need for raw materials like cotton and synthetic fibers, whose production processes are highly resource-intensive and polluting. This leads to a reduction in water consumption, insecticides, and pesticides used in cotton cultivation, contributing to water resource conservation and biodiversity protection.
Moreover, more targeted production reduces the energy needed for clothing production and transportation, with a consequent decrease in greenhouse gas emissions. The fashion industry is responsible for about 10% of global carbon emissions, surpassing those produced by international flights and maritime transport combined. Reducing the amount of unnecessary clothing produced also decreases energy consumption and associated emissions, contributing to climate change mitigation.
Less unnecessary clothing produced also means less textile waste: millions of tons of clothing end up in landfills every year, contributing to soil and water pollution. Using machine learning to accurately predict demand, companies can produce only what will actually be sold.
Adopting machine learning and artificial intelligence models for demand forecasting thus enables the creation of a more responsible production cycle where resources are optimally used, and waste is minimized, contributing to a more sustainable circular economy.
Machine learning not only combats overproduction but transforms the fashion sector into a more sustainable, efficient, and responsible industry.
Adopting and continuously improving these technologies is a crucial step towards creating a future where fashion can thrive without compromising our planet.

Supporting Scientific Sources:

  • Management Science (2023). “Improving Demand Forecasting with Machine Learning”. Management Science.
  • McKinsey & Company (2024). “The State of Fashion 2024”. McKinsey & Company.
  • Harvard Business Review (2022). “How AI Is Streamlining Demand Forecasting”. Harvard Business Review.
  • MIT Sloan Management Review (2023). “AI-Driven Demand Forecasting”. MIT Sloan Management Review.
  • IBM (2021). “The Role of Weather in Driving Sales”. IBM Report.
  • Deloitte (2017). “The Consumer Review: Digital Predictions 2017”. Deloitte Report.
  • The Conference Board (2020). “Economic Indicators: Understanding Their Importance”. The Conference Board.
  • Google (2020). “COVID-19 Community Mobility Reports”. Google Mobility Report.
  • Accenture (2020). “Consumer Behavior Research 2020”. Accenture Report.
  • Eventbrite (2022). “How Festivals Drive Fashion Trends”. Eventbrite Analysis.
  • Parrot Analytics (2021). “The Influence of TV Shows on Fashion Trends”. Parrot Analytics Report.
  • Ellen MacArthur Foundation. (2017). “A New Textiles Economy: Redesigning Fashion’s Future”. Ellen MacArthur Foundation.
  • WWF (2022). “Why it is Critical to See the Hidden Water in Clothing”. World Wildlife Fund.