This article was originally published in our PLM Report 2023 – the definitive instalment in fashion’s longest-running dedicated PLM market analysis. To read other opinion pieces, exclusive editorials, and detailed profiles and interviews with key vendors, download the full PLM Report 2023 completely free of charge and ungated.

Key Takeaways:

  • AI has the potential to automate tasks in PLM’s extended ecosystem and achieve greater efficiency, quality improvements, sustainability, and cost savings.
  • AI can collect, interrogate, compare, and analyse datasets; in fashion, we can use AI & ML to predict the latest trending silhouettes, style types and feature details, material types, popular colours, retail channel pricing strategies, and actionable data used by PLM users.
  • Data science, data mining, machine learning, deep learning, and artificial intelligence are related, but it’s useful to understand how they differ from one another in order to use them effectively in the fashion industry.

AI & ML politics

Before I get into the subject matter, I know that there’s a lot of news and hype, both negative and positive, on the implications of AI. In a lot of ways, even though the technologies themselves are new – at least in their current form – this isn’t a new argument. There are always good and bad ways we use any software or platform. Just consider how we all use the internet and the vast number of cloud technologies in our everyday lives, for a huge range of different purposes.

I agree we need to be cautious when implementing new technologies that have the potential to be as disruptive as AI does, but provided we can trust the data sources we use to train AI models, the potential risks to the fashion industry will be much smaller than what is being discussed in the news today.

Remember, too, that we all already use AI most days in our lives; every time we use Google Lens, Maps, Photos, YouTube, Assistant, Gmail, and Cloud, we all use AI to find answers to daily questions, receive content recommendations, tweak images, and converse with an assistant. As another example: finding your routes to a new location or your most efficient way home after a hard day’s work, those recommended routes and maps are coming from the cloud, and a complex model is measuring traffic flows and conditions, warning of any significant hold-ups on the highway/motorways, providing multiple route choices and supporting information on where we can charge our new electric cars… or where can find restaurants and other facilities along your journeys.

Whether we actively realise it or not, we are happily using AI daily to streamline our personal lives, so why not embrace the same idea at work? And more specifically, why not find ways to bring the best elements of AI into PLM to improve our design, development, and manufacturing workflows by surfacing new ideas, smoothing out manual work, and much more?

This article takes that next step as an inevitability, since Microsoft, Google, and other technology and productivity giants are already carving out a new, AI-assisted future of work that fashion technology vendors will surely follow. So I want to explore how AI (and its stablemate, Machine Learning (ML)) might help fashion brands and their value-chain partners to automate tasks in PLM’s extended ecosystem and achieve greater efficiency, quality improvements, sustainability, and cost savings.

When we brought PLM to the market over 20 years ago, one of the leading value propositions was savings in time and non-value-added administration tasks. As a work- study engineer, I estimated that around 25% of a person’s working day is spent locating information and filling out electronic views with repetitive common data. And although our systems have become richer and better- integrated since, that balance has not really changed – and in fact things might even have regressed, since we deal with so much more data in any given day than we did two decades ago.

Just imagine the potential size of the benefits if we could reduce this non-valuable time by 50% or even more across the entire workforce, and then revert the time to value-added purposes. Creativity, profitability, job satisfaction and more could all see a significant boost from sensitive enterprise-level adoption of AI.

Defining the different components

Data science, data mining, machine learning, deep learning, and artificial intelligence are the principal terms you’ll hear being used in any conversation around automation, efficiency, and process transformation today. They are related, but it’s useful to understand how they differ from one another, too.

Data science

The broad scientific study that focuses on making sense of data. Data science is a general umbrella term that links all activities and technologies that help build new models and supporting systems. Consider recommendation systems used to provide personalised suggestions to customers based on their search history. If, say, one customer searches for a blue pair of trousers and the other looks for trouser belts, there’s a good chance both customers will also be interested in purchasing each other’s choices. This is a very broad discipline, and also one of fashion’s fastest- evolving roles, but in essence it’s concerned with the architecture of information.

Data mining

Data mining is commonly a part of the data science pipeline. Data mining focuses on techniques and tools used to search out patterns in the data, previously unknown and make data more usable for analysis. These patterns can be used to support trends or demands in fashion products, when they’re applied to datasets from eCommerce channels and from the web, or they can be used to transform efficiency when they’re applied to in- house information.

Machine learning

There are lots of different approaches to machine learning, but the most current and most popular is the deep learning strategy that underlies today’s large language models and transformers. As an example, a generative adversarial network (GAN) is an ML model in which two neural networks compete using deep learning methods to become more accurate in their outputs and predictions. When these are paired with a large, domain-specific dataset, and given the time and capacity to train and improve, the results are dramatically better than prior approaches to machine learning. The fashion dataset is one example of this kind of single-domain training data, and is an extensive, freely available database of fashion images commonly used for training and testing various machine learning systems.

Artificial intelligence

An umbrella term meaning different things to different people; it’s difficult to define because it encompasses a wide range of phenomena and concepts, from simple mathematical algorithms that recognise patterns in data sets to complex systems capable of intelligent behaviour such as reasoning, natural communication, problem-solving, and learning from humans. Today, following the launch and rapid large-scale adoption of ChatGPT (and its underlying GPT-3 and GPT-4 models, which are now the engines behind Bing, the next generation of Office and so on) and text-to-image creation tools like Midjourney, people use “AI” almost exclusively to refer to these generative tools, since they represent such a significant leap in both capability and potential impact over prior generations of AI.

Predictive modelling

Predictive modelling uses mathematical and computational methods to predict an event or outcome based on set inputs and a wide spectrum of other variables. Imagine if PLM vendors-built models that would automatically recognise the data inputs and outputs, including the approval or disapproval gates. These models forecast a result at some future state or time based on changes to the model inputs. Predictive modelling would enable the automation of the end-to-end workflow process, including all notifications and communications inside and outside PLM.

Data sources

We have abundant data used within the fashion sector; data comes from a universe of system solutions and is of varying utility depending on its age. As a strong example, consider how, today, brands, retailers, wholesalers, and manufacturing partners will use historical data related to their customers’ purchasing habits. As we have already stated, this data tends to lag days or weeks behind the purchase data. In recent times using social media feeds, TikTok, Instagram trends, Twitter hashtags, clothing styles of the most popular fashion influencers, and celebrity fashion events, we can now obtain the same data in a matter of hours, providing the business with rich and insightful near-term data. Today, we can use AI to generate new ideas, colourways, style features, and even suggested retail pricing based on the same data source in hours – and ChatGPT plugins could soon help to get those same styles eCommerce- ready and to streamline interactions with consumers. New, AI developed ‘virtual styles’ can then be designed and tested on e-commerce sites before manufacturing physical samples.

AI supports category optimisation and expansion

By ingesting and analysing ERP (sales) and PLM (product data) information, AI can be used to help develop product categories, and to understand and optimise the success of your current product categories. This means automated identification of new opportunities, instant insights on when to remove certain underperforming products and when to introduce newness, and rapid recommendations on when to increase volumes on the winners.

And this really is just the tip of the iceberg. AI’s role here is surfacing insights from existing information at the right time, but as we’ve already seen, tapping into the generative possibilities of image and text models (provided these are also properly trained on brand- specific datasets) can then allow brands to shortcut from identifying a new slot to filling it with suggested possibilities.

AI-enabled trend analysis modelling within PLM

By its very nature, the fashion industry is fast-paced and constantly changing with the waves of trends that roll in and out of the short seasons. As a result, fashion brands, retailers, wholesalers, manufacturers, and the rest of the value-chain (Tier1-6) partners, must stay agile and responsive to the customer’s ever-changing demands. At the same time, they need to control design, development, cost, sourcing, manufacturing, quality, sustainability impact, and of course, costs, and not always in that order!

AI can collect, interrogate, compare, and analyse datasets; in fashion, we can use AI & ML to predict the latest trending silhouettes, style types and feature details, material types, popular colours, retail channel pricing strategies, and actionable data used by PLM users. AI can gather data from within your own country or from across the world, so whether you sell in your own country or are an international brand, artificial intelligence can augment regional or country-specific trend data.

We can use AI linked to demographic characteristics, amongst many other models, filtering the data and then using deep learning models to support specific (age, interest, values, gender, product preferences, fits, likes, dislikes, materials, colour choices, quality, prints, plains, sustainability, quality, product types, price points, etc.) requirements.

At the same time, they can be automated to build mood boards, create supporting digital assets (silhouettes, colours, prints, plains, stripes, plaids, fabrics compositions, trims, components, accessories) and provide more detailed supporting metadata. At the same time, designers can provide human expertise and experience, giving specific parameters within the moodboard software, including the product category, seasonality, product use, environmental sustainability, country availability, and target price point.

This approach could ultimately save a huge amount of time and human effort. Unlike traditional trend forecasting methods that are time-consuming, taking days or weeks to compile, AI solutions can deliver the same results in hours and are far less prone to mistakes or data errors that often come with human inaccuracies. Although I do want to point out that generative AI is currently prone to “hallucinations” and inaccuracies, so it will remain essential for the foreseeable future for brands to add human validation to the end of any cycle that starts with an AI recommendation.

AI-automated tech packs

One of the biggest challenges with the product lifecycle today is the need to translate different workstreams, ideas, data, and processes into a single output: a technical specification or “Tech Pack”. This is an area that brands have long turned to PLM (and earlier PDM) to assist with, but today the creation of a tech pack remains only partially automated.

This is going to change soon, though. PLM vendors are likely to already be looking at incorporating AI models linked to PLM datasets/libraries/workflow processes and APIs. And using these deep-learning models, we will soon be able to automate the building of style templates that feed off the dynamically generated trend analysis data I’ve already written about.

Designing new products is a complex process that involves multiple stakeholders operating across the value chain. At the same time, it includes numerous product iterations (industry average of 3-6 samples) before the style can be approved. ML will help streamline this process by automating specific tasks, such as creating and modifying synthetic product sketches. These deep-learning algorithms can be trained to recognise trending style features and automatically generate new sketch variations based on market analysis results.

A Tech Pack example would be the use of ML to generate accurate and detailed technical specifications for each product option, providing all of the essential data elements, including the product category, product type, gender, size range, size classification, points and measure, how to measure guides, automated bill of materials, product testing, material composition, sustainability requirements, bill of labour operations, a list of potential sourcing partners, etc.

Essentially, we would arrive at what we call the 80/20 rule, with 80% of the primary data being supplied automatically by AI, adding more data as we move ahead with the workflow; each new process decision would drive automation (material composition, plus colour choice, would automate the material lab dip process), the final 20% being completed by the creators, developers, and manufacturing value chain partners.


Undoubtedly, we can say that data-powered decisions driven using AI & Machine Learning will give brands an edge in the competitive fashion world. Some may say that AI is taking jobs away from the creative and commercial specialists that keep fashion brands running today, but the reality is that AI & ML can give the creative teams more time to be creative rather than being buried in administrative tasks that simply don’t add value to their day jobs, and it can allow commercial and technical teams to focus on the elements of their roles that hinge on human capability rather than data analysis and manipulation.

When I cast my mind forward even a short time, it’s clear to me – as someone who’s seen the technology journey the fashion industry has already undergone – that AI is going to be integrated into the core and extended solutions that make up the fashion technology ecosystem. But rather than replacing human skills, AI – used carefully – will instead help fashion to prepare for the future and will bring positive change to both core PLM and the broad spectrum of other solutions that integrate to it.