Expert insight: 5 essential data sources for fashion forecasting

In this Q&A with WGSN VP of Operations & Strategy, Jocelyn Corner, she explains the five data sets that WGSN uses for trend forecasting.

Q: Why is data important in fashion?

Fashion sometimes has a reputation for not being particularly data-driven but in truth, fashion and data go hand in hand. I remember when I started out my career as a junior retail consultant, one of things I was tasked with were manual store checks, when I would visit retail locations and attempt to discreetly count and photograph items on the racks. I would then enter this data into an Excel spreadsheet and analyse competitors’ range and price architecture.

Obviously the data and the methods available have improved exponentially since then, but SKU counts is just one example. If you work in the fashion industry, you are constantly looking at what’s showing on the catwalk, what influencers are posting on social media, what’s showing up in store or being worn on the street, what’s selling and what consumers are looking for (whether because they’re telling you or they’re typing it into a search bar).

All of these images, words and numbers are ultimately data points and over the past few years at WGSN we have been working hard on further quantifying this data to create the metrics to measure what you are seeing.

Q: How do you categorise this data at WGSN?

Internally, we fondly refer to these core data sets as the five S’s: Shows, Social, Shelf, Search and Sentiment. It’s the way we classify our data across all industries, not just fashion. Worth noting that the order is important as it broadly follows the lifecycle of a trend: from fashion leaders showing it on the catwalk (shows), to early adopters posting it on social media, to when it hits mass in stores, consumers search for it, purchase and review.

We analyse all of these data sets individually when it comes to fashion and they also input into our AI-driven models where the different lead times play an important part in forecasting trends up to 24 months out.

Q: What sort of proprietary data do you use?

With over 20 years in the industry, WGSN has access to an unrivalled wealth of proprietary data including almost two decades of catwalk shows, totalling over three million images and spanning every look from every major fashion show. From social media we analyse 100,000 posts a month across our expert-curated Influencer Map, which is segmented into cohorts according to how quickly they adopt trends. Our shelf data records product details from over 400 million SKUs on a daily basis, analysing key retail metrics such as new-ins, markdowns and out-of-stocks to give an indication of performance.

Alongside our sister brands at Ascential we are uniquely positioned to access extensive consumer search data, which enables us to track the magnitude of searches and monitor demand. We survey 17,000 consumers on a monthly basis to ask them directly about buying intent and brand sentiment. 

Q: How do you ensure quality as well as quantity?  

Quality is absolutely crucial with data. As the (polite version) of the expression goes, ‘garbage in, garbage out’! At WGSN we are building on our leading global trend forecasting expertise and we bring this human element into the loop wherever we’re using data, whether it’s working with our data science teams to ensure the code they are writing is picking up the right terminology, or when it comes to analysing the output of our models.

To give you an example, in order to compile our catwalk data, we have a team of fashion experts who have carefully identified and tagged the key attributes of every single outfit in our library of catwalk images, capturing the style, silhouette, material and design details. For instance, a green silk A-line dress with a puff sleeve and ruffle detail. Human expertise ensures near-100% accuracy in the data, which translates into an extremely powerful training data set for our image recognition and natural language processing technologies.

Q: How does the data help you forecast trends?

While each of these sources in themselves are incredibly rich, analysis shows that you cannot rely on one single source of data to forecast trends, and that combined they are significantly more powerful. Running machine intelligence on top of these billions of data points enables us to identify patterns – what’s growing, what’s declining, what’s flat – and to understand where a trend is in its lifecycle to project the next 12 or 24 months out.

These AI-driven forecasts underpin our recently launched TrendCurve+ service, which helps our clients understand when and how to trade a trend to maximise sales and minimise overstock. By combining the five S’s, we can now average over 90% forecasting accuracy.

Q: Which of the five S’s are most important?

It depends. Our data-science models continually analyse the different inputs and we adjust the future forecasts according to the correlation and lag of these inputs with the market, so some sources will be more important for some trends than others, depending on whether they’re more influenced by the catwalk or if they start out life on social media.

This is again where WGSN’s expertise comes in – it is part of our alchemy as trend forecasters to be able to flex the five S’s according to their relevance to a product or category and we see the relevance of the five S’s evolve over time. In fact, we have tracked the increasing importance of social as a predictor of a trend post-pandemic, whereas pre-Covid-19 it was a relatively poor-indicator of what would be successful at retail.

Q: Is AI going to ever replace traditional trend forecasting?

I don’t believe so! Data and AI are incredible tools but they’re not enough by themselves; we always need to augment them with human expertise. WGSN’s trend experts’ qualitative knowledge is there to contextualise and nuance the gaps in the quantitative data.

To take an example like the current cost-of-living crisis, yes, historic precedent does exist, but we would need data to go back to 2008 to replicate a similar scenario, when social wasn’t really a thing. This goes to show that even when the data is cyclical, it isn’t always comparable and we need our experts to account for the current context.

Q: What does having this data mean for the industry?

In the past, decision making has been much more based on the knowledge and experience of the industry and any reference to data has been historical. The tools available now are infinitely more powerful, providing a quantitative forward-looking view of consumer demand for the first time. 

However, data will always be just that: a way to quantify the signals out there. It is a tool to supplement your expertise and enhance the speed and accuracy of your decision making. Hopefully it also means that the next generation of retail analysts won’t have to sneak around stores counting SKUs.

For more information on TrendCurve+ and our data-driven forecasting, please visit our website or request a demo here.