Back to Blog
WGSN

Expert insight: The role of data in trend forecasting

Why does data matter in trend forecasting? We find out in this Q&A with WGSN Lead Data Scientist Jack Shipway, as he explains what trend forecasting is in the context of fashion.
WGSN Insider WGSN Insider

Why does data matter in trend forecasting? We find out in this Q&A with WGSN Lead Data Scientist Jack Shipway, as he explains what trend forecasting is in the context of fashion.

Jack Shipway

Q: Explain your role at WGSN.

JS: I’ve never perfected the art of telling someone what I do. In fact, at a barbecue recently, I received glances of disbelief when I said ‘fashion modelling’, though they did start to listen when I mentioned trend forecasting. My role at WGSN is to support qualitative forecasts with data-derived evidence, ensuring our clients are making decisions based on facts and our expert opinions.

Q: What’s a forecast?

JS: Forecasts are our expectations of events yet to unfold, and our attempts at describing a likelihood of them happening. In a world of uncertainty, we tend not to talk in absolutes; rather, we assign probabilities.

We as humans have been forecasting for millennia, but nowadays we concern ourselves with questions such as ‘Will it rain on Monday?’, ‘Will the FTSE hit a record high this month?’, ‘Will more of us be wearing white trainers at the next barbecue?’’ Predicting these events accurately can have a huge outcome.

Q: How do you define a trend?

JS: The subject of these forecasts are trends. To clarify, not everything is trend-ing (something exhibiting a sudden change in popularity). If we zone in on fashion, anything we wear could be a trend: jeans, T-shirts, sandals. Most items can also be split into styles: boyfriend jeans, skinny or relaxed.

Conversely, items can combine to form larger groups called categories: trousers, tops, footwear. This hierarchy defines the WGSN taxonomy and forms the blueprint for our forecasts. Each style and category can further be described in terms of colour, fabric, length, prints, patterns, embellishments – these are attributes and can, in theory, apply to any item in the taxonomy.

As with the stock market, styles, categories and attributes are in a constant state of flux and the ways in which they interact are fascinatingly complex. Trends ebb and flow in and out of fashion, while new products battle for market share. These are key elements of what we track and forecast at WGSN.

Illustration of data process
WGSN Original Image

Q: What does trend forecasting look like at WGSN?

JS: Our job in the Data Science team is to unearth the signals behind the noise and contextualise data with expert industry knowledge to help fashion buyers and merchandisers quantify future trends. What is my optimal assortment mix, how deeply should I invest, and when and what are my competitors doing?

Q: Why does this matter?

JS: Retailers make decisions months (if not years) before products hit the shelves; mistakes can be unprofitable and wasteful, especially with the spotlight on sustainability. It’s as important to know when a trend is going to peak so as not to overproduce stock, as it is to know if it will rise in the first place.

The problem is, without data, an educated guess is all we have. One in three of us are wearing white trainers at the barbecue – is that high? It is summer and it isn’t raining, but what happened to sandals? They were popular last summer, weren’t they? Sustainability is a key consumer focus; will more of us be wearing them at the next barbecue as a result? We need something more rigorous to back up our intuition, and that’s where WGSN data comes in.

Q: What data do we have?

JS: Quite a lot! Five years worth of proprietary shelf data. We have daily tracking of 400m SKUs across 10,000+ brands and 300 retailers, and this is underpinned by our expert-derived taxonomy, which allows us to define key retail performance metrics (new-ins, out-of-stocks, markdowns) for any trend. Add to this social media tracking, millions of tagged catwalk images, consumer sentiment and search data, and it’s a pretty comprehensive data selection.

Short-term spikes, key calendar events, seasonal shifts and long-term trajectories form the basis of time-series forecasts, where state-of-the-art machine-learning algorithms extract statistical patterns in the data and project them into the future. Our analyst-in-the-loop approach allows for minor tweaking to fine-tune our forecasts, and it is this unique approach of augmenting proprietary data with industry expertise that makes WGSN the world’s leading trend authority.

So, as you can see, data plays an integral part in trend forecasting and as our tools become ever more sophisticated, this role will only grow. As for myself, I'm getting better at explaining what I do... but I'm still not trendy.

Share:
Facebook
LinkedIn