The agency that used to set the weather
For decades, trend forecasting belonged to a small priesthood of agencies. WGSN, Trendalytics and Edited told the industry what would be in, and brands paid handsomely to hear it first. A WGSN subscription has long run around $25,000 per year, a price that put professional foresight behind a paywall and turned forecasting into a competitive moat reserved for large houses.
That moat has been partly dissolved by AI. The same predictive capability that once required a six-figure budget is now available through tools priced from $10 to $99 a month. The result is not that forecasting got cheaper alone; it is that the entire buying function had to be redefined around data it can no longer ignore.
How 94% accuracy is actually built
WGSN's TrendCurve AI claims 94% accuracy up to one year ahead. That number is not magic. It rests on a layered data engine: more than 2,800 third-party global data sources, a catwalk image library of over 5 million tagged photographs spanning 20-plus years, and shelf data tracking roughly 300 million SKUs across 200 retailers and 10,000 brands every day.
Crucially, the algorithm does not make the final call. WGSN uses an analyst-in-the-loop method in which more than 250 industry experts review the model's candidate predictions and judge whether they fit a specific market. The 94% figure, the company says, comes from continuously calibrating the model so it is neither over-fit to short-term noise nor lagging the market.
From instinct to instrument
The buyer's gut has not disappeared, but it now shares the desk with a score. Heuritech, which analyzes millions of social images monthly, reports accuracy above 91% for color and style predictions, with particular strength in European luxury. Style3D and similar systems push physics-accurate 3D simulation so a forecast can become a render before a sample is cut.
The shift is quantitative: instead of "velvet is coming back," a team sees a velocity curve, a regional split, and a confidence interval. That turns forecasting from a presentation into an instrument a merchandiser can query mid-season and act on.
The democratization problem
Cheap AI forecasting created a new risk. When every mid-market brand can read the same signals, they tend to forecast the same things, and the shelf fills with near-identical "trend" pieces. Sameness is the hidden tax of democratized data: the edge moves from who-knows to who-acts-and-edits.
The brands that win are not the ones with the most accurate model. They are the ones that use the model to de-risk a point of view they already hold. Forecasting tells you a color is rising; taste tells you what to do with it so it still looks like your brand.
What this means for the shopper
For shoppers, the algorithmic forecaster is mostly a quiet benefit. Arrivals are more likely to match what you actually want, restocks happen on winners instead of duds, and the lag between a look appearing and becoming available shrinks. The downside is homogenization: if every brand reads one forecast, every feed starts to look alike.
This is where curation earns its keep. A store with an opinion will use data to stock smarter, not to copy everyone else. At PHEFEE we read the signals but filter them through a rose-gold point of view, so the dresses in our dresses edit and the layering pieces in our tops edit feel chosen rather than computed.
There is a final caution worth stating plainly. A forecast is only as honest as the data behind it and the questions asked of it. Models trained on recent feeds can mistake a viral spike for a durable shift, and they can underweight markets where data is thin. The analyst-in-the-loop method exists precisely to catch those errors, which is why fully automated forecasting still carries blind spots. The shopper rarely sees this machinery, but she feels its results: when brands over-trust a single signal, shelves fill with lookalikes and the genuinely new gets crowded out. Good forecasting, like good editing, is as much about what you refuse to chase as what you stock.
Frequently asked questions
What is WGSN and how accurate is its forecasting?
WGSN is a leading fashion trend-forecasting platform whose TrendCurve AI model claims 94% accuracy up to one year ahead, built on 2,800-plus data sources, 5 million tagged runway images and daily shelf data across hundreds of retailers, with human experts validating the output.
How much does professional trend forecasting cost?
Enterprise platforms like WGSN have run around $25,000 per year, while AI-powered alternatives now start at roughly $10 to $99 per month, which is why forecasting has moved from a large-house privilege to a broadly available input.
What does analyst-in-the-loop mean?
It is a hybrid method where an AI model generates candidate forecasts and human industry experts review and validate them before they become recommendations, combining data scale with market judgment.
Can small brands compete with big forecasting budgets?
Yes, because affordable AI tools narrowed the data gap. The remaining edge is taste and speed of execution: using forecasts to support a distinct point of view rather than to copy the same trends as everyone else.
