AI: The 4 biggest impacts to demand forecasts

A commonly overlooked feature of AI is its ability to support demand forecasting. Generative AI steals the spotlight, thanks largely to the creative output it allows for. But there’s an argument that predictive AI models are actually more exciting — and more impactful for fashion brands’ success.

Predicting the future is never simple, but with AI advances, it’s becoming increasingly accurate, actionable, and scalable.

What is the demand forecast status quo?

A very common way to forecast demand today is using historical time series data — like previous sales data, for example. From there, the idea is to create an equation that spits out likely future values. One of the simplest ways to do this is using linear regression or ARIMA (AutoRegressive Integrated Moving Average) models.

But you don’t want to settle for simple, right?

More advanced, machine learning (ML) approaches build on this basic premise: math equations with slots for data variables. Two common types are ensemble methods and neural networks. Like all ML approaches, these are special in part because they can learn and improve over time.

If you’re interested in going deeper, check out our demand forecasting models cheat sheet, designed for the planning community.

How does AI change how fashion planners predict demand?

I. New data inputs unlock accuracy.

Many status quo approaches to demand forecasting are limited in their complexity, by design. To simplify repeating tasks or make Excel useable, simple models with limited space for variables are common — usually just historical sales data.

AI approaches to demand forecasting use more advanced models that can accommodate multiple variables. In addition to historical sales data, they can take into account other important internal inputs such as marketing activities, lookalike sales, or web traffic.

Excitingly, they can also incorporate external variables. Think local weather or events, geofenced social media trends — anything likely to have an influence on location-specific foot traffic or brand-level demand.

The end result? More accurate and specific predictions.

II. Smarter models increase granularity.

The simplicity of status quo approaches also limits how specific planners can be in their forecasts. Aggregate predictions, for example on the style level, provide OK signal to inform assortment or allocation strategies.

When forecasts are built with advanced AI models, planners get access to increased granularity. Some AI approaches are capable of predicting demand at the style x size x color x location level.

Instead of making do with signal like “this season we expect to sell 5,000 jeans,” planning teams can instead act confidently knowing that “at Store A this week, we expect to sell 5 black medium jeans.”

III. Co-pilot capabilities simplify large-scale forecasts.

Managing the demand of any given style typically requires a substantial amount of manual effort. Individual planners have their scope artificially limited — maybe one style, or even one region — simply to ensure there’s enough time in the day to get through everything.

Modern AI is getting better all the time at providing in-the-moment support, sometimes referred to as co-piloting. A planning team member might receive a batch of forecasts with associated allocation recommendations, simplifying an otherwise time-consuming activity down to easy verifications and approvals. They might even have the AI co-pilot run standard allocations automatically.

In either case, the benefit to the planning team member is twofold. First, it allows them to increase their scope, overseeing multiple categories or departments simultaneously. In addition, it frees time to focus on activities that usually get buried under routine daily tasks: longer-term strategy or creative problem solving, for example.

IV. Tools that learn with you support always updated strategies.

Standard demand forecasts only change if someone changes them. In isolation, this might seem like an obvious thing to say! But the consequence is that planning team members need to apply new knowledge back into forecast models at the end of each key period — for example, tweaking WOS constraints mid-season to reflect an insight the team member picked up.

ML approaches by definition learn over time. As new data enters and results are captured, the model can automatically adjust its approach to maximize accuracy while still meeting guidelines provided by the planning team.

Planners benefit from a tool that adapts in real-time, more efficiently hitting targets as the season progresses. The AI might even surface a radical opportunity to shift strategy — an opportunity for the planning team member to stand out as a forward-thinking shaper of the brand’s success.