Why most inventory forecasting tools struggle in fashion, and what to look for instead
There’s a pattern that comes up repeatedly in conversations with fashion retailers who use general inventory planning tools: the forecast is running, the system is switched on, and yet the merchandising team is still spending a significant part of their week working around it.
Manual overrides. Offline spreadsheets. Gut-feel adjustments that never make it back into the system. A forecast that the team has quietly stopped trusting.
If that sounds familiar, the problem is rarely the team, and is more likely to be the architecture underneath.
The core issue: forecasting at the wrong level
Most inventory tools forecast at the individual size level, treating every SKU as its own independent demand signal. The logic seems sound: if you want store-level precision, start at the most granular unit available. In categories with high rotation and stable demand, this works reasonably well. There’s enough data per size to build a reliable picture.
In fashion, however, size-level rotation is typically too low to generate a statistically robust forecast, especially for short life-cycle products. A size 38 dress that sells three units in its first week doesn’t give the system enough signal to work with. And when you multiply that problem across hundreds of SKUs in a seasonal collection, the forecast becomes unreliable almost by design. Therefore, the system is trying to read a signal that is too faint to hear clearly.
The result is a forecast that frequently diverges from what merchandisers know instinctively. And that’s not because the merchandisers are wrong, but because the underlying model is working with insufficient data at the wrong level of aggregation.
A more robust approach is to forecast at product level first, then apply a size curve on top. This produces a much more reliable base forecast, because you’re working with meaningful sales volumes before distributing across sizes. The size curve then handles the variation in a structured, fashion-specific way rather than asking the system to infer it from sparse data.
From there, the execution and decision-making layer operates at SKU-store-day granularity. Thus, every allocation, replenishment, and transfer decision is informed by what is actually happening at the level of each size, in each store/point of sale, each day. This combination of product-level forecast robustness feeding into SKU-store-day execution precision is what allows a fashion-native tool to act on early signals quickly and accurately.
The forecasting model itself wasn’t trained for fashion
The underlying forecasting model itself also matters, not just the level at which it operates. Machine learning forecasting is not a generic capability that transfers unchanged from one industry to another. A model trained and tuned for grocery demand (i.e. long lifecycles, stable velocity, consistent replenishment cycles) will behave very differently from one built specifically for fashion. The same applies to pharmaceutical or industrial inventory models, which operate under entirely different demand and lifecycle assumptions.
Fashion has structural characteristics that require a model built around them from the ground up: high and low-rotation products and businesses, short and variable product lifecycles, sparse early-season data, strong size curve dynamics, newness-driven demand patterns, and the need to act on signals before they are statistically mature. An ML model that doesn’t account for these specifics will produce forecasts that are technically sophisticated but practically unreliable for the decisions a fashion merchandiser needs to make.
The forecasting model underneath the tool is rarely visible in a product demo. But it is the single biggest determinant of whether the forecast earns team trust or becomes another number the team works around.
The calibration lag problem
Related to this is the question of how quickly a tool can react to early-season signals.
A forecasting model that relies on accumulated sales data to calibrate needs time (typically two to three weeks) before it can produce reliable outputs for a new product. In stable demand categories, this lag is manageable. because these products have long lives and the selling window is forgiving.
But in fashion, the peak selling window for a seasonal product is often measured in weeks rather than months. A bestseller that isn’t identified and acted on in the first few days of trading is frequently a bestseller that has already been missed. The size curve breaks, stock concentrates in the wrong locations, and the opportunity to replenish or redistribute has passed.
A fashion-native forecast acts on early sell-through signals from day one, as it doesn’t need to wait for the data to accumulate. That difference in speed is far from a minor operational detail, because it directly affects sell-through rates, margin, and end-of-season markdown exposure.
What happens when the in-season toolkit has gaps
Beyond forecasting architecture, it’s worth examining what the in-season toolkit actually covers and what it doesn’t.
Some tools position themselves as in-season solutions but don’t include first allocation as part of the package. This creates an immediate gap at the start of every season: the moment when getting stock into the right stores quickly is most critical, the tool isn’t covering it.
Similarly, store transfers are a standard feature in most platforms, but there’s a meaningful difference between a tool that proposes a movement and a tool that calculates the expected result of that movement. If the system tells you to transfer 20 units from Store A to Store B but doesn’t show you the projected sales impact, your team has to do that calculation outside the system. That means the tool is simply a suggestion engine that creates additional work.
Other common gaps in tools not built specifically for fashion include: linked lines (identical products with different reference codes so that decisions on one inform decisions on another), and reintroduction logic (the ability to bring back products from previous seasons with appropriate adjustments for changed demand patterns). These aren’t edge cases in fashion retail, they’re regular operational needs.
The breadth vs. depth problem
Some tools position themselves as end-to-end solutions, covering everything from pre-season buying and budgeting through to in-season replenishment and transfers. On paper, that looks like a compelling proposition: one platform, one data model, full visibility across the season.
But breadth and depth are different things, and a tool that spans pre-season planning and in-season execution is making a very different architectural bet than one built exclusively for how fashion moves in-season. A good pre-season plan still matters enormously, as it sets the boundaries within which in-season agility can operate. How much inventory is bought, how budgets are structured, and how assortments are shaped upstream all condition what’s possible once the season starts.
Pre-season planning is problematic when the logic that underpins a pre-season module is generic (e.g. built around historical averages, stable demand assumptions, and long planning cycles) and bleeds into the in-season layer. When that happens, the tool behaves accordingly: slow to react, resistant to early signals, and poorly suited to the pace of a seasonal collection.
The more useful question to ask of any tool is not how much it covers, but how deeply it handles the moment when your season is still alive and your decisions still matter. Pre-season sets the stage, while in-season is where margin is truly won or lost by way of faster reactions, better stock distribution, and a forecast that earns trust rather than requires constant correction.
That’s why the most meaningful gains tend to come from improving in-season execution. The mechanics of what that looks like and practice are covered in depth in our Agile Merchandising white paper.
Download the Agile Merchandising for Fashion white paper.
What “fashion-native” looks like in practice
The retailers who get the most from their inventory tools tend to share a few common characteristics in how those tools work:
- The forecast is trusted enough to act on: Merchandisers are making decisions based on what the system recommends, and not spending their time correcting it. When a team is running daily manual distribution reviews and maintaining offline calculations, that’s a signal that the tool isn’t carrying its weight.
- Fashion-specific rules are native and not add-ons or workarounds: Visual minimums, size curve protection, newness handling, linked lines, etc. should be part of how the system thinks. They should not be manual inputs that someone has to remember to apply.
- In-season results are measurable: ROI from allocation and replenishment decisions should be visible within a season and not dependent on complex external calculations or left as an open question at year end.
- There’s an active partner, not just a platform: Fashion retail moves quickly. A tool that requires heavy internal expertise to maintain or whose value collapses if the internal champion leaves is too fragile a dependency. Active support throughout the season, from a team that understands fashion operations, is part of what makes the technology work in practice.
A useful benchmark question
The retailers who navigate this well aren’t necessarily the biggest or the most resourced. They’re the ones who combine a disciplined pre-season approach with the in-season execution capability to adapt when reality diverges from the plan which in fashion, it always does.
Guess is a useful example: after shifting to a demand-driven, fashion-native approach, the brand saw a 5 pp full-price sell-through increase, alongside a 7.5% reduction in store coverage. All without having to completely rebuild the business, but rather just through a more agile approach to in-season execution.
If you want to understand the full operating model behind this kind of transformation, including what the technology, data, and team structures look like in practice, our Agile Merchandising white paper covers it in depth.
Download the Agile Merchandising for Fashion white paper.
And if you’d like a faster, more direct read on where your current setup stands, contact us to book a 20-minute Merchandising Assessment with our team.
