Seedlabs

BaseSeason: Long-Horizon SKU Demand Forecasts with a Base-Trend/Season Split for Retail Buyers

A tool for retail buyers that predicts product demand by separating long-term trends from holiday peaks.

A demand-planning tool that gives retail buyers 3–12 month-ahead SKU forecasts split into base-demand trend and seasonal/holiday components, so purchase-order sizing comes with visible reasoning instead of black-box numbers.

Computer ScienceTime Series Analysis and Forecasting
Hybrid Graph-Based Memory and SchedulingRetail & consumer goods — long-horizon SKU demand planning and seasonal buying

Concept

A demand-planning tool for retail merchandisers. It ingests multi-year daily sales history per SKU or SKU-region, decomposes it into a base-demand trend and recurring seasonal/holiday components, and produces 3–12 month-ahead forecasts with both components surfaced side by side — "base demand is up 6% quarter over quarter; the December peak is a separate, visible component." The Transformer layer handles local deviations such as promotion spikes. Buyers currently receive black-box numbers they hesitate to sign purchase orders against; the explicit split gives them auditable reasoning for order sizing, and forecasts export directly into existing ordering/ERP workflows. Long lookbacks across thousands of SKUs are affordable because compute scales linearly with history length.

Why now

FEDformer [0] shows that combining seasonal–trend decomposition with a frequency-enhanced Transformer improves long-horizon accuracy by 14.8% (multivariate) and 22.6% (univariate) over state-of-the-art baselines with linear complexity in sequence length, making multi-year daily lookbacks per SKU affordable at retail's SKU counts. AdaWaveNet [2] independently shows that explicit multi-scale frequency bases improve forecasting of non-stationary series, corroborating the frequency-decomposition wedge, and the survey [1] flags computational cost as a key adoption barrier for transformer forecasters — the exact barrier the linear-complexity result removes.

Written by GLM · 8 Oct 2026

AI assessment

Backed by 3 papers82

A high-utility application of FEDformer that solves the 'black-box' trust problem in retail procurement, though it is a direct sectoral port of the existing TrendSplit idea.

Evidence strength
5/5
The idea is tightly grounded in FEDformer's specific results on linear complexity and seasonal-trend decomposition, corroborated by AdaWaveNet's findings on non-stationary series.
Market pull
4/5
Retail buyers have a high-urgency need for auditable forecasts to justify multi-million dollar purchase orders, making the 'reasoning' aspect a strong value prop.
Novelty & moat
2/5
The mechanism (trend/seasonal split for planning) and value proposition are nearly identical to TrendSplit, simply swapping electricity utilities for retail SKUs.
Feasibility
4/5
The core architecture (FEDformer) is published and the data (SKU sales history) is standard, allowing for a rapid MVP.
Wedge clarity
5/5
The wedge is sharp: replacing black-box numbers with a visible 'base + seasonal' split to enable PO signing.
Simplicity / focus
5/5
The product is focused on a single, clear output (decomposed forecast) rather than an over-scoped supply chain platform.

Scored by Gemma 4 · 8 Oct 2026. Rated against a fixed rubric (evidence, market, novelty, feasibility, wedge, simplicity), relative to the closest ideas already in the catalog. A prior estimate to compare ideas before real-world signal arrives.

Persona discussion

AI personas trained on real people's expertise debate this idea as it evolves.

View the discussion →

Act on this idea

Ideas only matter if someone runs with them. Your message goes straight to the founder's inbox — nothing is stored on our servers.

Who benefits

  • Blue Yondercompany

    This supply-chain planning vendor could embed the decomposition-based engine as its demand-forecasting module, differentiating on interpretable trend/season splits for merchandise buyers.

  • o9 Solutionscompany

    Its demand-planning platform serves retail and CPG clients; a more accurate long-horizon forecaster with visible components strengthens the planning story for buyers and planners.

  • Walmartcompany

    Plans demand across tens of thousands of SKUs and stores on multi-month horizons; linear-complexity forecasting makes per-SKU long-lookback models affordable at that scale.

  • Inditexcompany

    Zara's buying teams commit to seasonal orders months ahead; forecasts split into base trend versus holiday peak give them defensible order sizing for fast-fashion cycles.

Research it builds on

  1. Fedformer: Frequency Enhanced Decomposed Transformer for Long-term Series Forecasting
    Tian Zhou, Ziqing Ma, Xue Wang et al. · 2026 · 213 citations
    All ideas from this paper →
  2. A survey of transformer networks for time series forecasting
    Jingyuan Zhao, Fulin Chu, Lili Xie et al. · 2025 · 45 citations
    All ideas from this paper →
  3. AdaWaveNet: Adaptive wavelet network for non-stationary time series forecasting via end-to-end learning
    Tiancheng Jin, Ji Yu, Ji Liu et al. · 2026 · 2 citations
    All ideas from this paper →
Leave feedback
feasibility