The Challenge
Omlet designs, manufactures, and distributes products directly to consumers across multiple markets. As their business scaled, the limitations of their existing forecasting approach became increasingly apparent. Their legacy system relied on traditional statistical methods, primarily linear regression and moving averages, which struggled to capture the complex, non-linear relationships inherent in modern e-commerce demand patterns.
This resulted in inventory imbalances: overstocking of slow-moving SKUs whilst understocking high-demand products. Fulfilment inefficiencies followed, with suboptimal warehouse allocation and shipping route planning. Manufacturing schedules became disconnected from actual demand signals, leading to missed revenue opportunities through stockouts during peak periods and promotional campaigns.
Omlet required a forecasting solution capable of learning from historical patterns, adapting to seasonal fluctuations, and incorporating multiple external variables to drive intelligent fulfilment decisions.
Our Approach
We designed and implemented an end-to-end machine learning forecasting pipeline, centred on the Temporal Fusion Transformer architecture, a deep learning model purpose-built for multi-horizon time series prediction.
Our data engineering work consolidated disparate data sources into a unified forecasting dataset, incorporating historical sales data across product categories and customer segments, promotional calendars and marketing campaign schedules, seasonal and cyclical patterns, and external variables including regional demand indicators.
The TFT model was selected for its ability to handle multiple time horizons, generating accurate predictions from one week to twelve months ahead. It provides interpretable outputs that quantify the influence of each input variable on predictions, captures complex temporal dependencies that traditional methods miss, and incorporates known future inputs such as planned promotions, holidays, and events.
The forecasting outputs were designed to directly inform inventory replenishment through automated reorder triggers, warehouse allocation through proactive stock positioning, manufacturing scheduling through demand-driven production planning, and logistics optimisation through improved carrier selection and delivery route planning.
The Outcome
Delivered in 3.5 months from project inception to production deployment, the solution achieved a 46% reduction in forecasting error compared to the baseline approach. This has a significant knock-on effect to inventory holding costs through optimised stock levels, while protecting against stockout incidents during peak trading periods, all allowing for improved production planning and better alignment between manufacturing output and demand.
Traditional forecasting methods were designed for a slower, more predictable retail environment. Modern e-commerce demands forecasting systems that can learn, adapt, and scale, capturing the nuanced patterns that drive consumer behaviour. Machine learning models like the Temporal Fusion Transformer represent a step-change in forecasting capability, enabling organisations to move beyond historical averages and embrace truly predictive fulfilment operations.