AI-Powered Supply Chain Forecasting for Rooms To Go

Project snapshot

Throughout our partnership with Rooms To Go, we have consistently been impressed by the team’s commitment to practical, value-driven technology adoption — enhancing what works, rather than replacing it for the sake of novelty.

Together, we set out to explore whether machine learning could match or exceed the accuracy of an existing rules-based forecasting system, and then apply that capability to one of the most costly challenges in retail logistics: getting the right inventory to the right distribution center before it ships.

Business challenge

Rooms To Go’s supply chain has grown significantly more complex over the years. Import lead times of three to six months, multi-DC inventory management, and an expanding product range had pushed the existing rules-based forecasting system to its limits — it could no longer process the full range of variables influencing demand and inventory placement at scale.

The practical impact was felt most acutely by the buying team, who were spending days at a time manually reviewing purchase orders in Excel to determine which containers to release and to which warehouse. Without a data-driven prioritization mechanism, the risk of routing inventory to the wrong distribution center — and incurring costly inter-DC transfers to correct the imbalance — was a constant operational reality.

At the same time, Rooms To Go was committed to AI adoption that was safe, measurable, and genuinely additive to existing systems. The challenge was not simply introducing AI but doing so in a way that could be validated, trusted by business users, and incorporated into live supply chain workflows without disrupting the UniVerse-based infrastructure that had underpinned operations for three decades.

Solution

Intellias proposed a phased approach designed to prove value before committing to scale. The first step was a tightly scoped proof of concept: a group of well-understood SKUs, two years of sales history extracted from the UniVerse ERP via Rooms To Go’s custom UVSync data synchronization layer, and a set of off-the-shelf machine learning models — LightGBM, XGBoost, CatBoost, and Meta’s Prophet — trained and evaluated with no real-time integration required. A simple web application visualized forecast curves against historical actuals, giving business stakeholders a clear, interpretable view of model outputs.

The POC was completed in three and a half weeks by a single data scientist. When actual November sales — including the Black Friday peak — were compared against the AI-generated forecasts in December, the accuracy exceeded expectations and validated the business case for an MVP.

The MVP focused on long-haul logistics optimization: the highest-ROI entry point in Rooms To Go’s supply chain. The application runs daily, ingesting updated inventory positions and demand forecasts to produce a prioritized, explainable recommendation for which purchase orders to release and to which destination warehouse — before containers ship from origin. The interface was purpose-built for buyers, allowing them to drill into individual SKU-level data and understand the reasoning behind every recommendation, building the business trust required for genuine adoption. The architecture — Python inference pipeline, PostgreSQL forecast store, and bespoke web application — was kept deliberately lean, running within Rooms To Go’s own Azure tenant with no new data residency risks.

Business outcomes

The proof of concept demonstrated that AI-driven demand forecasting could match or exceed the accuracy of the existing rules-based system on a well-understood SKU set, providing the confidence needed to move into production. The MVP is now accessed daily by the buying team, with AI-generated recommendations replacing a process that previously required days of manual purchase order analysis in Excel.

The solution has standardized the container release process and created a clear, auditable basis for routing decisions — reducing the reliance on individual buyer judgment for time-sensitive logistics calls. As inbound shipments routed by the model begin arriving over the coming months, Rooms To Go expects to see measurable reductions in inter-DC transfers, stockouts, and inventory imbalance across its distribution network.

Looking ahead, the forecasting model is designed as a shared foundation: a single source of demand intelligence that will extend across replenishment ordering, promotion planning, and inventory flow applications — giving every buyer, across every part of the supply chain, a consistent and trusted view of forward demand.