IN Brief:
- Dollar General will deploy RELEX forecasting, replenishment, and allocation across more than 21,000 stores and 34 distribution centres.
- Approximately 18,000 SKUs will be planned using common demand, ordering, lead-time, supplier, and fulfilment data.
- The rollout moves AI-based planning into day-to-day inventory control across one of North America's largest store networks.
Dollar General will deploy RELEX forecasting, replenishment, and allocation technology across more than 21,000 stores and 34 distribution centres, moving planning for approximately 18,000 SKUs into a common environment. The system will combine demand forecasting with store replenishment, ordering schedules, supplier lead times, allocation, and different fulfilment methods across the retailer’s North American network.
The scale makes coordination the central problem. A forecasting error repeated across thousands of locations can translate into substantial excess inventory or widespread availability problems, while individual stores can have markedly different demand patterns even when they carry similar assortments. Distribution centres must then reconcile those store-level requirements with inbound supply, available stock, transport schedules, and supplier performance.
RELEX will bring those decisions into the same planning process rather than treating forecasting, replenishment, and allocation as separate exercises. Sales patterns and other demand factors will feed the forecast, which can then influence ordering quantities and timing at store and distribution-centre level.
Allocation becomes particularly important when available inventory cannot immediately satisfy every location. The planning system then has to determine where constrained stock should be sent, taking account of demand and availability rather than simply distributing the same quantity across the network.
Dollar General has also identified supplier coordination and lead times as part of the project. A replenishment calculation is only as useful as the assumptions behind it: if a supplier regularly takes longer to deliver than the planning system expects, the resulting order may still arrive too late irrespective of forecast quality.
The same applies to fulfilment. Stores can be served through different processes and schedules, while distribution-centre capacity, transport frequency, and product characteristics influence how quickly inventory can move through the network. Putting those variables into one planning environment gives teams a better chance of working from consistent assumptions.
RELEX describes the platform as AI-driven, although the operational value will depend less on that label than on the quality of the decisions it can automate. A network containing more than 21,000 stores and 18,000 SKUs creates far too many item-location combinations for planners to assess individually, making routine automation necessary simply to reduce the volume of manual intervention.
The more useful role for planners is consequently exception management. If the system can handle ordinary forecasts and replenishment recommendations consistently, staff can concentrate on supplier failures, unusual demand, constrained stock, promotions, new stores, or other cases where historic patterns provide an incomplete answer.
That model still depends heavily on data quality. Historic sales can be distorted by previous stockouts, temporary closures, promotions, assortment changes, and local events. A product that recorded no sales because it was unavailable does not represent zero demand, and a forecasting engine needs sufficient information to distinguish between the two.
Implementation across such a large estate also creates an adoption problem. Planning teams and local operators need to understand which decisions the system will make automatically, which require approval, and how exceptions are handled. If users do not trust the recommendations, they can recreate local processes outside the platform and undermine the benefit of having one planning environment.
The project nevertheless reflects how inventory planning is moving away from isolated forecasting tools. Demand, supplier lead times, allocation, ordering, and fulfilment are increasingly being treated as connected decisions because changes in one part of the chain quickly affect the others.
A stronger demand signal can also improve planning outside the inventory team. Distribution centres gain earlier visibility of expected inbound and outbound volume, transport planners can anticipate changes in store replenishment demand, and suppliers receive requirements that are based on the same forecast used downstream.
Dollar General continues to expand its store estate, which makes that common planning architecture more valuable if the implementation works as intended. Every additional outlet creates another set of product-location relationships that have to be forecast, replenished, and supplied without allowing network inventory to grow indiscriminately.
The companies have not published expected reductions in inventory or improvements in availability for the Dollar General rollout. Those outcomes should therefore be treated as objectives rather than results. The useful measures will come after implementation through forecast accuracy, stock availability, inventory turns, supplier performance, and the volume of exceptions requiring planner intervention.
At more than 21,000 stores, even modest improvements would be multiplied across a substantial physical network. The harder part is ensuring that the system can make routine decisions at that scale without losing sight of local demand and supplier variability. Dollar General is now putting that proposition into normal replenishment operations rather than restricting AI-based planning to a limited trial.
