Demand Forecasting at a Pan-African FMCG — 22% Stockout Reduction Across 18 Countries
Demand Forecaster + Inventory Optimizer + KPI Monitor delivering granular demand forecasts across the African market complexity that defeats most global vendors.
The challenge
The client — a pan-African FMCG group with a portfolio spanning personal care, household and packaged-food categories across 18 African markets — was operating a demand-planning function whose performance had been chronically suboptimal due to the structural challenge of African demand patterns. African FMCG markets exhibit demand volatility and pattern complexity that exceed mature-market norms: FX-volatility-driven consumer-behaviour swings, weather-and-rainfall-driven demand patterns (particularly in the rural distribution segments), informal-trade demand-pattern variation, infrastructure-driven distribution-cycle variability, and the periodic supply-shock events (logistics disruption, port-and-customs delays, regional political events) that disrupt normal demand-supply matching.
The group's existing demand-planning approach was a combination of a global SAP IBP deployment (which the regional planning teams found inadequate for the African demand complexity) and a network of country-team Excel-based forecasting workbooks (which were inconsistent across countries and structurally unscalable). Forecast accuracy at the SKU-country level was averaging 16% MAPE, with the per-SKU-per-country variation reaching 35% or worse on the most volatile lines.
The combination of poor forecast accuracy and conservative inventory positioning had produced a structurally suboptimal outcome: simultaneous over-stocking (excess inventory in some categories and locations) and under-stocking (stockouts in others), with the working-capital trapped in the over-stocked categories not available to support the lines that needed it.
The approach
MindMap deployed Demand Forecaster (Df) as the SKU-market forecasting engine, Inventory Optimizer (Io) as the cross-country inventory-rebalancing layer, KPI Monitor (Kp) for the forecast-vs-actual tracking, and Data Lake Architect (Dl) for the underlying multi-country data integration.
Phase one was the data-integration build. The 18-country data estate was fragmented across local SAP instances, country-specific ERP installations and the regional SAP IBP deployment. The unified data layer was built to consolidate the historical sales data, the current inventory positions, the in-transit shipment data, the promotional-event calendars and the external macro data (FX, weather, port-and-customs status) into a single forecasting-ready dataset per country.
Phase two was the multi-model forecasting build. The forecasting platform uses a per-SKU-per-country ensemble: classical statistical methods (ARIMA, ETS) for stable mature lines, gradient-boosted trees for promotional uplift modelling, and a deep-learning Temporal Fusion Transformer for the high-volatility lines where the African-specific macro and weather signals drive forecast accuracy. The ensemble weighting is re-estimated per SKU-country combination weekly based on historical forecast-vs-actual performance.
Phase three was the cross-country inventory-rebalancing layer. Inventory Optimizer takes the country-level forecasts and the cross-country inventory positions and identifies cross-country rebalancing opportunities — where one country's over-stock could be transferred to another country's under-stock with reasonable logistics economics. The cross-country rebalancing has been the largest single contributor to the overall inventory reduction.
Phase four was the integration with the group's existing planning workflow. The platform's forecasts and recommendations feed into the existing SAP IBP environment, which the regional planning teams continue to use as the system of record. The platform does not replace IBP; it provides materially better forecast inputs to the existing planning workflow.
The pre-built building blocks
Rather than commission a ground-up build, the engagement leaned on MindMap's pre-built accelerator library — production-tested components that compress what would otherwise be a six-to-nine-month build into weeks.
Demand Forecaster
Per-SKU-per-country multi-model forecasting ensemble
Inventory Optimizer
Cross-country inventory-rebalancing layer
KPI Monitor
Forecast-vs-actual tracking with model-adjustment surfacing
Data Lake Architect
Multi-country data integration with per-country data-quality handling
The architecture
The platform runs on the group's Azure tenant with a per-country data residency where regulatory requirements apply (a meaningful fraction of the 18 African markets have data-residency frameworks) and consolidated processing where they do not. The cross-country data flow respects each country's data-protection framework.
The data-integration layer uses Kafka for the streaming data and Azure Synapse for the analytics warehouse. The data quality is materially variable across the 18 countries (some countries have clean, well-maintained source data; others have legacy systems with structural data-quality issues) and the platform's data-quality layer addresses each country's specific data-quality pattern.
Demand Forecaster's per-SKU-per-country ensemble produces a forecast horizon of 26 weeks at weekly granularity, with confidence intervals at each forecast point. The model architecture is the same across countries; the model weights and the ensemble weighting differ per SKU-country combination based on the local patterns.
Inventory Optimizer's cross-country rebalancing layer is a constrained optimisation that considers the country-level forecasts, the current inventory positions, the in-transit shipments, the cross-country logistics economics (cost per pallet per country pair, lead time, customs implications), and the per-country safety-stock policies. The recommendations are produced weekly and reviewed with the regional planning teams.
KPI Monitor's forecast-vs-actual tracking runs continuously, with per-SKU-per-country MAPE tracking, the contributing-factor analysis on the largest forecast misses, and the recommended-model-adjustment surfacing for the planner workflow.
The numbers behind the story
Forecast MAPE has improved from 16% to 8% on a rolling thirteen-week measurement across the 18-country portfolio. The improvement is most pronounced on the high-volatility lines where the deep-learning model's exogenous-signal integration (FX, weather, port-status) has the most impact.
Stockout rate has dropped approximately 22% across the portfolio. The improvement is most material on the high-velocity SKUs in the markets where the previous forecast accuracy had been weakest, and where the resulting under-stocking had been driving customer-experience and retailer-relationship issues.
Overall inventory has dropped approximately 14% across the portfolio. The reduction is driven both by the better forecast-accuracy (less defensive buffer needed) and by the cross-country rebalancing (over-stocked categories in some countries supplying under-stocked categories in others). The working-capital released has been redirected to the group's African-expansion capital programme.
Service-level (on-shelf availability at the retailer) has improved across the portfolio despite the lower overall inventory. The simultaneous achievement of lower inventory and higher service-level reflects the shift from buffer-driven stocking to forecast-driven stocking.
An unexpected outcome: the platform's per-SKU-per-country forecast accuracy has surfaced specific patterns the group's category-management teams had not previously had visibility into. Specific category-country combinations where the demand-pattern was structurally different from the regional baseline have been identified, with the category-management teams using the insights to refine the per-country product portfolio and the promotional approach.
“Global forecasting platforms have consistently failed to address the demand-pattern complexity of African FMCG markets. MindMap's platform delivered the per-SKU-per-country forecast accuracy that our planning teams had not previously been able to achieve, with the cross-country rebalancing capability that has materially improved both stockouts and inventory simultaneously. The platform changed our African-supply-chain economics.”— Chief Operating Officer· Pan-African FMCG
Why MindMap was chosen
The group had been quoted multi-year programmes by two global forecasting-platform vendors. The vendors' approaches were generic and did not address the African-specific demand-pattern complexity that the group's planning teams had identified as the structural issue. The previous SAP IBP deployment, supplemented by the country-team Excel workbooks, had been the group's previous compromise.
MindMap's accelerator-composition approach — bringing Demand Forecaster, Inventory Optimizer, KPI Monitor and Data Lake Architect together with the African-specific exogenous-signal integration and the cross-country rebalancing layer — was the structural differentiator. We could demonstrate the African-context forecasting at another pan-African manufacturer with comparable scope.
Our embedded African-FMCG expertise on the delivery team (two former demand-planning heads from peer pan-African FMCGs and a former regional-logistics director) was the third factor. The group's COO felt that the team understood the operational and demand-pattern reality of African FMCG, not just the forecasting technology.
Related deployments
Grocery Planogram Optimisation
An ML-driven planogram-optimisation platform delivered a 6.4% like-for-like sales lift across 1,200 stores — through per-store assortment and shelf-position tuning.
Global CPG Supplier Intelligence
Supplier Benchmarker + Anomaly Detector delivered continuous risk visibility across 12,000 Tier-1 suppliers — early warning on 84% of subsequent disruptions.
Pan-African Content Generation
Text Generator delivered 40x content throughput across 6 African languages and 18 markets — same brand voice, same compliance posture, structural cost reduction.
Want an outcome like this?
Start with a 2-week AI Readiness Sprint. We deliver a prioritised use-case backlog and business case grounded in what's actually buildable with our accelerator library.