Capability
Plan less with the gut. More with the distribution.
Demand forecasting, inventory optimisation, dynamic pricing, and routing — with confidence intervals, not just point estimates. The model lives inside your existing planning cadence, not in an isolated notebook.
Architecture
Forecast → decide → act.
What changes
Planning, before and after.
Before
Spreadsheet forecasts. Gut replenishment.
- Single-number forecasts that hide risk.
- Last-year-plus-10% as the only meaningful input.
- Stock-outs on top sellers, overstock on long-tail.
- Pricing changes are annual, not weekly.
- Nobody knows how accurate last month's forecast actually was.
After
Distributions. Decisions. Accountable.
- Forecasts with P50/P90 so risk is visible before committing.
- Weather, events, promos all modelled as signals.
- Service level and working capital both optimised, not traded off blindly.
- Pricing can be experimented weekly or daily with guardrails.
- Every forecast has a backtest score and a bias tracker.
Typical stack
Proven, not exotic.
LightGBM
NBEATS
Prophet
StatsForecast
MLflow
PuLP / Pyomo
Or-Tools
DuckDB
Postgres
Prefect
Airflow
Streamlit
How we deliver this
The same capability, three ways.
In an Audit
Forecast diagnostic
We benchmark your current forecast accuracy (WAPE, bias) on a holdout, identify where judgment adds versus subtracts value, and estimate the working-capital and service-level upside.
In an Integration
One forecasting system
Demand forecast at SKU/location, weekly regenerated, fed into a replenishment or pricing decision. Planner override UI with reason capture. 8–12 weeks.
In a Transformation
Integrated S&OP platform
Forecast + optimisation across demand, supply, inventory, and pricing. Cross-functional single source of truth with scenario planning and executive dashboards.
In production
One SKU, one week.
planning.forecast — SKU 48291 · Nairobi DC · W 2026-17
P50 forecast
4,820
units
P90 forecast
6,140
units
Recommended order
5,600
97% service
Top drivers
+18% mid-month salary effect · +12% promotion week 15 flow-through · −6% weather · baseline trend +3%
Next step
Benchmark our model against your current one.
Send us 12 months of SKU-level demand. We'll run a blind backtest and show you where the lift is — and isn't.