FORECASTING
Demand forecasting: plan for what customers will ask
Demand forecasting with TIMi estimates the volume of products or services that will be requested over a future period, and shows the factors behind each forecast.
Forecast per product
From one line to thousands of products or stores at once.
Calendar effects
Seasonality, promotions and holidays are taken into account.
Explained forecast
Each forecast comes with the factors that weigh in it.
HOW IT WORKS
How demand forecasting works.
Three steps, in one engine, on a standard server with no cluster.
1. Prepare the data
Anatella gathers the sales or consumption history, seasonality, known events (promotions, holidays) and relevant external signals (weather, economic context), then structures them into variables.
2. Build the model
Modeler builds a model that projects future demand from historical trends and the factors identified as influential.
3. Forecast and explain
Each forecast comes with its factors, such as seasonality, recent trend or the effect of an identified event, and not only a projected number.
Anatella
ModelerGETTING IT RIGHT
Choices that shape a demand forecast.
The right method depends on what drives your demand.
Pick the right method
For a clean seasonal series, Anatella’s Time Series action compares ETS, STLF, ARIMA and Holt-Winters side by side. When demand also depends on weather, neighboring stores or local dynamics, Modeler uses these as variables.
Choose the horizon
The horizon depends on the sector and on the representative history available. A short horizon (weeks) is usually more precise than a long one (months, seasons).
Plan for new products
Without a history of its own, a product is harder to forecast. A model can rely on the sales of comparable products for a first forecast, then adjust it with real data.
EXPLAINABILITY
A forecast says how much. The factors say why.
A forecast alone gives an expected volume, not what justifies it. A white-box model shows the factors in the projection, so planners can tell whether a rise comes from known seasonality, a one-off event or an underlying trend, and adjust their decision.
For planning and purchasing teams
A forecast can be questioned and adjusted with business context, instead of being followed as a black box.
For reliability over time
Understanding the factors of a forecast helps to detect earlier when a change of context makes it obsolete.
For model owners
A totally new event is hard to anticipate by nature. Modeler’s built-in Drift Report shows when a model loses accuracy, so you update it when needed instead of on a fixed calendar.
DATA
What data a demand model uses.
Anatella connects to more than 80 sources, so this data can be gathered without prior custom development.
Sales or consumption history
At the level of detail that matters.
Event calendar
Promotions, holidays, launches.
Seasonality
Seasonal factors already identified.
External signals
Weather, economic indicators, depending on the sector.
PROOF
Demand forecasting at scale.
Bavaria is our reference for demand forecasting. We do not yet have a second published customer story on this use case.
300,000 stores forecast with 30 million transactions
Demand depends on weather, neighboring stores and local dynamics, so classical time series models were not enough. The models score 2.4 million records per second. Reported forecast accuracy ranges from 81% to 92%.
FAQ
Common questions.
What horizon can demand be forecast on?
It depends on the sector and on the availability of representative history. A short horizon (weeks) is generally more precise than a long one (months, seasons).
Does it work for a new product with no history?
It is harder without a history of its own, but a model can rely on the sales of comparable products to build a first forecast, adjusted afterwards with real data.
How does the model handle an exceptional event?
A totally new event remains hard to anticipate by nature. Model drift monitoring detects quickly a large gap between forecast and reality, so the projection can be adjusted.
Per product or aggregated?
Both are possible. A fine forecast per product or store is usually more useful for operational planning, but it needs more data for each forecast entity.
RELATED USE CASES
Keep exploring.
Sales Forecasting
Forecast revenue from the sales pipeline and past conversion, with the factors explained.
Inventory Forecasting
Translate expected demand into stock levels that limit stock-outs and overstock.
Seasonal Model Recalibration
Adapt a model to predictable seasonal changes without rebuilding it each cycle.
SEE FOR YOURSELF
Test it on your own data.
Run TIMi on your own data with a free trial, or talk to us about your case.
