Demand Forecasting Basics for Operations
Build a sensible demand forecast, compare it to simple baselines and know how wrong it might be
A taste of a lesson
Our AI tool's forecast had 30 percent MAPE last month. Is that bad?
It depends, and the number alone cannot tell you. First, compare it to a simple baseline on the same period, such as using the same week last year. If the baseline scored 25 percent, the tool is worse than doing almost nothing. Second, check whether a few low volume products inflate the figure, because MAPE explodes when actual sales are tiny. Third, look at bias: are you consistently over or under? Exercise: calculate the seasonal naive error for your top five products last month and compare.
Written by the teacher as an example. In your lesson the tutor answers your own questions, and like any AI it can be wrong.
What you will be able to do
- Define a forecast by its decision, horizon and level of detail
- Prepare sales history while marking stockouts, promotions and one off events
- Compare any model against naive and seasonal baselines through backtesting
- Measure accuracy with MAE, MAPE and bias, and know their pitfalls
- Use forecast ranges and documented human overrides in planning
Lesson plan
- 1 What the forecast is for Tie the forecast to a decision, a time horizon and a level of detail. Start
- 2 Preparing the history Clean sales data and mark periods that do not reflect true demand. Start
- 3 Baselines first Build naive, seasonal naive and moving average forecasts as benchmarks. Start
- 4 Measuring accuracy honestly Backtest forecasts and measure error with appropriate measures. Start
- 5 Where models and AI help Know when statistical models, machine learning or AI assistants add value. Start
- 6 Ranges, judgment and new products Plan with forecast ranges and combine models with documented human input. Start
Try asking
About this tutor
For operations, inventory and planning staff who need forecasts for stock, staffing or production, and who are hearing that AI can predict demand. This tutor teaches the fundamentals first: what a forecast is for, how to prepare sales history, why simple baselines such as last period or seasonal averages are hard to beat, and how to measure accuracy honestly. You then learn where machine learning and AI assistants help (more variables, many products, explaining patterns, drafting analysis) and where they mislead (short histories, sudden shocks, new products). Lessons include forecast error measures, forecasting ranges rather than single numbers, and combining statistical forecasts with human knowledge of promotions and events.
Reviews
4.3
3 ratingsSample
- Emeka J.Sample
Practical and honest. I liked that it kept saying all forecasts are wrong and focused on measuring how wrong.
- Sunita R.Sample
Clear on MAPE versus MAE, finally. Marking stockout periods changed our history a lot. A bit more on safety stock maths would help.
- Viktor P.Sample
The baseline lesson was humbling. Our expensive forecast barely beat 'same week last year' for most products. We now report both every month.
About the teacher
Operations teaching: map the process first, then decide where AI earns a place
9 tutors 412 lessons taught Sample
I teach operations people how to improve real processes with AI, carefully. I come from operations and supply chain roles where a small error in a spreadsheet could mean a late shipment or a wrong payment, so I teach with a strong habit of checking. We map the process before touching any tool, measure where time and errors actually go,...
See Takeshi's profile and tutorsMore like this
Other tutors on the same or nearby topics.