Forecasting is estimating the future from the past, and understanding its nature (especially uncertainty) is critical. Let us cover: 1) What a forecast is: — An estimate of FUTURE values of a quantity based on PAST DATA (and possibly other factors). — Usually we work with a TIME SERIES: values of a quantity over time (sales by day/month, number of users by week). From the series history we estimate future points. — The goal: to help PLAN — inventory, budget, staff, capacity — for the expected future. 2) Where it is applied (very practical): — Demand/sales: how much to buy/produce (under-ordering = lost sales, over-ordering = frozen money/write-offs). — Finance: forecasting revenue, expenses, cash flow for budgeting. — Resources: how much staff/capacity for the expected load (a call center, servers). — Metrics: user growth, traffic, any indicators over time. — A good forecast saves money and improves decisions; a bad one (or blind faith in it) is a source of costly errors. 3) THE KEY PROPERTY — UNCERTAINTY (most important for the whole course): — A forecast is an ESTIMATE, NOT exact knowledge of the future. The future is not predetermined; it is affected by many factors, some of which are unpredictable. — Any forecast has ERROR. There is no exact number "sales will be exactly 1000" — reality will be around the forecast, with a deviation. — So an honest forecast includes NOT ONLY an "estimate" (e.g. 1000) but also a notion of UNCERTAINTY (e.g. "most likely between 850 and 1150"). A range/interval is more honest than one number. — The further into the future, the MORE the uncertainty (a forecast for tomorrow is more accurate than for a year ahead). — The danger: treating a forecast as an EXACT fact to building rigid plans on an uncertain estimate to costly errors when reality deviates. 4) What makes a forecast possible (and where the limits are): — A forecast works when the data has STABLE PATTERNS continuing into the future: a trend (the general direction), seasonality (repeating cycles), a level. — A forecast CONTINUES past patterns into the future, assuming they persist. — THE LIMIT: if something CHANGES FUNDAMENTALLY (a new product, a crisis, a market change, a one-off event), past patterns may not persist, and a forecast from history will be wrong. Forecasts poorly predict SHARP TURNING POINTS and unprecedented events — they extrapolate the past rather than foresee breaks. This is a fundamental limitation. 5) The role of AI and statistics (a course overview): — There are methods from simple (statistical — a moving average, extrapolating a trend/seasonality) to complex (AI/machine learning). — AI helps build forecasts but, as everywhere in the data course, speeds up the technique; understanding and honesty about uncertainty are with the human. — The course will cover: time series and their components, forecasting methods (from simple to AI), how AI helps, and CRITICALLY — how to evaluate accuracy and honestly treat uncertainty. Practical rule: forecasting is estimating future values from past data (usually a time series) to help plan. Its KEY property is UNCERTAINTY: a forecast is an estimate with error, not exact knowledge; an honest forecast includes a range/uncertainty, not just one number; uncertainty grows with the horizon; forecasts poorly predict sharp turning points (they extrapolate the past). Treat a forecast as a grounded estimate with error, not a fact, and plan accounting for uncertainty. This is the foundation of honest and useful forecasting that the course will develop: methods, the role of AI, accuracy evaluation — all around making useful forecasts while honestly understanding their limits.