How this track works
Six sessions, no code, one running story: Lumen, a skincare brand forecasting its website's daily sessions and signups so it can plan stock, staffing, and spend. You will learn to read a forecast, judge whether it is any good, and know when to trust it - the leader's job - while the builder track (b1-b10) writes the actual Python. Today is the foundation: why a business forecasts at all, what it costs to get wrong, and where the payoff is real. A running promise runs through every session: a forecast is a probability, not a guarantee, so we always ask for the range of outcomes, never a single confident number.
A forecast is a decision made early 6 min live
Strip away the technique and forecasting is simple: it is committing a resource before you know the demand. You order stock in March for the summer. You hire in Q2 for the Q4 rush. You reserve servers before the campaign spikes traffic. In each case you act now against a number that arrives later - and a forecast is your best honest estimate of that number, with a range around it.
LiveThe five places every business forecasts3 min▶
The same lever shows up across the whole business:
- Demand & inventory - how much to make or buy so shelves are neither empty nor overflowing.
- Staffing - how many people to roster for the call centre, the store, the warehouse next week.
- Cash-flow - how much cash will come in and go out, so you never run dry or leave money idle.
- Capacity - how much server, factory, or delivery capacity to provision before the load hits.
- Web traffic - how many sessions and signups to expect, so marketing, infra, and support are ready.
Lumen's one number, many decisions. Lumen forecasts its daily website sessions and signups. That single forecast tells the infra team how many servers to keep warm, tells support how many agents to schedule, and tells finance how much revenue to expect from new signups. One forecast, three teams planning off it - which is exactly why getting it right matters.
Self-studyForecast vs target vs plan - three different things2 min read▶
Leaders mix these up constantly, and it causes real damage:
- A forecast is what you honestly expect to happen, given the evidence. It is neutral.
- A target is what you want to happen. It is aspirational and often political.
- A plan is what you will do about the forecast - the stock, staff, and cash you commit.
When a "forecast" quietly becomes a target ("make the number say 20% growth"), you have stopped forecasting and started wishing - and the plan built on it will be wrong. Keep the forecast honest; argue about the target separately.
Being wrong costs money both ways 7 min live
The single most important idea in this whole course for a leader: a forecast error is expensive in both directions, and the two costs are almost never equal. Forecast too high and you waste - dead stock, idle staff, cash tied up. Forecast too low and you fall short - stockouts, missed demand, angry customers. The whole game is balancing those two costs, not chasing a perfect single number.
LiveLean toward the cheaper mistake, on purpose3 min▶
Because the two costs are unequal, the smart forecast is not "dead centre" - it deliberately leans toward the side that hurts less:
- If stockouts are brutal (a launch, a loyal customer you would lose) - carry a little extra. A bit of dead stock is cheaper than a lost customer.
- If waste is brutal (fresh food, perishable capacity, expensive inventory) - forecast lean and accept the occasional shortfall.
- Name the asymmetry before you plan. Ask: "which direction of error hurts us more, and by how much?" That answer, not the raw forecast, sets your stock and staffing.
Every planning question is a forecast question 6 min live
Forecasting pays off wherever you must commit a resource ahead of demand and the cost of being wrong is real. If the decision is reversible and cheap, a forecast is overkill. If it locks in money, people, or capacity for weeks, a good forecast - with its range - earns its keep many times over.
LiveWhen to forecast, and when to just use the baseline3 min▶
Building a real forecasting model is not always worth it. Before you commission one, ask:
- Is there a decision waiting on it? No decision, no forecast. A number nobody acts on is trivia.
- Does the error cost enough to justify the effort of building and maintaining a model?
- Is the past any guide to the future? A brand-new product with no history, or a genuine one-off event, cannot be forecast from a series that does not exist yet.
When the answer is "not really", use the baseline instead: "next week looks like last week" or "the same week last year, adjusted for growth". It is free, honest, and often good enough. The rule that runs through this whole course: a fancy model has to beat that simple baseline, or it is not worth the complexity.
The humility lesson from the M-competitions. The largest public forecasting bake-offs (Makridakis' M4 and M5) found again and again that simple methods - and combinations of them - are shockingly hard to beat. For a leader, that means: be sceptical of any vendor or team promising a complex model without showing it beats the boring baseline first.
Frame one decision as a forecast ◐ 10 min · together
No code here - just the leader's reasoning. Take one real decision from Lumen and walk it through the lens you just learned.
Name the decision. Lumen must decide how many support agents to schedule for next week. That commitment is made now, days before the demand arrives.
Name the number it waits on. The decision leans on forecasted daily website sessions and signups - more traffic means more support tickets.
Name the cost each way. Too few agents means long waits and churn (shortfall). Too many means paid, idle staff (waste). Which hurts Lumen more this quarter?
Decide the lean. If Lumen is protecting a fragile launch reputation, lean toward slightly over-staffing. If margins are tight, lean lean. State it out loud before anyone models anything.
A forecast is a probability, not a promise 3 min live
This is the rule that keeps you out of trouble, and it runs through all six sessions. A forecast is your best honest estimate of an uncertain future - it comes with a range, and it can be wrong. Anyone who hands you a single confident number with no range is selling you false certainty.
Try it yourself - this week ◐ 20-30 min total
- List three decisions in your org that a forecast feeds - stock, staffing, cash, capacity, traffic, anything you commit ahead of demand.
- For each, write the cost of being too high (waste) and the cost of being too low (shortfall) in one line each.
- Mark which direction hurts more for each decision - that asymmetry is where you will lean.
- Pick one decision and write a single sentence: does it deserve a real forecast, or is a simple baseline ("like last week / last year") good enough?
- Optional: skim the intro of FPP3 (Hyndman & Athanasopoulos, otexts.com/fpp3) - the free, honest textbook this course follows.
Sources covered (cited)
Framed from the forecasting canon and the field's biggest empirical evidence. This session covers:
Three questions before you go 🎯 ◐ 90 seconds
1 · What is a forecast, in one line?
A forecast is what you honestly expect, with a range - not a promise and not a wish. A target is what you want; a forecast is what you expect.
2 · Why does the "two-sided cost of error" matter to a leader?
Over-forecasting wastes (dead stock, idle staff); under-forecasting falls short (stockouts, lost sales). Naming which side hurts more sets the plan.
3 · When should you skip building a forecast and just use a baseline?
No decision, cheap error, or no history means a model is overkill. Use "like last week / last year" - and remember any real model must beat that baseline to be worth it.