Where we are
Last session: why a business forecasts, and the two-sided cost of getting it wrong. This session is the vocabulary you need for every session after it. A forecaster's whole job is to pull a series apart into its pieces, project the pieces that persist, and be honest that one piece - the noise - simply cannot be projected. If you can name the four building blocks and tell seasonality from a cycle, you can read any forecast in the building. Our running example stays Lumen, forecasting its daily website sessions and signups.
A series is four things stacked together 8 min live
Take Lumen's daily sessions and pull them apart. Under the messy line there is a trend (the slow drift, up as the brand grows), a seasonality (the weekly rhythm - quiet weekends, busy Mondays), a longer cycle (multi-month swings with the wider market), and noise (the daily wiggle nobody can explain). Add them back up and you get the observed line.
LiveThe four blocks, in one line each3 min▶
- Trend - the slow drift up or down over a long stretch. Lumen's sessions climbing as the brand grows.
- Seasonality - a pattern that repeats on a fixed period: daily (busy afternoons), weekly (quiet weekends), or yearly (holiday spikes).
- Cycle - a longer swing with no fixed length, tied to bigger forces like the economy or a category trend.
- Noise - the irregular wiggle left over. Real, but unforecastable - do not let anyone promise to predict it.
Why a leader cares about the split. When a team says "our forecast missed by 8% on Tuesday", the right question is: was that the signal we should have caught, or the noise nobody can? Missing a clear weekly pattern is a bug. Missing an unrepeatable random spike is just life. Knowing which is which tells you whether to push the team or defend them.
Self-studyLevel: the current height of the line2 min read▶
One more word you will hear: level. It is simply where the line sits right now - its current baseline height, before trend and seasonality push it around. Think of it as "roughly how many sessions a normal day gets at the moment". A forecast starts from today's level, then adds the trend and the seasonal shape on top.
You do not need the math. You just need to hear "level" and picture the current resting height of the line - the starting point everything else adjusts from.
Seasonality vs cycle: fixed clock vs loose swing 6 min live
These two get confused constantly, and the difference decides what a forecast can promise. Seasonality repeats on a fixed, known period - a week, a year - so you can project it confidently. A cycle swings over a longer, irregular stretch (think a market boom and bust) with no fixed length, so it is far harder to call. Same-looking wiggles; completely different forecastability.
LiveWhy this matters when you read a forecast3 min▶
The distinction is not academic - it changes what you should believe:
- Seasonality is a safe bet. "Weekends are quiet, December spikes" repeats on the calendar. A forecast that captures it deserves your trust.
- Cycles are a warning. When a team explains a big miss with "we are in a downturn", they are pointing at a cycle - real, but genuinely hard to time. Do not punish a team for failing to predict a turning point nobody could.
- Ask "is this on a fixed clock?" If yes, expect it to be forecast well. If no, expect a wider range and more humility.
Signal you can project, noise you cannot 5 min live
This is the most useful single idea for a leader. A series splits into signal - the persistent trend and seasonality you can carry forward - and noise - the irregular part that, by definition, cannot be predicted. A good forecast captures the signal and puts an honest range around the noise. A bad one pretends the noise is signal and promises precision it cannot have.
Read Lumen's series out loud ◐ 10 min · together
No code - just naming the pieces, which is a skill you use every time a dashboard lands on your desk.
Find the trend. Over the last year Lumen's daily sessions have drifted up as the brand grew. That slow climb is the trend - the piece a forecast should extend.
Name the seasonal period. Sessions dip every weekend and rise midweek - a weekly pattern on a fixed clock. That is seasonality, and it is safely forecastable.
Spot the cycle, carefully. A months-long dip during a category slowdown is a cycle, not seasonality - no fixed length, so expect a wider range there.
Point at the noise. One random Tuesday spike from a viral post is noise - do not expect any model to have called it, and do not build a plan around it repeating.
Try it yourself - this week ◐ 20-30 min total
- Pick a series you know well - web traffic, sales, tickets, headcount demand - and pull up its chart over at least a year.
- Name the trend in one line: is it drifting up, down, or flat?
- Name the seasonal period: is the repeating pattern daily, weekly, or yearly? Say the clock out loud.
- Name one source of noise: an irregular event that moved the line but will not repeat on a schedule.
- Bonus: is there a longer cycle you can see - a multi-month swing with no fixed length? Flag it as "forecast with caution".
- Optional: skim FPP3 chapter on time-series decomposition (Hyndman & Athanasopoulos, otexts.com/fpp3).
Sources covered (cited)
The building blocks come straight from the forecasting canon. This session covers:
Three questions before you go 🎯 ◐ 90 seconds
1 · What are the four building blocks of a time series?
Any series is a slow trend, a repeating seasonality, a longer irregular cycle, and random noise, stacked together. Forecasting projects the persistent pieces.
2 · What is the difference between seasonality and a cycle?
Seasonality is on a known calendar clock, so it is forecastable. A cycle (like a market boom-bust) has no fixed length and is far harder to call.
3 · Why does the signal-vs-noise split matter to a leader?
Noise is random and unrepeatable - it cannot be predicted, only bounded by a range. Anyone promising to nail every day is pretending noise is signal.