Pull any time series apart into trend + seasonality + residual. Adjust each component's strength and see how the observed series rebuilds itself.
OBSERVED Y(t) = T + S + R
TREND
SEASONAL (period = 12)
RESIDUAL (irregular noise)
0.20
8
2
Reading it: if the seasonal wiggle stays roughly constant over time, use additive. If wiggles grow as the trend grows (think retail sales), use multiplicative — then take logs to turn it into additive.