The 3-yearly moving average replaces each year (except the first and last) with the average of itself and its neighbour on each side, which smooths the series into a steadily rising trend.

Step 1 — Set up the 3-yearly moving totals. Take three consecutive years at a time, sliding forward by one year each time:

• 2016 (middle of 2015-16-17): $12+15+21 = 48$
• 2017 (middle of 2016-17-18): $15+21+24 = 60$
• 2018 (middle of 2017-18-19): $21+24+27 = 72$
• 2019 (middle of 2018-19-20): $24+27+33 = 84$
• 2020 (middle of 2019-20-21): $27+33+36 = 96$

Step 2 — Divide each moving total by 3 to get the moving average (trend value), and place it against the middle year of its block:

Verification: re-adding each block independently confirms the totals: $12+15+21=48$ ✓, $15+21+24=60$ ✓, $21+24+27=72$ ✓, $24+27+33=84$ ✓, $27+33+36=96$ ✓; dividing each by 3 gives the same trend values 16, 20, 24, 28, 32, which rise by a constant 4 each year — a smooth, believable trend compared with the more jagged original sales figures.

Forgetting that the moving average is placed against the middle year of the block, not the first or last year of it; trying to compute a trend value for the first and last years even though no full block of 3 years can be centred there; and adding the wrong three years (e.g. re-using a year twice) when sliding the block forward.