The calculation uses 95% confidence. If you don't enter passives, it makes the most cautious assumption: no passives at all. It assumes different people responded in each period. If your respondents don't represent your customers (for example, if only people with complaints replied), no calculation can fix that.
Adjusted Wald method (Lewis and Sauro, MeasuringU, 2021; Rocks, 2016). With few responses it pulls the result towards zero and widens the interval, which is why it is cautious. The method's authors note that they did not separately test its accuracy on small samples; with few responses, treat the result as a rough guide.
Each response counts as +1 (promoter), 0 (passive) or −1 (detractor); NPS is 100 times their average. Three virtual responses are added to each period (0.75 promoters, 1.5 passives, 0.75 detractors), so that not even a single response is treated as “certain”. The difference between the two periods is checked with a z-test: if |z| ≥ 1.96, the difference is significant at the 95% level.
The bands show each period’s 95% confidence interval. The difference can be significant even when the bands overlap; the verdict comes from the difference test, not from the bands.
Sources: Lewis and Sauro, “How to Statistically Compare Two Net Promoter Scores” (2021); Rocks, “Interval Estimation for the ‘Net Promoter Score’” (2016).