The meeting where the median won

Average order value was $84. It had been $84-ish for two years, it was on the wall, and the free-shipping threshold was set at $75 on the reasoning that most customers were nearly there already.

The median was $41.

The threshold experiment had failed twice and nobody could explain why. The explanation was that we had designed for a customer who did not exist. Most people were not "nearly there" — they were asking to double their basket for free delivery, which is not a nudge.

The conversation that changed it was not an argument. I added the median to the same tile as the mean, with the gap visible, and said nothing for a month.

select
  round(avg(order_total), 2)                                        as mean,
  round(percentile_cont(0.5) within group (order by order_total), 2) as median,
  round(percentile_cont(0.9) within group (order by order_total), 2) as p90
from orders;

Someone in a planning meeting pointed at it and asked why they were so far apart. That was the whole intervention. Explaining skew when a person has just noticed it themselves takes about forty seconds; explaining it unprompted takes a quarter and does not stick.

The threshold moved to $55. It worked.

I have since stopped trying to remove the mean from anything. Adding the median beside it does the job, and it does not require anyone to admit they were wrong.