First order
The size of a customer's first order is the one thing known on the day they arrive, and it carries most of the answer: the top fifth by first order went on to be worth ten times the bottom fifth.
The hard edge running diagonally across the chart is not an artefact. A customer who ordered once is worth their first order less whatever they sent back, so 1,326 of the 1,504 sit exactly on the line y = x, the other 178 fall below it, and none can rise above it. The gold band is that line; the blue cloud above it is everyone who came back, and returns pull only 21 of them below it.
Why the correlation coefficient is the wrong number here
The temptation is to quote a Pearson r, which is 0.643 and looks convincing. It is almost entirely one customer: the largest first order in the file is £33,168, from an account that went on to spend £233,804, and deleting that single row takes r down to 0.400. Below £5,000 of first order it is 0.348. A statistic that moves by a quarter of its range when you remove one of 4,868 points is measuring the outlier, not the relationship.
The rank correlation does not move at all: 0.598 with the outlier, 0.598 without it, 0.595 below £5,000. Ranks are indifferent to how far out the far end goes. That is the number to quote, and the quintile table is what it means in money.
Median lifetime value runs £190 → £406 → £705 → £892 → £1,904 across the quintiles. The top fifth spent 6.8 times what the bottom fifth did on day one (median £745 against £110) and went on to be worth 10.0 times as much, so the signal amplifies instead of merely persisting, and that fifth accounts for 45.1% of the cohort's revenue.
One wrinkle: the repeat rate is not monotonic. It climbs 58% → 64% → 74%, dips to 71% in the fourth quintile, then reaches 78% in the fifth. Whatever separates a £300 first order from a £400 one, it is not whether the customer comes back.
So what. This is the only genuinely forward-looking number in the retail section. The concentration curve says a small group carries the business; this says that group is largely identifiable on the day it arrives, from data that already exists at the point of sale. A first order over £508 puts a customer in the fifth that generates 45% of revenue, at just over 78% odds of ordering again.
Who acts: whoever owns acquisition spend and onboarding. Bidding the same amount for every new customer, or running the same welcome sequence at all of them, is leaving the difference between £190 and £1,904 of expected value on the table.
Cost of getting it wrong: the causation runs the other way just as easily. A big first order may signal a bigger business without creating a better customer. So this identifies who to look after; it does not say that pushing a first order from £200 to £500 buys you the difference in lifetime value. The customers here are shops, and a shop's opening order is mostly a statement about the shop.