Tool
Split the Bill Randomizer
One person pays the lot — and over enough dinners it evens out. Mostly.
Split the Bill Randomizer
One person pays the entire bill, chosen at random. Over enough dinners it evens out — and the tool does the arithmetic on how long "enough" actually is, which is longer than most groups expect.
The running tally is the honest part. It shows what each person has actually paid so far, the expected number of times they should have paid, and the standard deviation around that expectation.
That last number is why this method has a reputation. After ten dinners in a group of four, someone having paid five times is entirely ordinary — and it feels like being cheated.
How the Split the Bill Randomizer works — and why it's fair
The draw is a plain uniform pick over the names present, using the same unbiased index draw as the name picker. Nothing is weighted, nothing is remembered between draws, and the person who paid last night is exactly as likely to pay tonight.
The tool is fair in expectation, which is a precise and limited claim. Over n outings in a group of k, each person expects to pay n/k times, and each person's expected total spend equals their share of the total bills. That is the mathematical sense in which this is fair, and it is true.
What it is not is fair over any short run, and the tool computes exactly how unfair. The number of times a given person pays follows a binomial distribution with standard deviation equal to the square root of n × (1/k) × (1 − 1/k). For ten outings in a group of four, that is about 1.4 — so someone paying five times against an expectation of 2.5 is under two standard deviations out, which happens routinely.
Convergence is slow in a way that matters. The absolute gap between the luckiest and unluckiest person grows with the square root of the number of outings; only the proportional gap shrinks. After a hundred dinners the ratio has narrowed considerably, but the person who has paid twenty-nine times has still paid four more times than the person who paid twenty-five, and neither of them experiences that as a ratio.
Nothing is stored between sessions. The tally lives in the page and resets when you close it, which means the honest version of this method requires somebody to keep the record elsewhere — and a group that will not keep a record is a group where this method will eventually cause an argument.
When the Split the Bill Randomizer is fair — and when it is not
What it does guarantee
- Every person present has exactly the same chance on every occasion, with no weighting and no memory.
- Expected spend per person equals their fair share of the total, over a long enough run.
- The tally, the expectation and the spread are all shown, so nobody has to take the fairness on trust.
What it does not
- It is not fair on any given night, and not reliably fair over ten or twenty nights either.
- Someone who leaves the group early cannot be compensated. Their imbalance is simply permanent.
- The tally is not saved. Without an external record the group is relying on memory, which reliably favours whoever remembers loudest.
Two worked examples
A group of 4 over 10 dinners
- Each person expects to pay 2.5 times, with a standard deviation of about 1.4.
- Someone paying 5 times is under two standard deviations out — ordinary, not suspicious.
- The gap between the most and least frequent payer will typically be 3 or 4 occasions.
The same group over 100 dinners
- Expected 25 each, with a standard deviation of about 4.3.
- The absolute gap has grown — 31 against 20 is unremarkable — but the ratio has narrowed considerably.
- That is the shape of convergence: proportional fairness improves while absolute gaps widen.
Groups this works for, and groups it does not
It works for a stable group that eats together frequently, where the bills are of similar size and everybody expects to still be in the group in two years. All three conditions matter, and the third one is the one people forget.
It works badly for groups with very unequal bills, because expectation is about the count of times you pay rather than the amount. Paying once for a hundred-pound dinner and never again is not equivalent to paying four times for twenty-pound lunches, and nothing in the method notices the difference.
It works badly for groups with very unequal incomes, because the variance falls on everyone equally in absolute terms and unequally in real terms. A hundred-pound hit is a different event for different people, and randomness distributes it without regard for that.
For splitting a bill into random unequal shares that still sum to the total, the percentage allocator does that with a floor so nobody pays everything. For an even split, arithmetic is faster than any tool.
Frequently asked questions
Is this actually fair?
Fair in expectation, over a long run. On any given night one person pays everything, and over ten nights the gaps are large and normal.
How long until it evens out?
Proportionally, quite quickly; in absolute terms, never. The gap between the luckiest and unluckiest grows with the square root of the outings — only the ratio narrows.
Someone has paid five of the last ten times. Is it rigged?
Almost certainly not. In a group of four, the standard deviation over ten outings is about 1.4, so five against an expected 2.5 is under two standard deviations out.
Does the tool avoid picking last night's payer?
No. Every draw is independent, so the same person can pay twice running — which happens about once in every k draws.
Does it account for different bill sizes?
No, and this is its biggest real weakness. It counts occasions, not amounts, so one expensive dinner and one cheap one are treated identically.
Is the tally saved?
No. It lives in the page and resets when you leave, so a group relying on this method needs to keep the record somewhere else.
What if someone leaves the group?
Their imbalance is permanent and cannot be corrected. If people join and leave often, this method is a poor fit.
Can I weight it so someone pays less often?
Not on this page. The weighted random picker can do that, and it prints the resulting percentages so everyone can see the arrangement.
What is the standard deviation figure telling me?
The typical size of the gap between what someone has paid and what they were expected to pay. It is the number that makes an apparently unfair run look ordinary.