Who Owes What to Whom? How a Group Actually Settles Up

Someone always asks it on the last evening, usually while the table is being cleared: so, who owes what to whom? Four people then produce four different answers, and none of them is dishonest. They are simply answering a different question from the one that was asked.

The short answer. Nobody owes anything for a specific expense. Each person owes, or is owed, a single net amount: everything they advanced, minus everything they actually consumed. Work that out for each member and you get the balances. Then the group does not have to pay those balances one debt at a time, because most of them cancel out: a group of four can settle six potential debts with three transfers. The only case where this tidying must not happen is when several currencies are involved, because two currencies cannot offset each other without a conversion that changes the amounts.

Why nobody in the group gets the same number

Because everyone is tracking the wrong half of the problem. What you remember is what left your card: the hotel, the two dinners, the taxi from the airport. What actually decides your balance is that total minus your share of everything the group consumed, including the things other people paid for you.

The two numbers are only equal if you paid exactly your share, which basically never happens. In our own data, the average group has 3.68 participants and logs 21 expenses. Nobody holds twenty one expenses and four people’s shares in their head, so the group falls back on the part it can remember, and four people remember four different things.

Who owes what to whom, step by step

The method has three steps, and every expense splitting app uses the same one.

1. Add up what each person advanced. The full amount they paid, regardless of who it was for.
2. Add up what each person consumed. Their share of every expense, including the ones someone else paid.
3. Subtract. Paid minus consumed. A positive result means the group owes you that amount. A negative result means you owe the group.

That third number is the balance, and it is the only figure the whole table can agree on, because it does not depend on which expense you happen to remember. It also has a useful property: all the balances in a group add up to zero. If they do not, an expense is missing or someone was left out of a split.

Note what the balance is not. It is not a list of debts to specific people. You do not owe the person who paid for dinner, you owe the group. Who you actually transfer money to is a separate question, and a much more interesting one.

Why four friends end up with six repayments

Once you have the balances, you have to turn them into transfers. Done naively, expense by expense, the number of debts grows fast: in a group of n people, there are n(n-1)/2 possible pairs. Four friends means six debts. Six friends means fifteen. Groups of exactly four are 14.8% of our base, so this is the standard case, not an edge case.

Almost all of those transfers are pointless, because they cancel each other. Splitwise explained the principle on its own blog back in 2012, and it has not changed since: “if Anna owes Bob, and Bob owes Cathy, then Anna can just pay back Cathy directly”. The post gives the reason in one line: “Fewer payments = less waiting for your money = more good times.”

Mathematically, a group of n people never needs more than n-1 transfers to get everyone back to zero. Four friends, three transfers. That is what optimised transfers does in Banana Split, and every serious app in the category does some version of it. Nobody’s balance changes, only the route the money takes.

What changes when the trip crosses several countries

Here the tidying up above becomes actively wrong, and this is the part most apps get lazy about.

Say the group spends ringgit in Malaysia, then yuan in China. You owe 30 MYR to one friend and another owes you the equivalent of 12 CNY. Cancelling one against the other looks efficient, but it quietly does two things you did not ask for. It picks an exchange rate, on a date nobody agreed on, and it turns a debt you could settle in cash on the spot into a debt that has to survive a currency conversion.

This is rarer than travel blogs suggest, and much heavier when it happens. Among the groups where the currency is recorded, 2.21% mix several currencies. But 16.4% of those go past two currencies, one group reached ten, and they log 47.9 expenses on average against 21.1 for a normal group. A multi-country trip is not a marginal case, it is the biggest event most groups will ever track.

So Banana Split lets you choose how the balances are read. As soon as an event contains more than one currency, a selector appears at the top of the balance screen, with two options:

• Converted to the group currency, the default. One balance per person, everything brought back to a single currency at the daily rate. This is what you want when the trip is settled once, back home.
• One balance per currency. The settlement runs once per currency, on that currency’s expenses only, and the results are never mixed. Your 30 MYR debt stays a 30 MYR debt. It is not cancelled by a credit in yuan.

The second mode keeps everything else intact. Debts are still simplified, just inside each currency, so you still get the minimum number of transfers, currency by currency. The screen shows one table per currency, and a currency where everyone is square keeps its line and says so, which is genuinely useful information: it means Malaysia is closed and only China is left open.

One honest note: reading balances per currency is part of the multi-currency set, which sits behind the paid tier, with a free trial. Mixing several currencies inside one group is free. Splitting the balance apart, currency by currency, is not. We wrote up the whole multi-currency problem separately, including the repayment side, and you can see where it sits in the feature list.

Should you settle up as you go, or at the end?

At the end, in almost every case. Repaying during the trip means converting the same money twice, and it hides the picture: as long as expenses are still coming in, a balance is a snapshot, not a conclusion. The exception is the person who ends up advancing everything, which is common enough that we gave it its own article. If one person is carrying the group’s cash flow for a fortnight, a partial repayment halfway through is fair, and it costs one transfer.

What matters more than the timing is that everything is logged before anyone starts calculating. A balance computed on 90% of the expenses is not 90% right, it is wrong, and it is wrong in a way that always favours whoever forgot to log something.

If your group is about to do this the hard way, on a napkin, with four people who each remember a different half of the trip, download Banana Split and let the arithmetic happen on its own. We also keep an honest comparison of what we do and what Splitwise and Tricount do, including the rows where they win. And if your group already uses Tricount, here is how Tricount works out the balances and what it cannot do.

Where the numbers come from

The figures about group behaviour come from 186,833 groups tracked in Banana Split, each with at least two members and five expenses, 3.95 million expenses in total, exported on 15 September 2026. The currency statistics are computed on the 35% of those groups where the currency of each expense is explicitly recorded, so they describe the groups we can actually measure rather than the whole base. The data is aggregated and anonymous. Everything about Splitwise comes from the page linked in the text, read on 16 September 2026.