How Budgetto's monthly forecast works, and how many months it needs

The forecast takes all your transactions, income and expenses, and works out an expected amount per category for the next three months. That gives an expected "In", "Out" and "What's left" for each month. Refunds are subtracted from the expense first, so a returned order doesn't count as a cost.

Each category is forecast separately, and within it every shop with enough history of its own (at least three transactions in two different months) gets its own series. Otherwise a shop you have only been going to since May would have the months before counted as "zero", and that drags the average down. Odd one-off expenses go together in a catch-all group.

What isn't in it: your savings, your investments and the list of fixed costs. Why that last one is left out is explained in section 03.

The honest answer: two to say something, twelve to say it well. This is what happens at each threshold.

• Under two months. The app shows "1 of 2 months tracked" and no figure. The current month counts, so if you start on 28 August, you already have two on 1 September; the figure then still rests on very little.
• From two months. Fixed patterns are recognised and the trend is extended. Confidence is set to "low", and the app says so.
• From three months. The forecast is kept between 35 per cent and 250 per cent of your usual level, so one expensive month can no longer drag the figure along with it. Confidence "medium".
• From six months. Recent months get more weight than older ones, as long as your spending doesn't swing too wildly.
• From twelve months. The season joins in, and with a stable pattern confidence becomes "high". It never looks back further than 24 months.

There is a misunderstanding here that I caused myself, so let me be precise. The forecast doesn't read the list on the Fixed costs page. It searches your transactions themselves for what comes back: the same name, an amount that differs by 8 per cent at most, and a regular rhythm from weekly to yearly. Twice is enough; from three times the recognition is "high". Your salary may vary in amount; there only the name and the rhythm count.

So if you enter your rent as a transaction every month, you see it come back in the forecast. If you only put it on the Fixed costs page, you see it under "Upcoming" with the date it is due, but not in the forecast amount for the month. Those two lists are deliberately not added together: otherwise rent you have in both places counts twice, and counting twice is worse than not counting at all.

A pattern that stops is recognised too, cautiously: only after three quiet months that were active the year before, so a yearly summer break doesn't count as stopping. A stopped pattern still counts for a fifth, because subscriptions come back. How to get the list of fixed costs itself complete is explained in the guide on mapping out your fixed costs.

The anchor of the calculation isn't your average but your median: the middle month when you put them in order. An average gets dragged along by one month with a new washing machine; a median doesn't. A weighted average is mixed in, in which the last three months get roughly two thirds of the weight; how much that counts depends on how many months you have and how calm they are.

The trend is a straight line through your monthly totals, extended into the coming months, and capped at 15 per cent of your level per month. Without that cap, one outlier draws a line to the moon, and that is exactly what a naive forecast does.

The season is a factor per calendar month: how much more or less expensive December is for you than an ordinary month. That factor is damped according to how often you have already seen that month. One December counts for a third of its deviation, two Decembers for half. That is why the app says the season only joins in after a year: before that, there simply is no December to look at.

Every amount comes with a range, "from 380 to 520", calculated from how wildly your months differ from one another. It covers about 87 per cent of cases, deliberately not 95: a range that covers everything is so wide it no longer tells you anything.

On the Forecast page there is a block called "How close was the estimate?". The app forecasts each of the last three completed months again, using only the data from before them, and puts that next to what you actually spent. The percentage is the average deviation.

That figure is allowed to disappoint. If you have only just started, or moved house during those months, you see a low accuracy, and that is correct. A forecast without a margin of error is a guess in a smart suit.

Budgetto has been around since April 2026. On 23 September 2026 I measured, in aggregate, how people use the app. None of this can be traced back to a person, and I don't look at what anyone enters either, only at how much and when.

That table explains why the forecast stays silent for the first few weeks. Nearly two in three accounts don't get past three transactions, and for them any forecast is a number out of thin air. Some of them only signed up this week, but even among accounts more than two weeks old, fewer than half really get going. A confident number out of thin air is worse than no number.

What I learnt from it: the threshold isn't two months but ten transactions. Those who reach it stay. That is why, since September, the app first pushes you towards your first ten, with a card that keeps track of how far along you are, and only then towards the forecast. How that ties in with a budget that keeps holding up is explained in the guide on making a monthly budget.

The forecast is a calculation based on your own figures. It doesn't know that you are changing jobs next month, that energy prices are going up or that your sister is getting married. It doesn't tell you what to do. It says: if next month looks like your previous months, this is roughly what comes out. I don't promise more than that, and anyone who promises more is selling something.

I build Budgetto on my own, from Mechelen, with no revenue model and no bank connection. This guide is the explanation I would have wanted to read with every app that promised a "smart forecast" without saying what was underneath. If you see a month the calculation gets wrong, I would be glad to hear about it via the address at the bottom of the page.

Romain, maker of Budgetto. Algorithm described as it stood in the code on 16 September 2026; if the calculation changes, this text changes with it.

Create free account