Why we only forecast 30 days

The most common feature request we get is a longer projection. Sixty days. Ninety. A year. "Show me where I'll be at retirement." It's a reasonable thing to want, and we've said no to it four times now.

Here's the argument.

Forecasts decay fast

A projection is only as good as the events feeding it. On day one, we know almost everything: your rent is on the 1st, your salary lands on the 28th, your three subscriptions are dated, your card spending follows a pattern we've seen for months. By day forty, most of what we're showing you is extrapolated from averages.

We measured this. We took twelve months of anonymized data, generated projections at every horizon, and compared them to what actually happened.

At seven days, our median error was under 2%. At thirty days, it was 6%. At sixty days it jumped to 19%, and at ninety it was 34% — which is another way of saying the number is decorative.

The cliff between thirty and sixty days isn't gradual. It's where known recurring events stop dominating and discretionary spending takes over. And discretionary spending is, definitionally, the part nobody can predict.

A projection you can't act on isn't a projection. It's a mood.

The horizon that matches the decision

There's a second argument, and it's the one that actually settled the debate internally.

Ask what a forecast is for. Almost every real use is a decision with a deadline: can I book this, can I delay that, do I need to move money before the 12th. Those decisions live inside a billing cycle. Nobody makes a Tuesday decision based on where they'll be in July.

Thirty days covers a full cycle of income and obligations. It's long enough to see every recurring charge at least once, and short enough that we're mostly reporting facts rather than guessing.

What we built instead

Rather than extending the line, we made the short line more useful.

We surface the lowest point, not just the ending balance. Where you end the month is far less interesting than the worst moment inside it, because that's the moment something bounces.

We show what caused each dip. Tapping any day breaks down what moves that day and by how much, so the projection is auditable rather than magical.

We flag the threshold crossing rather than the trend. If you're going below your safe floor on the 10th, you'll know on the 3rd. That's a week to move something — which is the whole point.

The version of this we might build

There's a case for a longer horizon that we haven't ruled out: a goal projection rather than a balance projection. "At this rate, your emergency fund is funded in November" is a claim we can make honestly, because it depends on rules you've set rather than spending we're guessing at.

That's the difference. We'll happily extrapolate something you control. We won't extrapolate something you don't and call it a forecast.