Put a $3,000 bonus into a savings app and watch what the projected date does. In a lot of them it jumps forward by half a year.
That is a modelling failure dressed up as good news, and the user has no way to tell.
The Problem With Trusting the Last Deposit
Projecting a completion date is arithmetic: take what is left, divide by how much arrives per period. The hard part is the second term, because “how much arrives per period” is not a number you have. It is a guess about the future built from the past, and the whole question is how much past to use.
Use only the most recent deposit and the projection becomes extremely responsive and nearly meaningless. One bonus month implies you will produce a bonus every month. One lean December implies permanent poverty. The date swings hundreds of days on data that has not actually told you anything new about the person’s capacity to save.
Responsiveness feels like quality in software. Here it is the opposite. The user is not asking “what if this month repeated forever”, they are asking “when do I get there”, and those have different answers.
There is a behavioural cost on top of the statistical one. A date that lurches forward after a windfall quietly teaches people that windfalls are the strategy, when the thing that actually determines the outcome is the boring monthly contribution they can sustain. The app has rewarded the wrong behaviour with its most prominent number.
Smooth It, and Say That You Did
The fix is not clever. Average over a window instead of trusting the last point.
Hill Wallet, a planner for first-time home buyers, projects its date from the average of the last three months of logged deposits. A single large deposit moves the projection, because it should, but it moves it by about a third of what a naive model would do, and it keeps moving it for the next two months as the window rolls. Save less than your pace and the date drifts later. Save more consistently and it pulls in. The number tracks your behaviour rather than your most recent transaction.
Three months is a judgment call rather than a law. Shorter and you are back to noise, longer and the projection stops responding to a real change in circumstances, which is its own failure when somebody’s income genuinely drops.
The other half of the fix is disclosure. The product says the date is an estimate rather than a promise, that real prices and timelines will differ, and that it is not a bank, lender or advisor and gives no lending advice. Products that project dates should be loud about this, because a projection presented without its uncertainty gets read as a commitment.
What Else Smoothing Buys You
Once the projection is stable it can carry something more useful: a pace check. Not “you are behind” in the abstract, but the specific monthly figure that holds your current date.
That only works if the date was not jumping around to begin with. You cannot tell somebody what pace holds a target that moves every time they open the app.
Stability is the feature. The smoothing exists to make the date boring enough to act on.
Boring Numbers Get Believed
The temptation in this category is to make the number move, because movement looks like progress and progress drives engagement.
A savings date is one of the few numbers where the right design goal is for it to sit still most of the time. If it jumps whenever something nice happens, people stop reading it as a forecast and start reading it as a mood.
Then it is just decoration, and the arithmetic underneath may as well not be there.
