Round-ups count your transactions. They don't measure your spending.

A round-up is whatever it takes to reach the next dollar — so what you invest depends on how many times you tap a card, not on what you tapped it for. Buy ten small things and you invest ten times. Buy one large thing and you invest once.

The round-up column is identical. Both end in fifty cents, so the dollars are discarded before the calculation starts. Every percentage column scales by 66×, matching the 66× price difference. At the low end round-ups win a cent or two — the honest trade for a mechanism that responds at all.

Both products have a range. Only one is in your control.

Acorns' range is $0.01–$0.99 per transaction — fixed in dollars. Wealthstack's is 1–20% — fixed in proportion. Put both in the same unit and the difference stops being a preference:

A 6,500× swing in effective rate, with no way to choose where you land — the cents decide. The percentage band doesn't move, and picking it is the one decision the product asks for.

The ceiling isn't a policy. It's arithmetic.

Every round-up falls between one cent and ninety-nine, so a month's total is really transaction count × fifty cents. Acorns publishes the benchmark: an average of $45 a month, roughly 90 transactions. Count barely changes with income — earning more doesn't make you buy more things, it makes you buy costlier ones. The mechanism goes quiet exactly as you start to succeed.

Hold those 90 transactions constant, vary only what was spent, and put both ceilings side by side:

The Acorns column never moves. Five times the spending, same $45, and no setting raises it. Wealthstack has a ceiling too, at $200 — the difference is that it's yours: raise it to $1,000, or turn it off.

$3,045 is my own spending, not an average. The Acorns column holds transactions at 90 and the multiplier at 1×, because those are the two things that do move it — neither of which is what you spent. The cap only binds in an unusually heavy month — which is the point of putting it just above a normal one.

Answering the multiplier

Acorns lets you multiply round-ups by 2×, 3× or 10×, which looks like the answer to all of this. It isn't — a multiplier scales a number that was never connected to the purchase, so it multiplies the disconnection along with the amount.

At 10×, a coffee costs three times its own price and a grocery run costs five percent. No multiplier fixes this, because a multiplier is a constant and a percentage is a ratio. They agree at exactly one purchase size — 10× equals 5% at $100, 3× at $30 — and miss everywhere else.

Which leaves a setting nobody can reason about. Choosing a multiplier well means knowing your own average transaction size. Choosing a percentage means answering one question: how much of what I spend do I want invested?

The person who won't start, and the person who outgrew the app that started them.

The graduate's complaint wasn't the interface. Round-ups pay out in transaction count, so the only way to invest more is to buy more things — a mechanism asking you to shop more in order to save more. He hadn't outgrown Acorns' design. He'd outgrown its arithmetic. And he stopped using it. That's churn, not a feature request — the mechanism didn't annoy him into leaving, it just stopped being worth keeping. This is the finding that set the top of the range in 05.

Both quotes are verbatim from sessions, not composites.

One number. It just happens to be the right one.

1–20% of every purchase, 5% by default, set on a slider that shows what the rate does to a real purchase as you drag it. A cap sits behind it at $200, adjustable from $50 to $1,000. Two controls at first run; the rest live in Settings.

The defaults came from asking first

Eight participants informed this section — the same cohort whose sessions are detailed in 06. The numbers below came from asking them before the rate screen existed; the findings in 06 came from watching them use it afterwards.

I asked what they spend in a normal month and checked it against published figures. Both landed around $3,000–$3,500. At 5% that's $150–175 a month against the $45 round-ups deliver — and the cap sits at $200, just above a normal month, so it catches an unusual one without binding on an ordinary one.

I also asked what they'd actually accumulated on a round-up app. Most couldn't recall a number, only that it was low. Weak corroboration rather than a finding, but it pointed the same way as the published figure.

Set the default to be felt, and let it move both ways. 5% is the amount that makes the mechanism worth having. Anyone who finds it steep can step down to 1%; anyone who wants it working harder can go to 20%. What matters is where the dial starts — defaulting low and asking people to opt into meaning is how round-ups end up at $45 in the first place.

Why the band runs to 20%

5% and a $200 cap are the defaults; the band runs to 20% and $1,000 because of the graduate in 04. His problem wasn't that round-ups were badly designed — it was that the mechanism had a ceiling he'd passed, so it quietly stopped mattering.

A 10% ceiling would have reproduced exactly that. A range that stops responding once you're doing well is the failure this product exists to fix, so the ceiling sits at 20% and the cap at $1,000 — not because most people will go there, but because the range shouldn't be the thing that runs out.

The bet, and what could break it

5% is $150–175 a month leaving checking, against $45. That's the whole proposition and its biggest risk: $45 may be less a limitation of round-ups than the reason people tolerate them. The invisibility is doing work.

The open question isn't whether the math is better — it plainly is — but whether people hold a rate once they can feel it. That's the first thing I'd test with money actually moving.

A stack isn't a transfer

Your spending earmarks an amount and the app tallies it; nothing leaves your bank until a sweep. Nothing is held, so this sits outside money transmission — and verification can wait until money moves. A tally can outrun the balance behind it: stack $37, spend down to $12, sweep fails. The app takes what's there and says so. It never overdrafts.

Four changes, and none of them were free.

Going back through the built flow, I marked every screen decided or accepted. The onboarding order was accepted — I'd never chosen it.

I built a product about restraint and couldn't apply it to my own scope.

Sector tilting is the clearest miss — a control I'd criticised the category for, shipped because building it was more fun than deciding against it. A feature earns its place if it changes what the product is, not what it can do.

Solo has an obvious cost. Nothing here survived an engineer's estimate, a compliance review, or a stakeholder with a different incentive — only the infrastructure and the users told me no. What that taught me is which constraints don't negotiate: KYC requirements set the screen count, and money transmission rules are why a stack is a tally rather than a transfer. What I haven't done is defend a decision to someone who outranks me and isn't wrong.

What's next

Test the split onboarding against the old order with money actually moving. Instrument the carousel to find out whether anyone reaches the second slide. And find out what people think 5% of their spending costs them — right now, neither they nor I know.