My vision for what 12 million people would see in Apple Wallet on migration day
Chapter 1 redesigned the Apple Card and chapter 2 built the business case for moving 12 million Apple Card holders onto it. Neither answers the question a cardholder actually has on migration day, which is what happens to my Apple Card.
So this chapter builds the answer twice, both sides running the same five applicants through the same Migration Risk Engine, a five-factor score from 0 to 100. The Wallet screen a cardholder opens and reads in seconds, and the underwriting console a Chase reviewer would use to defend the same call. The reviewer sees the score. The cardholder never does.
What this chapter builds
Weights, thresholds, and ranges are spec decisions. The applicants are synthetic fixtures built to exercise them.
The Migration Risk Engine, weight by weight
Every applicant gets a number from 0 to 100, where higher means riskier. Five factors add into it, and where the total lands decides which of three things happens next.
Three conditions decline outright, whatever the range says. Utilization at 95% or above with a FICO under 570. A 90-day missed payment with a FICO under 580. Two 90-day missed payments with utilization at 90% or above.
The credit-washing flag is the fourth way to be declined. Credit washing is dressing up a file so it reads better than the behavior underneath, usually by disputing accurate negative entries until they fall off. The flag looks for the signature rather than the act: utilization at 85% or above, either a 90-day missed payment or two or more resolved lates, and a FICO of 700 or above. All three at once, because that combination should not occur honestly, and a single condition on its own never triggers it.
Five applicants, five different reasons
Each applicant exists to exercise a different part of the policy. Two clear the Apple Card (2028), one routes to Chase Freedom Rise, and two are declined for opposite reasons.
Synthetic applicants. Names and profiles are invented to exercise the policy, not drawn from any real portfolio. Select any of them in the console to see the score computed factor by factor.
The pair that proves the policy works
When it comes to underwriting a credit card, it is typical for the higher credit score to be the safer bet. Here the higher score is the one declined. This is the comparison I would put in front of a credit committee, because it is where the policy either holds or falls apart.
On score alone Serena lands at 40, inside the Chase Freedom Rise range, so the model was about to issue a card. But utilization sits at 86% with a 90-day delinquency behind it, which trips all three parts of the credit-washing flag at once. The flag declines the application regardless of what the arithmetic says.
Riley has a weaker credit score and a past late payment, which on a cutoff would sink the application. But utilization is low and nothing is currently delinquent, so the behavior is a recovery rather than a warning. It clears to the Apple Card (2028).
Why one number is not enoughThat pair is the whole case for scoring five factors rather than one. A cutoff at 700 approves Serena and declines Riley, which is exactly backwards, and it is a plausible account of how the outgoing issuer ended up with a portfolio it did not want. Every account that looked fine on paper and behaved badly in practice arrives through that door.
The easy halfCatching Serena is straightforward. Any flag looking for high utilization alongside a delinquency would find her, and an issuer that only wanted to stop obvious manipulation could stop there.
The harder halfNot catching Riley is the difficult part. A flag that catches manipulation but also punishes honest recovery is worse than no flag, because the person paying down debt after a rough year is exactly the customer worth keeping, and declining them is how a portfolio loses the people who would have become its best accounts.
The mirror riskThe opposite error is labeling distress as manipulation. Marcus trips two of the three conditions, 96% utilization and a 90-day missed payment, and only his 561 score keeps the flag from firing. A looser test firing on any two would call his file manipulation when it is simply distressed. He is declined by two hard floors instead, which is the honest reason.
The trade-offThat is what requiring all three buys, and it is the part I would defend hardest. It is also the part that costs something. A conjunctive test is deliberately easy to evade, since anyone who keeps utilization at 84% clears it, and I would rather miss a manipulated file than decline a recovering one. That is a policy choice rather than a technical limit, and it is the choice I would want a credit committee to argue with me about.
The reviewer console
This is what a credit reviewer at Chase sees, not what a support agent sees. Pick an applicant and the engine shows the score, where it lands, and which factors drove it. A frontline agent would see the outcome and the written reason instead, because the score is internal and putting a number in front of a cardholder turns the conversation into an argument about the number rather than the behavior behind it.
Nothing here is hardcoded. Change the applicant and the meter, the outcome, the flag, and the factor breakdown all recompute, which is what lets a reviewer defend a specific call rather than a category.
The same decision, from the cardholder side
Three outcomes produce three different screens, and the middle one is the hardest to get right. An approval is easy to write and a decline has a legal template. A counter-offer has to land as an offer rather than a consolation.
The decision arrives where the card already is. A cardholder never opens an application, never sees a score, and never leaves Wallet: the offer, the terms, and the choice all render on the card they are already carrying.
That constrains the design in a useful way. There is no room for a page of explanation, so every screen has to name the issuer, state the terms, and give one clear action. The decline gets the same treatment as the approval rather than a quieter version of it.
Screen behavior described here is my own proposed flow, not an Apple or Chase design.
The credit line and the full terms, then acceptance adds the card to Wallet and Safari AutoFill and sets it as the preferred card at Apple. The whole thing happens in one screen.
Designing the decline
This is the outcome a cardholder experiences as a judgment about them, so it gets the most design attention rather than the least. Three decisions make it defensible.
Federal fair lending law requires a lender to state the specific reason, not a general one. Scoring five factors is what makes a specific reason possible to give, and the same factors say what would have to change to qualify.
The reasons are ranked by how much each one contributed for that applicant, so the notice is true for them rather than true in general.
The account winds down rather than switching off, and the notice says how to reapply. Reapplying runs the same three outcomes again, so someone who improves usually lands on Chase Freedom Rise rather than straight onto the Apple Card. A decline today is not a permanent judgment. These are still future customers, and consumer finance is the one place Apple cannot afford a bad story, so a migration that reads as mass account closures writes that story for it.
Why ranked reasons matterRanking by contribution has a side effect worth naming. Once the reasons are ordered by what actually drove the score, the policy becomes auditable. Anyone can ask why this applicant was declined and get an answer that traces back to the weights rather than to a reviewer's judgment.
What a decline actually costsClosing an account does not just take away a card, it ends the credit history attached to it, and that history is not something another issuer hands back. Which is exactly why the route back matters. The door that stays open is the one at the bank that already knows them.
Which is the whole argument in one line. In consumer lending, the small details show up in how you say no.
Apple Card Savings, and why it is gated
Why savings matters hereSavings is the part of the product that turns a card into a banking relationship, and it is the reason the issuer cares about this migration beyond interchange. The proposal gates it. The elevated rate is offered to cardholders who route direct deposit to the account, and the standard rate applies otherwise.
A cardholder who routes their paycheck to the account earns 3.50% APY. Everyone else earns 0.50%. The issuer earns slightly more than that deploying the balances, so the gated rate leaves a thin spread. The gate is not a penalty, it is a price on the thing the issuer actually wants, which is being the bank the paycheck lands in.
The business case prices this line, and the net across three years is almost nothing as a spread. The value is not the spread. It is a low-cost deposit base that does not leave, and a shot at primary-bank status.
Why I would gate itSavings is the quietest piece of this migration and possibly the most valuable. A card relationship ends when someone stops carrying the card. A deposit relationship where the paycheck lands is much harder to leave, and it is the only part of this product that makes the issuer someone's bank rather than someone's card. I would gate the rate rather than pay it to everyone, because paying a premium APY to balances that sit for a quarter and leave is the worst version of this line. The gate makes the rate a purchase rather than a giveaway.
The limits of this chapter
The applicants are invented. Five fixtures built to exercise five paths through the policy is not a portfolio. It demonstrates that the logic behaves as specified, and says nothing about how often each path fires in a real population. The distribution is chapter 2's problem, and it is modeled there rather than observed.
The score is not the only gate. Chase is widely reported to decline applicants who have opened five or more cards across all issuers in the past 24 months, a rule it has never published. The console does not model it, because it is a portfolio-level screen rather than a credit judgment, and it would decline some applicants this engine approves. Anywhere a decision is an application rather than a migration, that screen sits above everything here.
The weights are asserted, not fitted. Nothing here calibrates the five factors against realized default. They encode a policy view, that behavior should outweigh score, rather than an estimate of what predicts loss. A real deployment would fit them and would probably move them.
I did not test the decline language. The spec says a written reason is required and ranks the causes. It does not say whether the wording a person receives actually reads as fair, and I have not put it in front of anyone to find out. That is research, and it is missing.
The savings gate is unpriced. Section 06 proposes gating the elevated rate on direct deposit, and nothing here estimates how many cardholders would route a paycheck to it. That take-up rate decides whether the gate is a real acquisition tool or a rate most people never see, and it is the first thing I would instrument alongside the decline language.
A screen is not an implementation. Everything here is a specification rendered as an interface. It does not touch a bureau, run against real files, or handle the cases a live system spends most of its time on: thin files, frozen reports, disputed items, and identity mismatches.
The terms on this page
Every term of art the engine and the screens use, defined once.
What is published, and what is mine
Independent professional-development project. The Migration Risk Engine, the applicant fixtures, both interfaces, and the savings gate are a proposed concept, not an announced plan, and this is an Apple-inspired concept rather than an official Apple product. Not affiliated with, authorized by, or endorsed by Apple, JPMorgan Chase, Goldman Sachs, Mastercard, or Visa. The public facts relied on are listed in the sources. Every applicant shown is invented, and the credit lines, rates, and offer terms on the screens are model outputs rather than quoted products. None of this is legal, financial, or investment advice.