Cover
Writing the Public-to-Secure Banking Journey
“The best banking copy sounds like a calm, competent person explaining your account, not a legal disclaimer reading itself aloud.”
A content design case study covering a public product page, a secure account dashboard, a security alert flow, multi-channel notifications, content governance, and a data-informed recommendation.
Problem
Strategy, not a screenBanking content has two jobs at once, and most of it only does one.
Every sentence in a banking product has to calm someone who is anxious about their money, and survive a compliance review. Those jobs pull in opposite directions, and the second one has a reviewer attached to it. So the copy defaults to the second job: legal language reading itself aloud, technically unimpeachable, and no help at all to the person holding the phone at 2:41 in the afternoon wondering why their card stopped working.
I do not think that trade is real. The precision compliance asks for and the calm the customer needs are the same discipline pointed at different audiences, and this case study is my argument for writing them as one sentence instead of two.
Problems like this show up as avoidable support calls, most often the ones asking a question the interface should have already answered.
No figure in this case study is real. Fernwell is a fictional institution, and the numbers exist to show how I would size and test the argument, not to claim a result.
Research, grounded in published findings
Four sources, four decisionsGrounded in published research, not just instinct
Every major content decision below traces back to a real, citable finding, not a guess.
Financial worry itself measurably narrows a person’s available cognitive bandwidth, independent of the person. Not a trait of anyone in particular, a temporary state anyone under financial stress experiences.
The security-moment copy stays short, one idea per sentence, no jargon, because the reader has less cognitive bandwidth available in that exact moment.
Effective error messages state the problem specifically, appear near its source, avoid blaming language like “invalid” or “illegal,” and give a constructive next step.
The card-lock message’s structure: what happened, what we already did, what to do next, in that order, with no blaming language.
Channel choice should match urgency and permanence: SMS for time-sensitive reference-worthy alerts, push for lower-urgency app-directing updates, email for detailed, less time-critical information.
The one-event-three-channels notification strategy, where each channel gets a different compression of the same fact, not a copy-pasted message.
The CFPB’s own consumer testing on its plain-language redesign of mortgage disclosures found a 29% gain in comprehension, up to 42% improvement in comparing loan offers, and up to 28% improvement in catching cost discrepancies, versus the old jargon-heavy forms.
The case for plain language across Fernwell overall. This is regulatory-grade evidence that plain language isn’t just friendlier, it measurably performs better.
Research plan, what I’d validate next
Future work, explicitly not findingsWhat I’d test with real users next
This is a research plan, not completed findings. Fernwell is a self-directed concept project without real users yet, this is exactly what I’d run first if it became a live product.
“Does naming what the system already did, before asking for confirmation, actually reduce anxiety in the moment, or does it read as presumptuous?”
5 to 6 moderated usability sessions on the card-lock flow prototype, think-aloud protocol, with a before and after self-reported anxiety scale.
“Which notification channel do people actually check first when they suspect fraud on their own account?”
A short survey or diary study with people who have experienced a real card decline or hold, asking what they saw first and what they trusted most.
“Does the AI agent permission pattern build real trust, or does the extra detail just add friction?”
An A/B concept test comparing the specific-scope permission screen against a simpler generic on/off toggle, measuring comprehension and stated trust in each.
Persona
A believable composite, not a real individual“Just tell me what happened and what to do. I can handle the rest.”
Know immediately if something is wrong with her money, without having to decode what happened.
Her income arrives in lumps, so a held charge and a missing payment feel identical until something tells her which one it is.Past experience with a card decline that gave no reason and no next step. She had to call support to find out it was a routine fraud hold.
Twelve minutes on hold to be told nothing was wrong. The call was not the failure, the silence before it was.User journey, before Fernwell
The card-lock scenario as it happens todayThe system knew what happened before the customer did, and did not say so.
Every fact the support agent eventually read aloud existed at the moment of the decline. The journey's low point is not a technical failure, it is a writing decision.Solution
The three principles artboard 13 turns into component rulesThe same discipline, aimed at what the customer needs to know first.
The best banking copy sounds like a calm, competent person explaining your account, not a legal disclaimer reading itself aloud.
User flow
Structure only · each box names the artboard that shows itTHIS
YOU?
Public product page
Unauthenticated · assumes nothing · has to earn the click
“Unlock a world of limitless possibilities with rewards that work as hard as you do.”
“3% back on groceries and transit. 1% on everything else. Paid as a statement credit each month.”
Rationale: the cut line promises something we can’t price or prove. The replacement is the same length and lets a shopper compare us to the card already in their wallet.
Secure account dashboard
Authenticated · already convinced · here to do one thingExplains the product from zero. Defines terms (“statement credit”), sets expectations about credit checks, names the person it isn’t for. Sentences run longer because they’re carrying an argument.
“3% back at grocery stores and on transit, including buses, trains and rideshare.”
Assumes the product is understood and answers the two questions people sign in for: what do I owe, and what’s that charge. Labels shrink to nouns; the copy that stays is the copy that resolves doubt.
“Credited to your statement on Sept 28. Nothing to redeem.”
Same voice, different job. The public page persuades; the dashboard confirms. The only place the secure side gets wordier is uncertainty, the pending hold gets a full explanation because an unexplained $340 is what generates a phone call.
Security moment: card locked
Interactive, choose a response“Yes, I made the $340 charge at Vermeer Home Supply, unlock my card”
Why: screen reader users often tab a form’s controls out of context, and “Yes, that was me” read alone answers nothing. The label restates the amount, the merchant and the consequence, so the button is unambiguous even if the heading was never heard. The visible label stays short because sighted users already have the charge on screen.
Rule: an aria-label may only add context, never contradict or replace the visible words, the visible label stays the first words of the accessible name.
“Your card has been suspended due to suspicious activity.”
Three problems in nine words: “suspended” sounds punitive and permanent, “suspicious activity” names no activity, and the passive voice hides that we did this and we can undo it. The shipped version gives the amount, the place, the reassurance, and one action, in that order, because that’s the order people ask.
Notification strategy
One event from artboard 11, four channelsCompression: 41 characters of title, one sentence of body. Amount and merchant survive; the “we’ll keep it paused” clause is cut because the tap resolves it.
Compression: the only channel that must carry its own anti-phishing line, because SMS is the channel scammers imitate. Costs characters, keeps trust.
Compression: subject states the fact, preview text does the reassurance, they’re read as one line in the inbox, so they can’t repeat each other. Arrives last on purpose: it’s the searchable receipt.
Compression: the only place the full flow can finish, so the copy is shortest, the screen behind it already says where you are. Persists until resolved; the others don’t.
Push and the in-app banner carry the call to action; SMS carries a fallback link for the person whose notifications are off; email informs and becomes the record. One fact, a $340 charge in an unusual city, held constant across all four. What changes is how much context each channel can afford, and only one sentence is identical everywhere: your card is safe. That’s the line the whole strategy exists to deliver.
Content governance in the design system
Rules live on the component, not in a separate guide nobody opens“Your transaction could not be processed at this time. Please try again later or contact customer service.”
“This payment didn’t go through, your bank declined it. Try a different account, or pay a smaller amount now and the rest by Oct 16.”
Form error, v1.8: say what’s wrong and how to fix it in one sentence. “At this time” and “later” are the two words the rule exists to delete, they postpone without informing. The rewrite names the decider (the bank, not us) and gives two real exits.
Data-informed recommendation
ILLUSTRATIVE, NOT REAL DATAIf we replace “suspended due to suspicious activity” with copy that names the amount, the merchant city, and states the card is safe, then inbound calls per 1,000 card locks will fall, because the most common caller question is one the message can answer before the phone is picked up.
All three are questions about state, not fraud. Nobody called to ask what fraud is; they called because the message described a policy instead of their card.
“Your card has been suspended due to suspicious activity.”
“We paused your card after a $340 charge in a city you don’t usually shop in. Your card is safe. Confirm it was you, or we’ll keep it paused.”
Sentence two exists solely to answer the 14%. It earns its place in the character budget because it deletes a phone call.
Figures above are invented for this case study. No real Fernwell data exists, Fernwell doesn’t.
Accessibility, WCAG 2.1 AA
Artboard 11, annotated · click a step to trace focus orderWe paused your card after a $340 charge in a city you don’t usually shop in.
Your card is safe. Confirm it was you, or we’ll keep it paused.
“Alert dialog. Card paused. We paused your card after a $340 charge in a city you don’t usually shop in. Your card is safe. Confirm it was you, or we’ll keep it paused. Two buttons.”
The status word “Card paused” is read before the sentence, so the outcome lands before the explanation. “Two buttons” tells the listener a decision is waiting, the reason the container, not a button, takes focus.
Paused state is carried by three signals at once: a lock icon, the words “Card paused”, and a bordered container. Remove the amber entirely and the screen still says what it means, which is the test (WCAG 1.4.1). Motion is limited to a 2-second fade on the alert and is suppressed under prefers-reduced-motion.
AI agent permissions and the audit trail
Interactive, pause a permission and watch the trail record it“Allow Fernwell Assistant to manage my payments.”
“Pay my statement in full on the due date, from Checking ••3092, up to $2,500 per payment.”
Rationale: a customer cannot consent to a range they cannot picture. “Manage my payments” covers a $12 charge and a $12,000 one. The shipped scope is one action, one account, one ceiling, which is also what makes the audit trail writable in plain past tense.
Any AI action taken on a customer's behalf gets this pattern: a specific, legible scope, a plain-language audit trail, and an equally easy pause action, never a vague on or off toggle.
The thumb slides in 200ms and the track cross-fades over the same interval, so a permission change looks like a mechanism moving rather than a screen redrawing. New trail rows fade and rise 8px, staggered 70ms, which is the only motion in the case study triggered by a write: if something moved, something was recorded.