FINTECH
CONCEPT STUDY
CASE 01
An AI agent that pays household bills automatically. It worked, but users switched it off after the first surprise payment. We audited it against the TRUST Framework and redesigned the moments that decide trust.
AGENT
Pays household bills from a linked bank account
USERS
Busy households and freelancers
ENGAGEMENT
Trust Audit + UX Sprint
FOCUS
Supervision · Reversibility · Traceability
Concept study: a composite of patterns common in consumer fintech agents, not a client project. Scores come from our TRUST audit of the “before” state.
01 — THE PROBLEM
The agent paid the right bills most of the time. But “most of the time” isn’t good enough when it’s your money, and the interface gave users no way to tell the difference.
✕
Silent payments
The agent paid bills in the background and reported only “Task completed”.
TYPICAL USER REACTION
“Wait, which bills did it just pay?”
✕
Blind approvals
Approval prompts said “Approve payment?” with no amount, payee or reason.
TYPICAL USER REACTION
“I just tap yes. I have no idea what I’m approving.”
✕
No way back
A wrong payment meant a support ticket and a long wait for a refund.
TYPICAL USER REACTION
“One mistake and I turned it off for good.”
02 — THE AUDIT
Each action was mapped to the Risk Ladder, then scored against 40 checks across the five TRUST pillars.
TRUST SCORE · BEFORE
38
Untrustworthy
Transparency
38
Reversibility
25
User control
44
Supervision
31
Traceability
50
Read account balance
L1
Reads silently and logs it
OK
Draft a payment
L2
Drafted and sent without review
Gap
Pay a routine bill under the limit
L3
Pays, with no undo window
Gap
Pay a new payee
L4
Pays automatically, notifies after
Critical gap
Pay above the usual amount
L4
“Approve payment?” with no context
Critical gap
Change the funding account
L5
One tap, no re-authentication
Critical gap
CRITICAL FINDINGS
F1
S2 · Supervision
Approvals without context
Users approved blindly, so supervision was only a formality.
F2
R2 · Reversibility
No undo after payment
Every mistake became a support ticket and a lost user.
F3
TR2 · Traceability
Logs written for engineers
Users couldn’t check what happened, so they assumed the worst.
03 — THE REDESIGN
We redesigned the three moments that decided whether users kept the agent on.
F1
Approval cards with full context
→ Amount, payee and reason on every approval
→ Flags for new payees and unusual amounts
→ Approve, edit or skip in one tap
BEFORE
Bill Agent
Approve payment?
Yes
No
AFTER
L4 · APPROVAL NEEDED
City Electric Co.
$142.60
vs last month
+38%
Why I’m asking
Over your $100 limit
Approve
Edit
Skip
F2
An undo window after every payment
→ 30-minute undo on reversible payments
→ A clear receipt with reference and timing
→ One tap to reach a real person
BEFORE
✓ Task completed
3 payments processed
AFTER
$142.60 paid to City Electric Co.
From
Checking ••4821
Reference
CE-20931-774
Changed your mind? 28 min left
Undo
Something wrong? Talk to a person
F3
A log people can actually read
→ Every action in plain language
→ The reason, and who approved it
→ Weekly summary and export for disputes
BEFORE
txn_8f2a PAY_OK 142.60
txn_8f2b PAY_OK 59.99
txn_8f2c PAY_OK 41.20
AFTER
What your agent did
Paid City Electric Co. $142.60
Over your limit · You approved
Paid FiberNet $59.99
Routine bill · Auto-paid
Held a $24 late fee
On your “always ask” list
04 — THE RESULT
We re-audited the redesigned concept with the same 40-point checklist.
BEFORE
38
Untrustworthy
AFTER
89
Trusted
Transparency
38 → 86
Reversibility
25 → 90
User control
44 → 88
Supervision
31 → 92
Traceability
50 → 90
Before
After
WHAT WE’D MEASURE AFTER LAUNCH
01
Agent retention
Share of users who keep auto-pay on after 30 days
02
Time to approve
How quickly users decide on an approval card
03
Undo rate
How often payments are reversed, and why
04
Support tickets
Tickets per 1,000 payments about unexpected charges
05
Approval load
Share of payments that need a human decision