Infrastructure
Card-network integration would need 2–3 years and policy changes neither side controlled.
Featured project · 2022·7 min read·Apr–Aug 2022
Pay with Your Biometrics
Bringing biometric payments in-house at KFTC, Korea’s financial-clearing institute, without breaking employee trust or PIPA, Korea’s privacy law.
Try this · the friction
Every dot below is one employee turning back at the cafeteria because they forgot their card.
TL;DR
Problem
- 30 forgotten-card incidents a day
- 15 min lost each time
- 7.5 hours of org-wide productivity gone, every single day
Strategy
- Rejected card-network integration (24–36 mo)
- Chose wage deduction (1 mo)
- Separated DBs, required verification on every charge, built consent in from day one
Impact
- 1,270+ hrs productivity recovered annually
- 0 false charges
- 99.9% authentication accuracy
- Featured in YTN Science
Problem
01. People left their physical credit cards
when going to the cafeteria.
Employees often forgot credit cards heading to the in-office robotic cafeteria. They were used to frictionless authentication everywhere else (mobile pay, face unlock, airport check-in), but the cafeteria still demanded a physical card.
- •~30 incidents/day
- •~15 minutes lost per incident
- •~7.5 hours of org-wide productivity, daily
Before · the typical incident
- 1
Heads to the cafeteria
normal trip
- !
No card on hand
+5 min realize & queue out
- 2
Walks back to the desk
+5 min
- 3
Returns to the cafeteria, with card
+5 min
Wasted (steps !, 2, 3)15min× ~30 incidents/day = 7.5 hrs lost org-wide, daily
Opportunity
02. Biometric authentication removes the card entirely.
- •Always available, no card or phone
- •Familiar already; mobile, banking kiosks, airports
- •Right-sized for frequent, low-value internal payments
Challenge
03. Deploy biometrics for payments
without losing trust on the first false charge.
Trust
One false charge could permanently break confidence in the whole system.
Privacy
Biometric data demanded Korean PIPA compliance, explicit consent, and tight governance.
Key decision
04. Skip the card network entirely.
Instead of waiting on card-network integration, we deducted the cafeteria charge directly from each employee's wages, only after biometric verification at the kiosk.
This was only possible inside an in-house system. But it delivered the same user experience (cardless payment, one tap, one second) in a fraction of the time.
ApproachProsConsDecision
Biometric + Card integration
industry standard path
- Broad acceptance
- Existing infrastructure
- 2 to 3 year timeline
- External policy lock-in
- High cost
✗Rejected
Biometric + Wage deduction
in-house path
- 1-month delivery
- No external dependency
- Forced privacy isolation
- Only works in-house
✓Selected
How I built · trust-by-design
05. Four guardrails.
Separated the biometric DB from the payroll DB.
Only the employee ID crosses between systems. A breach in one couldn't compromise the other.
How the data is isolated
on-premise · encryptedBiometric DB
- Palm vein template
- Face template
Employee ID onlyno biometric leaves this roomexisting HR systemPayroll DB
- Employee ID
- Wages, role
A breach in one couldn't compromise the other.
Required an explicit verification screen before every charge.
“You are [Name], purchasing [Item] for [Amount].” False payments became structurally hard, not just statistically rare.
Established opt-in-only enrollment with 30-day deletion for inactive users.
User agency was built into the data lifecycle itself, not just the onboarding flow.
Implemented a False Acceptance Rate threshold tuned for payment security.
Accuracy was budgeted by the trust requirement, not by convenience. The bar is higher when money moves.
The user flow · where the architecture becomes visible
- 1
Registration
Employees enroll palm vein + face biometric data at a dedicated booth (one-time, ~3 min).
- 2
Authentication
Scan biometric (palm vein OR face) at the kiosk.
- 3
Verification
Screen shows name + purchase + amount.
- 4
Payment
Purchase amount automatically deducted from monthly payroll.
- 5
Confirmation
Receipt shows the transaction. No additional steps required.
Impacts
06. What it moved.
1,270 hours
Annual productivity recovered, org-wide15 min × ~20 fewer incidents/day × 251 working days
0
False charges in year 1
at 99.9% authentication accuracy
2–3yr→1mo
Development timeline reduction
card-network integration vs. wage-deduction path
99.9 percent
Authentication accuracy
at an industry-compliant False Acceptance Rate
What I’d do differently
Bring security into phase one, not into the review.
Security got thought through, but in later steps. Sitting down with security operations at the start would have been the more efficient path: data storage methods, encryption, and data mapping were all decisions we could have settled early, when they were still cheap to change. Treating security as something to satisfy at review rather than design from means you discover the constraint after you have already built around its absence.
Key takeaway
Responsible tech isn't about avoiding constraints. It's about reframing them.
The card-network timeline forced a wage-deduction architecture that turned out to be safer and more efficient than the original plan. The on-premise privacy requirement, originally a regulatory cost, became the trust differentiator itself.
Keep exploring
You just viewed: Pay with Your Biometrics