Payments risk · Product strategy · Interactive system
Digital Goods Merchant Risk Strategy Lab
A payment-platform framework for controlling fraud and contingent exposure without suppressing legitimate digital-goods growth.
I designed an employer-neutral decision system for direct web payments across short drama, AI services, web fiction and games. It connects merchant behavior, customer obligations and control economics to explainable risk actions.
Strategy demonstrator · Aggregated and synthetic data · No confidential platform or merchant data
Illustrative results from aggregated synthetic scenarios; not observed merchant performance or industry benchmarks.
- 8
- synthetic merchant scenarios
- 32
- period-specific assessments
- 4
- controlled mechanism experiments
- 56
- automated tests
Illustrative outputs demonstrate decision logic, not predictive accuracy.
01 · Decision problem
Digital-goods risk is interaction-driven.
Small tickets do not necessarily mean small platform exposure. A merchant can settle funds while customers still hold unused coins or credits; value can be consumed before a dispute arrives; and content, account-security or service-continuity failures can interrupt fulfillment without first appearing in payment-loss rates.
The strategy therefore evaluates what can go wrong, what protection already exists and which control keeps residual exposure within appetite at the lowest sustainable cost to the merchant and platform.
Use the least-restrictive control that keeps residual risk within appetite.
02 · Decision architecture
From merchant evidence to an explainable action.
Establish the boundary
Check legitimacy, sanctions and prohibited-activity constraints before commercial optimization.
Diagnose risk mechanisms
Separate payment conduct, credential abuse, contingent obligation, fulfillment, content continuity and evidence uncertainty.
Measure protection independently
Evaluate tenure, service reliability, support, liquidity and evidence quality without allowing strength to erase risk.
Compare control economics
Size protection across 30-, 60- and 90-day horizons, then compare mechanism-specific controls, operating cost and merchant liquidity burden.
Return an accountable decision
Provide controls, reserve treatment, reasons, release conditions and escalation triggers.
The system recommends balanced growth when it is within risk appetite and economically near-equivalent to a more restrictive eligible posture.
03 · Category applications
One framework, two distinct exposure structures.
Short-drama coin systems
Short drama compresses acquisition, coin purchase, episode unlock, consumption and dispute into a short interval. The key platform question is whether unused purchased value, post-consumption disputes and viral cross-border growth jointly create losses the merchant cannot absorb after settlement.
- Purchased-coin obligation and ledger reconciliation
- Post-consumption dispute evidence
- Monetization and renewal clarity
- Content-rights and continuity interruption
- Viral growth before cohort quality is observable
Track purchased and promotional coins separately, retain episode-level fulfillment evidence and link reserves to the uncovered obligation gap—not to category labels alone.
AI subscriptions and credits
AI services combine subscription conduct, prepaid credits, usage metering, account or API-key compromise, service reliability and variable compute exposure. The correct response depends on whether risk is customer-wide, account-specific or linked to an outstanding credit obligation.
- Trial, renewal and cancellation conduct
- Purchased, consumed, refunded and unused credits
- Usage-meter reconciliation and bill shock
- Account and API-key compromise
- Provider dependency, reliability and postpaid exposure
Contain localized key or account abuse before restricting the entire merchant, and connect reserve or processing controls to measurable release conditions.
04 · Scenario evidence
Stress the mechanism, not the category label.
| Scenario | Synthetic change | Decision result | Strategic lesson |
|---|---|---|---|
| Viral cross-border short drama | Volume, cross-border share and outstanding obligations rise faster than evidence | Mechanism-specific controls and exposure-based reserve treatment | Growth changes the observation burden; geography alone is not treated as misconduct. |
| Post-consumption dispute deterioration | Complaints, disputes, integrity concerns and unused value worsen together | Manual review with horizon-based protection and explicit counterfactuals | Combined integrity and complaint deterioration can bind before the aggregate score does. |
| AI account/API-key abuse | Fraud, anomalous usage and disputes concentrate during an abuse event | Targeted account, authentication and usage-reconciliation controls | Target the abuse mechanism, then relax controls after verified remediation. |
Dollar results use a configurable synthetic merchant with $1 million in monthly attempted payment volume. They are illustrative, not benchmarks.
05 · Interactive lab
Explore how the decision changes.
Select a synthetic merchant and observation period, then inspect the decision, 30/60/90-day exposure decomposition, mechanism-specific controls, economics, binding constraint, counterfactuals and sensitivity results. Controlled experiments show whether the system responds to the intended mechanism.
Connecting…
Demonstration data are aggregated and synthetic. Thresholds and policy effects are illustrative.
Representative synthetic result — Viral cross-border short-drama growth (Stress)
Decision
MANUAL REVIEW
Risk exposure
32.4
Merchant strength
58.9
Commercial value
93.3
Recommended controls
- enhanced monitoring
- rolling reserve
- progressive processing limit
- manual underwriting review
Illustrative reserve
10% · $357.2K
- Illustrative $ approved
- $3,572,000
- Per $100 reserved
- $9.40
- Recommended posture
- balanced growth
Demonstration data are aggregated and synthetic. Thresholds and policy effects are illustrative.
06 · Findings
Five conclusions from the system.
Interaction effects can be more decision-relevant than isolated thresholds.
Unused customer-funded value is contingent exposure, not automatically an expected loss.
Merchant strength should reduce uncertainty without erasing observed risk.
Restrictive controls should be mechanism-specific, temporary and tied to release conditions.
The strongest control is not automatically the best strategy once approval opportunity and merchant liquidity are considered.
07 · Build
Research, product judgment and technical execution in one workflow.
Risk and product strategy
- MECE research architecture
- Risk taxonomy
- Indicator prioritization
- Risk appetite
- Control design
- Reserve logic
- Release conditions
Model and analysis
- Synthetic scenario construction
- Explainable rules
- Commercial simulation
- Sensitivity logic
- Assumption governance
- Limitations and model-risk thinking
Engineering
- Python
- Pydantic
- FastAPI
- REST APIs
- Versioned YAML
- HTML/CSS/JavaScript
- Automated unit and contract testing
- Vercel serverless deployment
- Frontend/API integration
I defined the product question, research structure, scenarios, decision policy, commercial simulation, API, test suite, interactive interface and case-study narrative.
08 · Methodology
Transparent assumptions, bounded claims.
- No confidential merchant, customer or payment-platform data is used.
- Thresholds, policy effects and dollar outcomes are synthetic.
- The system demonstrates methodology and judgment; it does not claim predictive accuracy.
- Production use would require internal calibration, legal and privacy review, fairness testing, model-risk governance and controlled experimentation.