Economic Model
SentryFlow's economic case rests on three levers: lower infrastructure cost-per-transaction, higher fraud catch rate, and faster policy response to emerging attack patterns.
Comparative unit economics
Note: catch rate and FPR figures below are modeled targets. Observed values on synthetic simulation data are reported by the training pipeline on each make train run. Do not cite these as production guarantees.
| Driver | Legacy Vendor Stack | SentryFlow (Integrated) |
|---|---|---|
| API cost (per tx) | $0.45 (3–4 vendors) | $0.12 (infrastructure) |
| Fraud catch rate | 72% | 89% (model target) |
| False Positive Rate | 4.2% | 1.8% (model target) |
| Policy deployment | 14–21 days | < 10 minutes |
| Engineering overhead | 2 FTEs (hard-coded rules) | 0.5 TPM (self-serve) |
3-year TCO model
Assumes $1B annual transaction volume, 2% baseline fraud attempt rate ($20M at risk).
Legacy scenario
| Cost driver | Annual |
|---|---|
| Direct fraud loss (28% slippage) | $5.6M |
| API fees (1M tx × $0.45) | $450k |
| Good-user churn from false positives | $4.2M |
| Engineering overhead | $400k |
| Total | ~$10.7M |
SentryFlow scenario
| Cost driver | Annual |
|---|---|
| Direct fraud loss (11% slippage) | $2.2M |
| Infrastructure COGS (1M tx × $0.12) | $120k |
| Good-user churn (1.8% FPR) | $1.8M |
| Operations (0.5 TPM) | $100k |
| Total | ~$4.2M |
Net delta: ~$6.5M saved annually (varies with actual catch rate and transaction volume).
Velocity of protection
The economic case extends beyond static costs. When a new social engineering script goes viral:
- Legacy: Engineering ships a hard-coded fix in 14 days. Exposure window: ~$1.2M in losses.
- SentryFlow: Risk Manager identifies the pattern via SHAP feature attribution, authors a JsonLogic rule in the dashboard, submits for 4-eyes review, deploys in under 10 minutes. Exposure window: <$50k.
Policy agility is an asymmetric advantage — each hour of response time advantage compounds during a viral fraud wave.
Infrastructure cost model
SentryFlow's $0.12/tx COGS reflects self-hosted infrastructure:
- FastAPI on a single instance handles ~500 req/s at < 30ms p99
- Redis for DIBB signal caching
- Metaflow for batch training (runs weekly or on-demand)
The marginal cost per transaction is compute (not vendor API fees), making the unit economics improve at scale rather than degrade.