What Banks Track (and how they currently do it):

Deposit Composition:
├── Core deposits      (stable, relationship-driven, low rate — low "deposit beta")
├── Non-core deposits  (rate-sensitive, likely to leave — high "deposit beta")
├── Uninsured deposits (>$250K, highest flight risk)
├── Concentration risk (% from top 10 depositors)
└── Surge balances     (rate-chasers with significantly higher betas)

2023 shift: time deposits grew 42% YoY, reaching ~25% of total deposits (rate-chasing)

Key Metrics:
NIM (Net Interest Margin) = (Interest Income − Interest Expense) / Avg Earning Assets
  → Typical community bank NIM: 3.0–3.5%
  → Compressed NIM = funding cost problem = ModernFi's value prop

Liquidity Gap = Assets maturing − Liabilities maturing (per time bucket)
  → Positive gap: more assets maturing than liabilities (reinvestment risk)
  → Negative gap: more liabilities maturing (funding risk)

"Accounts at Risk" — banks lose ~15% of customers/year (5-10% is acceptable):
  → Low engagement scores = 20-30x greater chance of leaving
  → Top drivers: high fees, unattractive rates
  → ML models can reduce attrition by 7% in a single quarter

How SVB Failed (ALM perspective):
1. 89% uninsured deposits (highest of any major bank vs ~40-50% industry avg)
2. $91.3B in HTM securities (44% of assets), $86B in MBS with 10+ year maturities
3. Had breached own internal IRR limits on and off since 2017
4. Rates rose → HTM portfolio had $15.1B unrealized losses (HTM accounting hid them)
5. March 8: announced $1.8B realized loss + $2B capital raise
6. March 9: $42B withdrawn in a single day — fastest bank run in history
7. March 10: seized by FDIC