TL;DR
Three KPIs matter most: cycle time application-to-fund (industry: 45 days; target: 12–18), production cost per loan (industry: $11,800; target: $5,400–$7,200), and critical defect rate (industry: 1.79%; target: under 0.5%). The other seven below are leading indicators that explain movement in those three. Each KPI in the table includes the calculation, the industry baseline, the modern target, and the modernization fix when it goes red. Confer customers track these monthly via a built-in operations dashboard.
The 10 KPIs at a glance
| # | KPI | Industry baseline | Modern target |
|---|---|---|---|
| 1 | Cycle time — application to fund (days) | 45 days (MBA Q4 2024) | 12–18 days |
| 2 | Production cost per loan (USD) | $11,800 (MBA Q4 2024) | $5,400–$7,200 |
| 3 | Critical defect rate (%) | 1.79% (ACES Q4 2024) | <0.5% |
| 4 | Pull-through rate (%) | 70–82% typical mid-sized | 85–92% |
| 5 | Application completion rate (%) | 55–70% typical | 85–92% |
| 6 | TRID compliance rate (%) | 95–98% typical | 100% (zero violations) |
| 7 | Average documents per loan | 30–60 documents | Use as workload signal, not target |
| 8 | Loan officer productivity (loans/LO/month) | 4–6 (purchase-heavy), 8–12 (refi-heavy) | 10–18 with full automation stack |
| 9 | Pre-close exception rate (%) | 8–15% typical | <5% with stable production |
| 10 | HMDA filing prep time (FTE-hours) | 80–120 hours typical | <10 hours review |
Deep-dive: when each KPI goes red, what fixes it
Cycle time — application to fund (days)
Calculation: Median elapsed time from application creation to disbursement.
Industry baseline: 45 days (MBA Q4 2024)
Target: 12–18 days
When it goes red: Document processing, condition clearing, and TRID waiting periods are sequential rather than parallelized.
The fix: Parallelize with AI agents in document processing, income calculation, and condition automation. Confer customers target 12–18 day cycle.
Production cost per loan (USD)
Calculation: Total origination operating cost ÷ funded loans for the period.
Industry baseline: $11,800 (MBA Q4 2024)
Target: $5,400–$7,200
When it goes red: Manual document classification (40–60 min/loan) + manual income calc (90+ min/loan) drives the largest single line item.
The fix: Document AI + deterministic 1084 income calc. The two together account for the largest cost compression at mid-sized lenders.
Critical defect rate (%)
Calculation: Critical-defect loans ÷ sample size in post-fund QC.
Industry baseline: 1.79% (ACES Q4 2024)
Target: <0.5%
When it goes red: Income/employment defects (#1 category per ACES) caught post-fund rather than pre-close.
The fix: 100% pre-close continuous QC with reproducible income calculation. Investor repurchase exposure tracks this number directly.
Pull-through rate (%)
Calculation: Funded loans ÷ applications submitted in the cohort.
Industry baseline: 70–82% typical mid-sized
Target: 85–92%
When it goes red: Application drop-off at credit pull or document upload; condition-cycle delays trigger borrower abandonment.
The fix: Progressive borrower wizard, soft credit pulls early, AI document classification on upload, Voice AI for status calls.
Application completion rate (%)
Calculation: Applications submitted ÷ applications started in the cohort.
Industry baseline: 55–70% typical
Target: 85–92%
When it goes red: URLA forms force completion before any verification; borrowers abandon before credit pull.
The fix: 42-step progressive wizard with state persistence. Verification integrations (Plaid/Finicity) early in flow.
TRID compliance rate (%)
Calculation: Loans with on-time CD delivery ÷ total closings in the period.
Industry baseline: 95–98% typical
Target: 100% (zero violations)
When it goes red: Cron-based or queue-based TRID timers fail under server restarts, deploys, or worker crashes.
The fix: Durable workflow engine (Confer uses Temporal). Timer state persists across all failure modes.
Average documents per loan
Calculation: Total documents in the loan file at close ÷ funded loans.
Industry baseline: 30–60 documents
Target: Use as workload signal, not target
When it goes red: Climbing average suggests redundant requests for the same document or condition cycles.
The fix: Single intelligent document store; classification + de-duplication via AI; conditions tied to documents, not loan stages.
Loan officer productivity (loans/LO/month)
Calculation: Funded loans ÷ active LOs in the period.
Industry baseline: 4–6 (purchase-heavy), 8–12 (refi-heavy)
Target: 10–18 with full automation stack
When it goes red: LOs spend time on status calls, condition follow-up, and re-keying instead of new applications.
The fix: Voice AI for status calls. Real-time borrower portal for self-serve. Single state machine eliminates re-keying.
Pre-close exception rate (%)
Calculation: Loans flagged for exception review ÷ total in pipeline.
Industry baseline: 8–15% typical
Target: <5% with stable production
When it goes red: AI document classification confidence below 90% triggers human review at unsustainable rates; income calc variance flags fire too often.
The fix: Tune classification thresholds; ensure deterministic income calc; review exception routing rules quarterly.
HMDA filing prep time (FTE-hours)
Calculation: Staff hours spent on annual HMDA LAR preparation and validation.
Industry baseline: 80–120 hours typical
Target: <10 hours review
When it goes red: HMDA fields collected as year-end batch rather than continuously during origination.
The fix: Auto-populate 110+ LAR fields from origination data. FFIEC edit-check validation runs continuously, not in March.