Deep Dive
Step-by-step breakdown
Step 1. Development Sample Vintages and Economic Context
FICO 8 was developed on mid-2000s pre-crisis credit data during a period of rapid credit expansion. FICO 9 used early 2010s post-recession data capturing mass defaults, short sales, and recovery patterns. This vintage difference means the models calibrated risk predictions against different economic realities.
The coefficient structure established during development persists throughout a model's production life. FICO 8 still carries coefficients optimized for pre-crisis behavior after 15+ years in production. The fundamental risk relationships are stable, but precise calibration may not perfectly match current conditions.
FICO validates each version against contemporary out-of-time data but does not recalibrate deployed models. This is why FICO periodically releases new versions, but slow lender adoption means older models persist long past their development data vintage.
- FICO 8 trained on pre-crisis data; FICO 9 on post-recession data
- Coefficient structures persist through the model's entire production life
- FICO 8 has been in production 15+ years without recalibration
- Fundamental risk relationships are stable even as calibration ages
- FICO does not recalibrate deployed models; it releases new versions instead
Step 2. Collection Account Treatment: The Core Divergence
FICO 8 ignores collections under $100 original balance but penalizes all others, including paid ones. The penalty for a paid collection under FICO 8 is reduced but not eliminated. The design logic is that the event of going to collections is itself predictive, regardless of subsequent payment.
FICO 9 ignores all zero-balance collections. The design logic is that resolving a collection demonstrates willingness to address debt, and the predictive value diminishes once the obligation is satisfied. This change can produce 25-75 point improvement for consumers with paid collections.
FICO 9 also differentiates medical from non-medical collections for unpaid accounts, reflecting research showing medical debt is less predictive of default on other obligations. This dual treatment is the primary reason FICO 9 scores are systematically higher for consumers with collection tradelines.
- FICO 8 ignores only collections under $100; FICO 9 ignores all zero-balance collections
- The paid collection treatment can produce 25-75 point score gap between versions
- FICO 9 weights medical collections less than non-medical for unpaid accounts
- FICO 8's logic: the event of collections is predictive regardless of payment
- FICO 9's logic: resolving collections demonstrates willingness to address debt
Step 3. Rental Payment Data: FICO 9's Inclusion Innovation
FICO 9 was the first FICO version to incorporate rental payment data into scoring. When landlords report rent payments to bureaus through reporting services, FICO 9 evaluates this as a positive tradeline. FICO 8 does not recognize rental data and ignores it entirely.
The impact is most significant for thin-file consumers lacking traditional accounts. FICO estimated rental data could improve scores for approximately 10 million consumers with limited histories. However, most landlords do not report rent payments, and consumers must typically opt in through a rent-reporting service.
Rental data reporting is not universal and appears only at the bureau to which it is reported. This fragmented landscape means the benefit is available only to a subset of renters who actively pursue it, limiting the population-level impact of this FICO 9 feature.
- FICO 9 incorporates rental payments; FICO 8 ignores them entirely
- Rental data has greatest impact for thin-file consumers with limited traditional credit
- FICO estimated 10 million consumers could benefit from rental data inclusion
- Rental reporting requires active opt-in through a rent-reporting service
- Data may appear at only one bureau, creating cross-bureau inconsistency
Step 4. Authorized User and Piggybacking Detection
Both FICO 8 and 9 include safeguards against authorized user piggybacking. FICO 8 introduced algorithmic detection using signals like address mismatch, file inconsistency (thin file with a single 20-year-old high-limit card), and single-tradeline reliance.
FICO 9 refined the detection algorithms but uses the same probabilistic approach. The filter is not deterministic: some genuine AU accounts may have scoring contribution reduced, while some piggybacking may pass undetected. Legitimate family-member AU relationships still contribute positively.
The practical impact is that consumers using authorized user strategies should understand that both versions may discount the AU account's contribution if the relationship appears inconsistent with the overall file profile. Neither version has eliminated piggybacking entirely, but both have reduced its effectiveness.
- FICO 8 introduced the first algorithmic piggybacking detection
- Detection signals include address mismatch, file inconsistency, single-tradeline reliance
- Both versions use probabilistic detection: some genuine AUs may be discounted
- Legitimate family-member AU relationships still contribute positively
- Neither version has eliminated piggybacking entirely
Step 5. Utilization Sensitivity Differences
FICO 8 increased the utilization penalty gradient at the 30%, 50%, 70%, and 90% thresholds compared to prior versions. FICO 9 maintained this sensitivity but refined the low-end treatment, more clearly defining the 1-9% optimal range with a small penalty for zero utilization.
The practical utilization difference between FICO 8 and 9 is relatively small compared to the collection treatment divergence. Both respond similarly to utilization changes at the individual consumer level. The primary utilization-related variance comes from their interaction with other factors: FICO 9's more generous collection treatment can shift scorecard assignment, changing the utilization coefficients applied.
This secondary effect is important: removing collections from the scoring calculation under FICO 9 can move a consumer from a derogatory scorecard to a clean-file scorecard, where utilization coefficients may produce a different score contribution than on the derogatory scorecard.
- FICO 8 increased utilization penalty gradients at 30%, 50%, 70%, and 90% thresholds
- FICO 9 refined the 1-9% optimal range and zero-utilization penalty
- Utilization calibration differences are minor compared to collection treatment
- Scorecard reassignment from collection handling creates secondary utilization effects
- Both versions apply a small penalty for zero utilization across all revolving accounts
Step 6. Adoption Landscape: Why FICO 8 Still Dominates
FICO 8 remains dominant for credit cards, auto loans, and personal loans despite FICO 9's technical improvements. Lender inertia is the primary reason: underwriting systems, pricing models, and compliance documentation are calibrated around FICO 8. Transition costs include system recalibration, regulatory review, and staff retraining.
Mortgage lending skipped FICO 9 entirely. Fannie Mae and Freddie Mac use classic versions (2/4/5) with a planned transition directly to FICO 10T. FICO 9's primary adoption has been in tenant screening, insurance, and some non-mortgage lending applications.
For consumers, the practical impact is understanding which version their specific lender uses. A consumer with paid collections who sees a high FICO 9 on a monitoring app should not assume their lender sees the same number. If the lender uses FICO 8, paid collections still contribute negative weight.
- FICO 8 remains dominant for credit card, auto, and personal loan underwriting
- FICO 9 adoption limited by system recalibration costs and regulatory burden
- Mortgage lending skipped FICO 9 entirely, jumping from classic versions to planned FICO 10T
- FICO 9 found adoption in tenant screening, insurance, and some non-mortgage lending
- Consumers must verify which version their specific lender uses
