Credit Analysis # Same score. Same profile. Completely different risks. Twin brothers — same age, same income, same credit score. The only difference: where they live. And that difference changes everything. See how geospatial data reveals what the bureau can't capture. Score ≠ Risk same score, opposite risk contexts 1 coordinate reveals socioeconomic environment invisible to the bureau 0 CPFs required — 100% location-based enrichment The problem ## The credit score doesn't tell the whole story Traditional bureaus assess payment history, indebtedness, and registration behavior. But they systematically ignore the applicant's **territorial context** : where they live, what the socioeconomic surroundings look like, whether the declared income is consistent with the region's reality. Two clients with identical bureau profiles can represent radically different risks when assessed by what location reveals. 0 territorial context variables in most credit engines Scores treat clients in opposite socioeconomic contexts the same way 72.5M delinquent Brazilians — but scores don't identify local economic pressure Source: Serasa Experian, Delinquency Map, 2024 R$ 462B in delinquent debts in Brazil — much of it avoidable with better risk assessment Source: Serasa Experian, 2024. Consumer negative debts The scenario ## Twin brothers. Same bureau profile. Location changes everything. Imagine twin brothers — men, 40 years old, same declared income of R$ 10,000, identical family composition, same credit history and score. For any traditional bureau, they're the same profile. But they live in different neighborhoods in Joinville/SC. And that difference is decisive. What bureaus see — identical profile **Gender:** Male **Age:** 40 years **Income (bureau):** R$ 10,000 **Family composition:** Identical **Credit history:** Similar **Credit score:** ~700 (close) → For the traditional credit engine, both receive the same decision. Twin A — Attention signal ### Boa Vista — Joinville/SC R. Antonio de Oliveira Silva, 283 · [-26.301319, -48.819026] **Declared income:** R$ 10,000 **Regional median income:** R$ 2,156.39 **Z-score:** +2.39 (income far above local standard) ⚠ **CDF (percentile):** 99.17% ⚠ → Declared income exceeds 99% of regional residents. Strong socioeconomic incompatibility. Requires additional verification. Twin B — Consistent profile ### Downtown — Joinville/SC R. Pastor Fritz Buhler, 73 · [-26.307350, -48.849506] **Declared income:** R$ 10,000 **Regional median income:** R$ 14,190.73 **Z-score:** -0.14 (near local median — compatible) **CDF (percentile):** 44.37% → Income near the local median. High-income region, stable territorial context. Proceed. Without PrismaData, both have the same score and receive the same decision. With it, Twin A is immediately flagged — income far above the local standard, incompatible socioeconomic context. The credit engine gains an intelligence layer the bureau doesn't offer. The insight ## Risk isn't just about the person. It's about the context. A z-score of +2.39 doesn't automatically deny credit. It raises a question the bureau can't ask: **is the declared income compatible with the real surroundings where the applicant lives?** #### Income × surroundings incompatibility Declared income of R$ 10k in a region with a R$ 2.2k median is statistically anomalous. It may indicate registration inconsistency, inflated income, or document fraud — signals the bureau doesn't detect. #### Regional financial pressure Regions with lower median income and higher debt concentration exert real economic pressure on the applicant — cost of living, local unemployment, and financial stress that impact payment capacity. #### Early risk signals An elevated z-score is an early signal — before the first late payment, before negative listing. Location anticipates what the score will only capture months later, when the loss has already been recorded. The solution ## One coordinate. Socioeconomic intelligence ready for the credit engine. Send the applicant's address or coordinate. PrismaData returns the regional median income, compatibility z-score, comparative percentile (CDF), and sample size — numeric, continuous variables, ready to feed policy rules or ML model features. GET /v1/location/personal_income_pdf Twin A — Boa Vista, Joinville/SC { "renda_pessoas_com_renda_mediana": 2156.39, "renda_comparada_cdf": 99.17, "renda_comparada_z_score": 2.39, "renda_pessoas_com_renda_qtde": 435 } ⚠ Declared income exceeds 99% of the region. Z-score +2.39. Twin B — Downtown, Joinville/SC { "renda_pessoas_com_renda_mediana": 14190.73, "renda_comparada_cdf": 44.37, "renda_comparada_z_score": -0.14, "renda_pessoas_com_renda_qtde": 289 } ✓ Income near the region's median. Z-score -0.14. Socioeconomic consistency. renda_pessoas_com_renda_mediana R$ — real median Census tract median income. Base: Census 2022, Continuous PNAD, IPCA and proprietary models. z_score Income standard deviation How atypical is the declared income vs. the surroundings? Continuous variable, ideal for ML. CDF (percentile) Relative position What percentile of the region is the applicant in? 98% = alert. 6% = compatible. Integration ## Additional intelligence layer. Doesn't replace — complements. PrismaData works as a pre-score: it adds geospatial context variables before (or during) the decision engine processes the credit analysis. Three ways to apply: 01 #### Model feature Use `z_score` e `cdf` as numerical variables in ML models. Continuous, normalized, ready for training. 02 #### Policy rule Create rules in the engine: _"if z_score > 2.0, escalate for manual review"_ or _"if cdf > 95%, request additional income verification"_. No black box. 03 #### Combined auxiliary score Combine income z_score with PrismaData's border, prison, and community data to create a proprietary geospatial score that complements the bureau score. **Important:** the responsibility for the final decision remains with the partner's credit engine. PrismaData provides explainable and auditable variables — it doesn't make decisions. Expansion ## Income is just the beginning. Combine with other layers. The same coordinate used to validate income can be enriched with border, prison, and community data — each layer is an independent endpoint, combinable by coordinate. /v1/location/border Fronteira Border strip, twin city, nearest country. Flag for AML/CFT and compliance. /v1/location/prison Presídio Proximity to 1,500+ prison facilities. Detection of operations originated from inside prisons. /v1/location/slum Comunidade Community proximity. Context variable — can increase or decrease risk depending on the scenario. Governance ## Territorial data, not personal. LGPD-compliant. Auditable. #### No personal data The query key is the geographic coordinate — not CPF, name, or any personal identifier. The data returned describes the profile of a **region** , not an individual. #### LGPD-compliant There is no personal data processing in the enrichment layer. The input is a coordinate; the output is territorial context. This simplifies the compliance discussion. #### Auditable and explainable Every variable is traceable — income comes from Census 2022, Continuous PNAD, IPCA and proprietary models. Z-score and CDF are reproducible statistical calculations. Ideal for internal and regulatory audits. Results ## More context in the decision. Fewer surprises in the portfolio. Pre-score Geospatial enrichment before the decision engine processes Contínuo Z-score and CDF — numeric variables, not binary labels. Ideal for ML. 0 CPFs 100% enrichment by coordinate — no personal data, no regulatory friction **References:** ¹ Income data returned by PrismaData API (personal_income_pdf endpoint) for actual Joinville/SC coordinates. ² 2022 Demographic Census (IBGE) — median income by census tract, with PrismaData proprietary PDF models. ³ Serasa Experian, _Delinquency Map of Brazil_ , 2024. 72.5 million delinquent Brazilians. ⁴ PrismaData institutional presentation — Credit Analysis, twins case in Joinville/SC. ## What if you were making credit decisions with only half the information? Talk to PrismaData. We can run a test with your own addresses in minutes — and you decide if it makes sense for your engine. [ Talk to PrismaData ]() [ Explore the API → ]()