NATIONAL GIS DAY / RESEARCH & INNOVATION

Welcome to
CAP Banking Intelligence

Explore the places around banks: how CAP scores relate to nearby reported crime, how the view changes with distance, and what the surrounding neighborhood looks like.

Start exploring

15Cities of different sizes
A variety of perspectives

FROM GSX TO GIS DAY

Two examples became
a broader exploration.

This project builds on the Charlotte and Atlanta examples shared at GSX. For GIS Day, it expands to 15 cities of different sizes, including agencies with public data portals and agencies without them.

Why these 15 cities?

These 15 cities were selected to illustrate a variety of community sizes and geographic settings across the United States, including jurisdictions with and without public crime-data portals. The selection provides varied examples for GIS Day exploration; it is not a nationally representative sample.

SPATIAL DATA, CONNECTED

Different layers. A shared location.

Start with a bank

Bank locations from the Federal Deposit Insurance Corporation (FDIC) provide the starting point for each branch.

Add risk and reported crime

2026.1 CAP scores describe crime risk. Separately, 2025 CAP-standardized crime records describe reported events in half-mile and one-mile buffers around each bank.

Understand the neighborhood

The Census Bureau’s American Community Survey (ACS) adds population and housing context. WorldPop adds optional modeled population and a grid map. Workplace job density shows jobs per square mile. City values provide a reference for each neighborhood.

CAP INDEX’S APPROACH TO AIHuman expertise.
Scientific rigor.
Practical innovation.

At CAP Index, we take a practical and intentional approach to artificial intelligence (AI). CAP Index was founded by academic criminologists, and our crime risk models are grounded in criminological research, empirical data, and rigorous statistical methods. Scientific rigor and subject matter expertise remain central to how we assess and forecast crime risk, reinforced by our continued investment in Research and Innovation.

AI is not used in our crime risk score modeling approach. Full stop.

Instead, we use AI to support innovation, productivity, and efficiency across our business while keeping human judgment and subject-matter expertise at the center of our work. Across research, product development, sales, marketing, and daily operations, we use AI to explore ideas, accelerate proof-of-concept development, support problem-solving, and streamline routine tasks. We focus on applications that meaningfully improve workflows, reduce administrative burdens, and give our teams more time to address complex problems.

We view AI as a tool to enhance human expertise, not replace it. By applying emerging technologies thoughtfully, we can continue advancing our products and services while delivering greater value to our clients.

CHOOSE YOUR STARTING POINT

Three ways to explore

See the overall picture

Start with the study sample and compare half-mile and one-mile results across CAP score ranges. The Methodology page explains the GIS terms.

Explore a city

Select a city and bank. See its buffers, nearby reported crime and how much of each buffer falls inside the reporting boundary.

Meet the neighborhood

Explore one site or compare two in the same city. See neighborhood measures beside city reference values.

THE STUDY AT A GLANCE

CAP scores and nearby reported crime

Two measures, two purposes. The CAP Index is a crime-risk score. Nearby reported-crime records describe events already reported around each bank. Comparing the two provides geographic context; it does not establish crimes at the bank or predict an individual incident.

Branches included in comparisons1,875/ 2,127This matched sample includes banks with at least half of each buffer inside the reporting jurisdiction. The same banks are compared at both distances; some circles extend beyond the boundary.

Crime reporting period2025Full calendar year

Half-mile rank association—CAP and serious-violence density

One-mile rank association—Same 1,875 banks

The Overview focuses on selected serious violence: homicide, aggravated assault and robbery reported in 2025. Each eligible source row counts separately, even when several rows belong to one incident. Counts are unweighted: each row contributes one count. City View also offers the original five crime categories and exploratory weighting.

A QUICK GIS GUIDE

Four terms that help you read the results

GIS—geographic information systems—connects data to places. These terms explain how location shapes this comparison.

Buffer: A circle drawn around a bank at a set distance. This study uses half-mile and one-mile buffers.
Jurisdiction: The police agency’s reporting area. We count only records from that agency that fall inside its boundary and the bank’s buffer.
Crime density: Reported records divided by the buffer area inside the jurisdiction, expressed as records per square mile. This adjusts for how much area is included.
Median: The middle value when bank results are ordered from lowest to highest. It describes a typical branch in the selected group.

For example: 20 eligible records in 0.5 square miles gives a crime density of 40 records per square mile. This is a teaching example, not a study finding.

POOLED STUDY FINDINGS

Crime density across CAP score ranges

Compare the median crime density for banks in each CAP score range.

How to read this chart: Find a CAP score range along the bottom, then compare its two bars. Dark purple shows the half-mile buffer; light purple shows the one-mile buffer. A taller bar means a higher median crime density. The branch count under each range shows how many banks are in that group.

Each bar shows the middle (median) branch-level density in that CAP score range. CAP bands follow the CAP Crime Risk Score Scale; they are score ranges, not probabilities. Small groups need careful interpretation: only three banks are in the highest CAP range. These 15 selected jurisdictions do not represent all U.S. communities.

DISTANCE AND AREA

Geography changes the view

The one-mile buffer includes the half-mile buffer and reaches farther into the surrounding area. A larger buffer can contain more reported records while having a lower crime density, because those records are spread across a larger observed area.

Nearby records describe the area around a branch. They do not establish that an offense occurred at the bank.

JURISDICTION COVERAGE

Boundary coverage matters

252 banks have less than 50% of at least one buffer inside the reporting jurisdiction, so they are excluded from the matched comparison. Houston accounts for 229 of these exclusions.

Explore Houston’s boundary coverage →

THE 15-CITY COMPARISON

Compare the study jurisdictions

A jurisdiction is the police agency’s reporting area. Click a city to explore its banks, buffers and nearby reported crime.

Open City View

City / state	Branches included / total	Median CAP	½-mile crime density	1-mile crime density	Rank ½ mi	Rank 1 mi

Density means median reported records per inside-jurisdiction square mile. Rank values are Spearman correlations between bank CAP Index and nearby density. Small city samples and overlapping buffers limit inference.

NATIONAL CAP CONTEXT

CAP scores by community type

The national bank dataset includes 75,738 banks. This table groups their CAP scores by Urban, Suburban and Rural community types, using the same national CAP benchmark.

Median is the middle score; average is the sum of scores divided by the number of banks. Community types come from the bank dataset. This national view compares CAP scores only; the nearby reported-crime study still covers the 15 selected jurisdictions.

Community type	Banks	Median CAP	Average CAP

FROM GIS DAY TO EVERYDAY DECISIONS

Explore CAP Index

CAP Index provides address-specific crime risk intelligence, reports, and tools to support security planning and location decisions.

Continue the conversation about your locations, your portfolio, and the questions your team needs to answer.

Explore CRIMECAST reports Learn about CAP Index Talk to CAP Index

CAP BANKING INTELLIGENCE / CITY VIEW

Atlanta, GA

Explore how CAP scores relate to reported crime around bank locations.

2025 reported crime
2026.1 CAP Index scores

Study jurisdictionAtlanta Police Department

Distance from bank

½ mile1 mile

Crime categoriesHomicide, aggravated assault, robbery

Branches included in comparisons121/ 123Same qualifying branches at both distances. Partly covered buffers can qualify.

Median CAP Index—National benchmark

Median nearby records—Selected serious violence

CAP–density rank association—Compares individual branch ranks by CAP and density.

THE SPATIAL CONTEXT

Bank branches across the jurisdiction

CAP IndexCrime density

Basemap

Select a bank branch. Scroll to zoom. Drag to pan.

Selected block group Census place boundary

Show 2 excluded banks Agency jurisdiction Selected bank’s buffer

Colors show CAP Index at bank locations. The boundary represents the reporting agency’s jurisdiction.

SELECTED BRANCH

In matched sample

—

CAP Index—

Nearby records—

Records per square mile—Inside-jurisdiction buffer area

Selected bank crime categories and counts
Reported category	Records

Boundary coverage

—

SELECTED BRANCH / CENSUS CONTEXT

The neighborhood compared with its city

ACS 2020–2024 · five-year estimates

Measure / geography	Neighborhood	Census place / city	Difference

Residential stability uses the containing tract. All other neighborhood measures use the full containing block group. “Same home one year ago” describes residents age 1+, not household tenure. City values use the associated Census place, which can differ from the reporting jurisdiction.

± values are 90% margins of error (MOEs); derived percentage MOEs are approximate. “Wide uncertainty” means MOE is at least 50% of the estimate. Unavailable values remain unavailable. Open-ended medians retain their bounds. A minus B compares neighborhood estimates for the two sites, with no significance test; pp means percentage points. Population and household totals describe geography size. These are residents, not bank customers.

Preview the downloadable profile graphic

Sources: U.S. Census Bureau, 2024 ACS five-year detailed tables B01003, B11001, B01002, B19013, B25003, B25077, B15003 and B07003; 2024 TIGERweb geography. Income is in 2024 inflation-adjusted dollars. Neighborhood data are independent of the selected crime distance, categories, and weights.

SELECTED BRANCH / CITY BRANCH BENCHMARKS

The branch in local context

Compare with matched study branches at both distances.

CAP retains the national benchmark. Density benchmarks follow the current category and weighting settings. Excluded branches receive no local density ranking.

THE RELATIONSHIP

CAP scores and nearby crime density

Each point is one matched bank. Select a point to locate it on the map.

—rank association

—

THE BRANCH COMPARISON

Explore individual bank branches

Bank institution / branch					Comparison eligibility

—

Definitions, sources, and interpretation

What this prototype covers

The study covers 15 reporting jurisdictions and 2,127 original bank locations in calendar year 2025. The pooled comparison retains the same 1,875 banks whose half-mile and one-mile buffers each have at least 50% coverage inside their assigned jurisdiction. Each city view uses that city’s subset of the matched sample. The national CAP context includes all 75,738 bank locations and uses the existing CommType field. It does not apply the buffer filter.

Counting and geography

Each eligible source row counts separately, including rows that repeat an incident ID. Counts use the same ORI and locations inside the assigned jurisdiction. Density divides that count by the square miles of buffer inside the jurisdiction. One-mile buffers include half-mile buffers. Overlapping bank buffers can share source records.

Crime scope and weighting

Selected serious violence includes homicide, aggravated assault, and robbery. The original five-category option also includes simple assault and disorderly conduct. These are operational selections, not a complete definition of violent crime. Exploratory weighting changes density to weighted points per square mile. Weights do not establish relative harm or risk multipliers.

Census neighborhood methodology

Each branch is assigned by its saved point geometry to its full containing 2024 Census block group. Statistics are not clipped or apportioned to the jurisdiction or bank buffer. Block groups crossing the associated Census place are flagged. The city benchmark uses the full official Census place; Lexington uses Lexington-Fayette urban county. Residential stability uses the containing tract because B07003 is not published at block-group level. It measures residents age 1+ in the same home one year earlier. Neighborhood and place estimates both use ACS 2020–2024. Survey estimates have uncertainty; unavailable estimates are not replaced with zero. Percentages are calculated from matching table numerators and denominators, with approximate 90% MOEs using the Census subset-proportion formula and its fallback. Medians in open-ended intervals retain their bounds and receive no exact difference. These profiles provide resident context, not customer characteristics or explanations of crime. Census outlines are simplified for display only.

Neighborhood workplace jobs

Workplace jobs use 2024 LODES 8.5 WAC All Jobs (JT00). Published workplace-block internal points are assigned to the full 2024 profile block groups and Census places. Whole-block point assignment reconciles the geography vintages without assuming the release’s 2025 place boundaries match the 2024 profiles. Density uses 2024 Census land area. Jobs are not daytime population or foot traffic. Detailed geography, source variables and ACS margins of error remain in the neighborhood CSV and expandable methodology.

FEMA natural-hazard context

Each bank is spatially assigned to its containing FEMA National Risk Index tract, retaining FEMA’s boundary vintage separately from ACS. The December 2025 v1.20 snapshot covers 18 hazards. Overall risk and hazard scores are national relative measures, not probabilities. Leading building-loss contributors use positive tract-wide building expected annual loss. These values are not branch-level losses or flood-zone determinations. FEMA and CAP scores remain separate.

FBI agency reporting context

The FBI CDE panel follows the crime-category selector: summarized murder and nonnegligent manslaughter, aggravated assault, and robbery, plus NIBRS simple assault in five-category mode. Disorderly conduct is omitted from the FBI panel because FBI Group B arrest data are not interchangeable with offense counts. It remains in the supplied CAP five-category analysis. The official 2024–2025 participation file confirms 12 reported months for all 15 ORIs in 2025. Atlanta has only 10 reported months in 2024; Biloxi is not listed for 2024, so neither receives a year-over-year comparison. Other comparisons require full published reporting coverage and complete monthly source series. These totals do not alter the CAP source-row counts, are not allocated to buffers, and do not validate the geocoded extract’s completeness. Rates use the FBI agency population for each year.

Interpreting the comparison

2026.1 national current CAP Index scores describe a different measure from 2025 nearby reported records. Spearman correlation compares bank rankings. The study does not establish crimes at banks, causation, or prospective predictive accuracy. Area adjustment does not account for population, visitors, land use, or differences in reporting. Crime reporting completeness and the underlying CAP reference period require separate verification.

Source run: Banking_2025_20261103_083023_487395. Read-only exports from All_Banks_Combined.csv and saved Results.gdb geometries. Map geometry is simplified for display; analysis retains the original exported geodesic areas and counts. CAP Index branding reference: capindex.com. Street basemap: Esri, HERE, Garmin, OpenStreetMap contributors and GIS User Community. Online street tiles require internet access. The boundary map, study data, benchmarks and comparison exports work offline. Map display uses Web Mercator; analytical geodesic areas and counts remain unchanged. Boundary-map SVG exports omit online tiles. Local percentiles use (branches below + half of ties) / reference count × 100. The selected matched branch is included in its reference group. This is a research prototype, not the CRIMECAST product.