PatentCensus transforms raw USPTO patent data into a structured view of where innovation is occurring, how it is evolving, and which technologies are shaping regional ecosystems.
PatentCensus is designed to deliver clear, decision-ready insights while maintaining rigor in how data is processed, normalized, and analyzed.
All data is sourced from the USPTO Patent File Wrapper bulk datasets. These datasets include detailed information on applications, inventors, assignees, classifications, and filing timelines. PatentCensus retains the full dataset and applies normalization, geographic mapping, and industry classification to ensure consistency across regions and over time.
Innovation in a region should be understood across two primary dimensions: level and direction.
Innovation Level is measured by the Innovation Insight Index™ and reflects the overall strength of innovation activity. Innovation Direction is measured by Economic Momentum™ and reflects how that activity is changing over time relative to other regions.
A region may have a high level of innovation but slower relative growth, indicating a mature and established ecosystem. Conversely, a region with lower overall activity but strong growth may represent an emerging innovation hub.
Innovation is inherently distributed. A single patent application often reflects contributions from inventors in multiple locations.
PatentCensus accounts for this by attributing each application proportionally across all inventor locations. Each inventor is mapped to a geographic region, and the application is distributed based on the share of inventors in each region.
Example
This preserves total national counts while enabling fair comparisons across regions.
Industries directory leading locations use a different measure. Card and breakdown location lists rank US states by distinct application counts in Agg_Industry_Stats (GeoType = STATE), built from ApplicationGeography address rows linked to the application. Ingestion can include inventor, applicant, correspondence, attorney/assignee, and related addresses, so attribution is mixed. An application may count in more than one state. That ranking is not inventor-fractional attribution and does not identify headquarters or firm offices exclusively. ApplicationGeography does not store a reliable address-role field for every row (InventorSequence may be set for some inventor-derived rows, but a null value is not proof of a non-inventor address), so individual leading-location counts cannot be partitioned by original address role from stored geography alone. Rankings are not adjusted when a state appears frequently; labels match this source and measure.
PatentCensus groups applications using CPC-based technology categories. These categories describe patent subject matter, not necessarily an applicant’s business industry.
Applications may belong to more than one technology category. Technology totals and percentages should not be added together as if the categories were mutually exclusive.
Mapping source: the live CPCIndustryMap table. Matching uses dbo.fn_CpcIndustryMapMatch: longest mapped CPCPrefix that is a prefix of the application's classification key wins (ORDER BY LEN(CPCPrefix) DESC). Industry analytics and classification breakdowns read the same map (aggregates via the match function; prefix lists via CPCIndustryMap for the selected industry bucket).
Classifications with no map match: fn_CpcIndustryMapMatch returns no row. Industry aggregate rebuilds join through that match, so unmatched classifications do not contribute to industry totals. That is separate from Traditional / Other, which is an explicit industry bucket with its own mapped prefixes in CPCIndustryMap (a catch-all assignment for listed codes, not an automatic label for every unmatched symbol). The public Industries directory hides Traditional / Other cards while still including that bucket in assignment-share denominators when present in the analytics response.
Map revisions and snapshots: the runtime mapping identity is the current contents of CPCIndustryMap. The product does not expose a formal mapping version number. Dated backup tables such as CPCIndustryMap_backup_20260403_1106 are point-in-time snapshots for compare/rollback, not a published version identifier. After map edits, industry aggregates must be rebuilt for counts to reflect the change.
Pharma vs Biotech: in the current map, A61K is assigned to Pharma (with other Pharma prefixes such as A61P and C07*). Biotech is defined by a different prefix set (for example C12N, C12Q, C12P, C12M, G16B). There is no shared A61K map rule sending the same prefix to both buckets. An application can still appear in both industries when it carries classifications that map to Pharma and classifications that map to Biotech. Earlier overview cards that showed A61K on both industries used mock/stale stand-in codes for display, not the live map.
Official CPC scheme titles are not stored on the industry map or breakdown fields used today; classification breakdowns therefore show mapped symbols without invented titles. Open Industries and choose View classification breakdown to inspect the mapped prefixes for a category.
PatentCensus combines a small set of proprietary metrics so you can read level and direction together—without turning the page into a spreadsheet.
The Innovation Insight Index™ summarizes how strong and diversified a region's patent activity is relative to peer regions: recent activity, intensity per resident, and whether filings are spread across industries or concentrated in a few. It is shown on a 0–100 scale with High, Moderate, and Low bands.
Economic Momentum™ describes whether innovation activity is accelerating, steady, or cooling relative to other regions over a multi-year window. We prioritize completed filing cohorts so publication lag does not distort peer comparisons.
Local Industry Signal™ uses a location quotient (LQ) to compare a region's industry mix with the national mix over the same completed filing-year window. Local industry share is that industry's filings in the geography divided by the geography's total filings in the window (filings follow the inventor-location attribution described under Geographic Attribution). National industry share uses the same years from state-level industry aggregates. The implemented LQ is (local share + ε) ÷ (national share + ε), with ε = 0.0001 added to both shares—not an unsmoothed local÷national ratio. That smoothing matters when shares are small. Display groups use LQ ≥ 1.5 and 1.1 ≤ LQ < 1.5 as existing descriptive cutoffs; their rationale is undocumented and is not an established standard for industry strength, growth, or quality. LQ is specialization relative to the nation, not commercial performance.
Each region should be understood through two primary dimensions: level and direction.
The Innovation Insight Index reflects the level of innovation activity, while Economic Momentum reflects the direction of that activity. A region may have a high level of innovation but slower relative growth, indicating a mature and established ecosystem. Conversely, a region with lower overall activity but strong growth may represent an emerging innovation hub.
PatentCensus combines these signals to provide a more complete view of regional innovation dynamics, helping users distinguish between established centers, emerging regions, and areas of accelerating change.
To simplify interpretation, PatentCensus generates a concise narrative summary for each region. This summary is based on the region's innovation level, growth trajectory, leading industries, and long-term trends.
These summaries are generated using structured logic to ensure consistency with the underlying data. The goal is to provide immediate context so users can quickly understand what defines a region's innovation profile.
Innovation Context summaries are deterministically generated — they follow strict rules tied to each region's III bucket, EMI category, and top industry signals, ensuring every summary is consistent and data-accurate.
PatentCensus maintains more than a decade of historical patent data, enabling long-term trend analysis and identification of structural shifts in innovation.
Data is updated regularly as new USPTO records become available. Because patent data is subject to publication timing and processing delays, recent filing years can be incomplete until applications publish. Trend and momentum metrics use completed filing cohorts so comparisons reflect real multi-year behavior; headline charts may still show partial recent years where labeled.
10+ Years
Historical Coverage
USPTO Bulk Data
Data Source
Regular Updates
Update Cadence
PatentCensus is designed to make innovation data accessible without requiring users to interpret raw patent datasets. The platform emphasizes clarity, consistency, and comparability across regions.
While the underlying models incorporate multiple inputs and normalization techniques, the focus is on delivering insights that are intuitive, reliable, and useful for strategic decision-making.
Apply these metrics across states, metros, counties, and ZIP codes on the interactive map.