Patent Representation Index
We built a composite score that ranks 91 tech companies on how well they translate women's technical work into patent inventorship. The score combines five sub-indices into a single 0-100 number, normalized across the dataset, weighted to reward both workforce diversity and patent inventor representation.
This is version 2 of an index we first published in April 2026. The first version measured the gap between the share of women in a company's STEM workforce and the share of women among its patent inventors, but that metric had a structural flaw: a small gap could mean a company was doing well, or it could mean the company was male-heavy across the board with little distance between two low numbers. The composite score fixes that. A company can no longer rank well simply because both its workforce and its inventor pools are male-heavy.
Patent data comes from Google BigQuery (patents-public-data.patents.publications), covering all US granted patents. Workforce data comes primarily from government-mandated EEO-1 filings (OFCCP FOIA release, 2016-2020, plus each company's own more recent published EEO-1 where available) supplemented by company diversity reports. The two datasets do not identify the same individuals, and the inventor pool spans a company's full patent history while the workforce figure is a recent snapshot, so this is a company-level representation comparison, not an estimate of which employees became inventors.
Published April 2026, version 2. Patent data last refreshed July 2026 (US grants through 2025); workforce data uses each company's most recent published EEO-1 where available (2021-2024 for 54 large firms), falling back to the OFCCP FOIA release (2016-2020) otherwise. This is a patent-side representation index, not a complete innovation census or a claim about any individual company's culture. See methodology for the formula, sub-indices, and limitations.
The Big Picture
Only companies with both patent and workforce data are eligible for the composite. The 91-company subset is drawn from a larger 183-company patent dataset by joining with EEO-1 workforce filings.
The Composite Score Leaderboard
The top 10 companies in the index, ranked by composite score. Each score combines five sub-indices: women in STEM workforce, women inventors as a share of all inventors, the ratio between them, the trend over the last 24 months (sample-size weighted), and inventor concentration. Companies need 100 or more total US patents to appear in the leaderboard. All 88 leaderboard-eligible companies are in the full ranked table below. One caveat to read the ranking with: women's baseline participation in patenting varies widely by technology field (higher in pharma and biotech, much lower in semiconductor and mechanical fields), so this raw ranking partly reflects what a company invents, not only how well it credits women. It is a descriptive comparison, not one adjusted for each company's technology mix.
How to read a high score in context
The index is intentionally two-dimensional. It separately rewards having women in technical roles (Foundation) and crediting them on patents (Output and Ratio). Some companies will score well on one and poorly on the other. That tension is not noise. It is exactly why we publish the underlying numbers next to the composite score.
Cray is the cleanest illustration. Cray sits at #10 on the leaderboard despite having the lowest share of women in its STEM workforce of any company in the 91-company dataset (14.3%). The composite puts them in the top 10 because their representation ratio is the highest in the index: Cray's women-inventor share is about 39% of its women-in-STEM share, the closest proportional alignment in the dataset. That is a comparison of two aggregate shares, not a count of which employees became inventors. Cray has very few women in engineering, but the few they have are being recognized.
This is not a workplace-culture ranking. A high composite should not be read as "this is the best place for a woman engineer to work." Foundation matters too, and the full ranked table shows both Foundation and Output next to each composite score so readers can hold both numbers in mind at once.
The Bottom 10
The lowest composite scores in the index. Almost entirely semiconductor, networking infrastructure, storage, and legacy industrial companies. All have women inventor rates between 1.4% and 2.5%.
Why we built v2: the Tesla story
Version 1 of this index used a single metric: the gap between the share of women in a company's STEM workforce and the share of women among its patent inventors. By that metric, Tesla had one of the smaller gaps in the dataset at 15.5 points (ranked 7th-smallest of 91), well below the median gap of 25 points. On v1, Tesla looked like a relative leader.
But Tesla's STEM workforce contains roughly 60 percent the share of women that IBM's does (17.4% vs 32.2%). The "small gap" was small because Tesla had so few women to begin with. Of the distinct inventors named on Tesla's patents, only 1.9 percent are women, which is in the bottom 11 of the women inventor rates we measured. The gap metric was treating Tesla as a leader because both its workforce and inventor pools were male-heavy, not because it was doing anything well.
The composite score fixes this. Tesla now ranks #82 of 88 on the leaderboard. IBM, which appeared "worse" on the v1 gap metric (gap of 26.6 points), is now the highest-scoring big tech company in the index at #27 of 88, because IBM has both the highest women STEM workforce share in big tech (32.2%) and the highest women inventor share (5.6%). The composite rewards balance, not the absence of women.
How the Score Works
The composite combines five sub-indices, each scaled to 0-100 against fixed anchors, then weighted and summed. A score of 100 would mean a company sits at the top of every anchor band; a score of 0 would mean the bottom of every band. Real companies fall between roughly 16 and 75.
Foundation (25%)
Women in the professional and technical workforce, as a percentage. From EEO-1 Categories 2 (Professionals) and 3 (Technicians). These are broad job categories used as a proxy for the technical workforce, not an exact count of engineers or R&D staff (Professionals also includes some non-technical roles). We label it "women in STEM" for brevity. It is the baseline: you cannot have inventor representation without workforce representation.
Output (35%)
Women inventors as a percentage of all distinct inventors at the company. The most heavily weighted sub-index because it is the actual outcome being measured: whether women in the engineering workforce are being credited as inventors when patents are filed.
Representation Ratio (25%)
How a company's women-inventor share compares, proportionally, to its women-in-STEM share. Computed as Output divided by Foundation. A value near 1 means women are named as inventors roughly in proportion to their presence in the technical workforce. This compares two aggregate shares from different populations (the all-time inventor pool and a recent workforce snapshot); it is not a count of which individual employees became inventors, which public data cannot show. Caveat: the ratio can favor companies that credit a high share against a small women-in-STEM base, which is why it should always be read alongside the Foundation and Output values.
Trend (10%)
Relative change in the women inventor rate over the last 24 months, weighted by the size of the recent patent sample. We use relative change (not absolute) so improvement from a low baseline counts as real progress: 2% → 4% is a 100% relative gain, not a 2-point one. Sample-size weighting pulls noisy small samples toward zero so a few patents cannot dominate.
Distribution (5%)
The inverse of overall inventor concentration. Specifically, 100 minus the share of patents accounted for by the company's top 5% most prolific inventors of any gender. A less concentrated inventor pool gives more engineers a path into the patent system, which may create more room for women to appear. This is a weak proxy signal, not a direct gender measure, and the 5% weight reflects that it is a tiebreaker rather than a primary metric.
The formula
Composite = Foundation × 0.25 + Output × 0.35 + Ratio × 0.25 + Trend × 0.10 + Distribution × 0.05
Each sub-index is scaled to 0-100 against fixed anchors (a constant band bracketing the realistic range of each measure) before weighting, so a company's score depends only on its own numbers and stays comparable across refreshes as companies enter or leave the dataset.
How we handle small samples in the trend
The trend sub-index uses relative change rather than absolute change. Going from 2% women inventors to 4% counts as a 100% relative improvement, not a 2 percentage-point absolute one. We chose relative change deliberately so the index recognizes meaningful improvement from low baselines, where most companies in this dataset still sit.
But relative change has a known weakness: it explodes near zero. A company that filed only 3 patents in the last 24 months and happened to credit one woman would look like a 33% women inventor rate, swamping any real signal. To handle this we apply sample-size shrinkage toward zero, the same regularization technique used by baseball sabermetrics, IMDB Top 250 ratings, and small-sample polling. The formula:
adjusted_change = relative_change × (recent_patents / (recent_patents + 50))
This pulls noisy small samples toward zero (no change) while leaving large samples close to face value. A company with 5 recent patents has its trend signal trusted at 9 percent of face value. A company with 5,000 recent patents has its signal trusted at 99 percent. Big tech with thousands of recent patents shows their real trend. Small companies with noisy samples have that noise neutralized. We chose k = 50 based on the dataset's median recent-patent count of roughly 220, where it gives 81 percent confidence at the median.
Relative change is also capped at the bounds +2.0 (200% improvement) and -1.0 (100% decline) before shrinkage is applied. The cap exists precisely because relative change near zero is unstable, and prevents any single outlier from dominating the trend score even if shrinkage is incomplete.
Robustness Check: Are the Rankings Stable?
The natural critique of any composite index is that the rankings might be artifacts of the specific weights chosen. To test that, we ran 4,000 sensitivity analyses across two Monte Carlo regimes plus six named alternative weight schemes. The findings give us confidence that the headline rankings are robust to plausible reweightings.
Test 1: Bounded perturbation around our weights
We sampled 2,000 weight vectors with each sub-index allowed to vary up to ±5 percentage points from its baseline. This asks: if our specific weights are slightly wrong, do the rankings hold?
9 of the 10 leaderboard companies remained in the top 10 in at least 80 percent of runs. 7 of the 10 bottom companies remained in the bottom 10 at the same threshold. Tesla stayed in the bottom 10 in 100 percent of runs. The only top 10 entry that moved meaningfully was Cray (in the top 10 in 53 percent of runs), reflecting that Cray's score depends heavily on the Representation Ratio sub-index — exactly what the Cray section above explains. We use 80 percent as a practical stability threshold here, not a formal statistical cutoff.
Test 2: Pure random weights (Dirichlet)
We then sampled 2,000 weight vectors from a symmetric Dirichlet distribution, allowing any combination of weights summing to 1.0 with no constraints. This is a worst-case stress test, not an alternative design universe. It includes degenerate cases like 100 percent on a single sub-index, which is not what a thoughtful designer would actually publish. We treat it as a stress test rather than a competing methodology, but the result is meaningful: it answers how much of the leaderboard survives even under weights nobody would choose on purpose.
Under this much harsher regime, 5 of the 10 top companies remained in the top 10 in at least 80 percent of runs. Starbucks (99.8%), Etsy (99.7%), Airbnb (91.1%), Procter & Gamble (88.4%), and Pinterest (88.8%) were the most robust top 10 entries. Mid-pack companies moved more, which is expected: when companies sit close together on a composite, small weight changes naturally swap their relative positions.
Test 3: Named alternative weight schemes
Finally, we ran six specific alternative weighting schemes that a reviewer might propose, including equal weights, Output-dominant (60% on Output), Foundation-dominant (50% on Foundation), Foundation + Output only, composite minus Distribution, and Ratio-heavy. Here is how the headline anchor companies fared across all six:
| Company | Baseline | Equal | Output-dom | Foundation-dom | F + O only | No Distribution | Ratio-heavy | Range |
|---|---|---|---|---|---|---|---|---|
| Starbucks | #1 | #1 | #1 | #1 | #1 | #1 | #1 | 1–1 |
| Etsy | #2 | #2 | #2 | #2 | #2 | #2 | #2 | 2–2 |
| Procter & Gamble | #3 | #4 | #3 | #5 | #3 | #3 | #3 | 3–5 |
| IBM | #27 | #51 | #23 | #28 | #23 | #23 | #23 | 23–51 |
| Microsoft | #36 | #44 | #35 | #38 | #35 | #35 | #37 | 35–44 |
| Apple | #48 | #66 | #46 | #49 | #43 | #40 | #47 | 40–66 |
| Amazon | #54 | #59 | #54 | #45 | #46 | #50 | #56 | 45–59 |
| Tesla | #82 | #79 | #83 | #87 | #87 | #86 | #77 | 77–87 |
| Lumentum | #85 | #77 | #86 | #72 | #78 | #83 | #86 | 72–86 |
Starbucks ranks #1 in all six alternative schemes. Etsy ranks #2 in all six. Tesla stays in the bottom 10 of every scheme. Lumentum stays in the bottom 10 of every scheme. The Range column makes the asymmetry visible: the top three move 0–2 positions, the bottom two move 10–14 positions, and the mid-pack big tech names move 9–28 positions.
What the sensitivity analysis actually shows
Reading the three tests together, the most defensible interpretation is more specific than "the rankings are robust." It is:
- The top of the index is fairly stable. Starbucks, Etsy, and Procter & Gamble survive almost any reasonable weighting.
- The bottom of the index is fairly stable. Lumentum, Marvell, and Tesla stay near the bottom under almost any reasonable weighting.
- Mid-pack positions are meaningfully more sensitive. Big tech (Apple, Microsoft, Amazon, Intel, Google) sits in a region where small changes to the weights can swap rankings.
- Cray is notably less stable than the other top 10 names (in the top 10 in only 53 percent of bounded perturbation runs). That is honest about what the Cray section above explains: Cray is in the top 10 because of a single sub-index, not because of balanced strength across all five.
- Tesla remains in the lower tier under almost anything reasonable. The v1-to-v2 correction holds.
What this analysis tests and does not test: The sensitivity tests above examine robustness to weight choices. They do not test robustness to other modeling choices, such as the fixed-anchor normalization bands, the choice of which sub-indices to include, or the gender-from-first-name estimation method. A more complete robustness program would test all of those independently. We treat this analysis as a meaningful first robustness layer, not the final word.
The Full Ranked Index: 91 Companies
All 91 workforce-matched companies, ranked by composite score. Companies marked "small portfolio" have fewer than 100 patents and are excluded from the curated leaderboard but appear here for transparency. Their scores should be interpreted with caution because percentage estimates from small samples have wide confidence intervals.
| Rank | Company | Composite | Women in STEM | Women Inventors | Representation Ratio | Trend | Patents |
|---|---|---|---|---|---|---|---|
| 1 | Starbucks | 74.7 | 49.0% | 16.9% | 34% | ↑ | 252 |
| 2 | Etsy | 63.5 | 52.6% | 13.0% | 25% | ↑ | 113 |
| 3 | Procter & Gamble | 60.7 | 44.2% | 12.5% | 28% | ↑ | 19,285 |
| — | Expedia (small portfolio) | 54.2 | 47.8% | 9.9% | 21% | ↓ | 43 |
| 4 | Airbnb | 53.7 | 27.5% | 9.3% | 34% | ↑ | 226 |
| 5 | Johnson & Johnson | 50.8 | 47.4% | 9.4% | 20% | ↑ | 4,555 |
| 6 | Tableau | 49.9 | 30.2% | 8.8% | 29% | ↓ | 302 |
| 7 | Intuit | 48.4 | 60.6% | 8.4% | 14% | ↓ | 5,286 |
| 8 | 46.5 | 43.5% | 7.7% | 18% | ↑ | 140 | |
| 9 | Corning | 44.2 | 29.3% | 7.4% | 25% | ↑ | 22,657 |
| 10 | Cray | 43.6 | 14.3% | 5.5% | 38% | ↓ | 423 |
| 11 | Nordstrom | 43.6 | 72.1% | 4.4% | 6% | ↓ | 287 |
| 12 | Palantir | 41.9 | 32.3% | 6.6% | 20% | ↑ | 1,678 |
| 13 | Lyft | 41.1 | 34.4% | 6.5% | 19% | ↑ | 566 |
| 14 | Zillow | 40.6 | 41.5% | 6.1% | 15% | ↑ | 135 |
| 15 | AT&T | 40.3 | 29.7% | 6.3% | 21% | ↑ | 44,407 |
| 16 | Xerox | 38.8 | 32.2% | 5.0% | 16% | ↑ | 39,148 |
| 17 | Salesforce | 38.4 | 38.3% | 5.7% | 15% | → | 4,712 |
| 18 | Twilio | 36.5 | 31.0% | 5.5% | 18% | ↑ | 352 |
| 19 | Docusign | 36.4 | 38.9% | 5.2% | 13% | → | 181 |
| 20 | Uber | 36.2 | 38.2% | 4.5% | 12% | ↑ | 963 |
| 21 | Coinbase | 36.2 | 31.6% | 5.2% | 16% | ↑ | 143 |
| 22 | Bloomberg | 36.1 | 27.9% | 5.5% | 20% | ↑ | 166 |
| 23 | Netflix | 36.0 | 40.9% | 4.7% | 11% | ↓ | 529 |
| 24 | Boeing | 35.9 | 27.2% | 5.1% | 19% | ↑ | 24,291 |
| 25 | Weyerhaeuser | 35.7 | 31.7% | 5.3% | 17% | → | 1,291 |
| 26 | 3M | 35.4 | 37.4% | 4.9% | 13% | → | 11,037 |
| 27 | IBM | 35.0 | 32.2% | 5.6% | 17% | ↓ | 177,221 |
| 28 | Conduent | 34.8 | 49.8% | 4.0% | 8% | ↓ | 341 |
| 29 | Meta (Facebook) | 34.7 | 32.2% | 4.9% | 15% | ↑ | 10,415 |
| 30 | Alphabet (Google) | 34.5 | 32.1% | 5.0% | 16% | ↑ | 40,669 |
| 31 | Splunk | 34.4 | 34.2% | 4.8% | 14% | ↑ | 1,975 |
| 32 | Snap | 34.2 | 32.1% | 4.9% | 15% | ↑ | 3,593 |
| 33 | HP Inc. | 34.0 | 32.4% | 5.0% | 15% | ↑ | 44,070 |
| 34 | Rapid7 | 33.7 | 28.9% | 4.5% | 16% | → | 310 |
| 35 | Raytheon | 33.2 | 25.1% | 4.4% | 18% | ↑ | 14,149 |
| 36 | Microsoft | 32.9 | 32.3% | 4.7% | 15% | → | 57,420 |
| 37 | eBay | 32.8 | 33.7% | 4.6% | 14% | ↓ | 3,766 |
| 38 | Workday | 32.8 | 43.7% | 3.4% | 8% | ↑ | 285 |
| 39 | Lockheed Martin | 32.7 | 24.1% | 4.5% | 19% | ↓ | 6,237 |
| 40 | PayPal | 32.2 | 37.1% | 4.4% | 12% | ↓ | 2,742 |
| 41 | Northrop Grumman | 31.9 | 23.9% | 4.0% | 17% | ↑ | 3,616 |
| 42 | Zoom | 31.6 | 29.0% | 4.3% | 15% | → | 729 |
| 43 | Cloudflare | 31.2 | 31.2% | 4.1% | 13% | ↑ | 331 |
| 44 | Garmin | 31.2 | 18.0% | 3.7% | 21% | ↑ | 880 |
| 45 | KLA | 31.1 | 18.8% | 3.5% | 19% | ↑ | 17,233 |
| 46 | MicroStrategy | 30.9 | 31.5% | 4.0% | 13% | ↑ | 355 |
| 47 | Adobe | 30.8 | 35.5% | 4.0% | 11% | ↑ | 7,093 |
| 48 | Apple | 30.5 | 29.4% | 4.3% | 15% | ↑ | 42,190 |
| 49 | CrowdStrike | 30.3 | 23.7% | 3.6% | 15% | ↑ | 187 |
| 50 | Oracle | 30.1 | 36.1% | 3.9% | 11% | → | 13,618 |
| 51 | ServiceNow | 30.0 | 36.1% | 3.6% | 10% | ↓ | 1,146 |
| 52 | TE Connectivity | 29.8 | 28.2% | 3.8% | 13% | ↑ | 1,882 |
| 53 | Electronic Arts | 29.3 | 29.7% | 4.0% | 13% | ↓ | 628 |
| 54 | Amazon | 29.1 | 32.0% | 3.8% | 12% | ↑ | 23,915 |
| 55 | SAP | 29.0 | 37.3% | 3.0% | 8% | ↑ | 9,682 |
| 56 | Applied Materials | 28.9 | 22.4% | 3.9% | 17% | ↑ | 14,663 |
| — | T-Mobile (small portfolio) | 28.1 | 43.0% | 1.8% | 4% | → | 27 |
| 57 | Dell | 27.7 | 30.0% | 3.4% | 11% | ↑ | 13,970 |
| 58 | Honeywell | 27.6 | 22.8% | 3.3% | 14% | ↑ | 27,301 |
| 59 | Seagate | 27.4 | 23.5% | 3.6% | 15% | ↓ | 8,883 |
| 60 | Motorola Solutions | 27.1 | 23.5% | 3.1% | 13% | ↑ | 2,245 |
| 61 | Micron Technology | 27.1 | 19.8% | 3.3% | 17% | ↑ | 41,620 |
| 62 | Intel | 26.2 | 25.9% | 3.5% | 14% | → | 84,365 |
| 63 | Keysight | 25.9 | 31.0% | 2.7% | 9% | → | 698 |
| 64 | NetApp | 25.8 | 26.9% | 2.7% | 10% | ↑ | 3,346 |
| 65 | Cadence Design Systems | 25.7 | 21.3% | 2.5% | 12% | ↑ | 2,578 |
| 66 | Cisco | 25.5 | 30.0% | 2.9% | 10% | → | 23,304 |
| 67 | Analog Devices | 25.4 | 23.8% | 2.8% | 12% | ↑ | 4,877 |
| 68 | Broadcom | 24.7 | 28.0% | 2.6% | 9% | ↑ | 11,447 |
| 69 | AMD | 23.9 | 21.4% | 3.3% | 15% | ↓ | 13,104 |
| 70 | Texas Instruments | 23.7 | 19.7% | 2.9% | 15% | → | 32,244 |
| 71 | Caterpillar | 23.0 | 26.5% | 2.4% | 9% | ↑ | 15,245 |
| 72 | Fortinet | 22.8 | 20.2% | 2.5% | 12% | ↑ | 1,068 |
| 73 | Cirrus Logic | 22.6 | 18.8% | 2.1% | 11% | ↑ | 2,695 |
| 74 | F5 Networks | 22.6 | 24.0% | 2.2% | 9% | → | 486 |
| 75 | Snowflake | 22.5 | 29.4% | 2.4% | 8% | ↑ | 1,053 |
| 76 | VMware | 22.1 | 29.9% | 2.1% | 7% | ↓ | 5,613 |
| 77 | Palo Alto Networks | 21.9 | 31.4% | 1.7% | 5% | → | 789 |
| 78 | NVIDIA | 21.4 | 18.6% | 1.9% | 10% | ↑ | 5,742 |
| 79 | Juniper Networks | 21.4 | 24.9% | 1.8% | 7% | ↑ | 4,785 |
| 80 | General Electric | 21.1 | 22.0% | 1.8% | 8% | → | 1,147 |
| 81 | Sandisk | 20.9 | 27.3% | 1.9% | 7% | → | 13,622 |
| 82 | Tesla | 20.1 | 17.4% | 1.9% | 11% | ↓ | 853 |
| 83 | Equinix | 20.1 | 24.2% | 1.7% | 7% | ↓ | 169 |
| 84 | QUALCOMM | 20.0 | 21.4% | 2.5% | 12% | → | 44,009 |
| — | Rackspace (small portfolio) | 19.9 | 21.3% | 1.5% | 7% | → | 88 |
| 85 | Lumentum | 19.8 | 27.7% | 1.5% | 5% | ↓ | 466 |
| 86 | Synopsys | 19.8 | 21.5% | 1.8% | 8% | → | 2,549 |
| 87 | Akamai | 18.4 | 25.7% | 1.6% | 6% | ↓ | 635 |
| 88 | Marvell Technology | 16.2 | 20.8% | 1.4% | 7% | ↑ | 9,067 |
Composite scores are normalized 0-100 across the dataset. Trend arrows: ↑ improving, → stable, ↓ declining. Hover for sample-size confidence.
Want the deeper data?
The full data explorer for all 183 companies includes 24-month trend movements, top improvers and decliners, CPC technology areas where women are filing, and the underlying year-over-year breakdown.
The National Trend
Women Inventor Rate from USPTO data, 2005-2024. Progress is real but slow. At the current rate of roughly 0.3 percentage points per year, parity would take more than 100 years.
USPTO reported 12.8% benchmark Estimated
Methodology and Limitations
Patent Data
Google Patents Public Datasets on BigQuery (patents-public-data.patents.publications). Complete coverage of all US granted patents with disambiguated inventor and assignee names. 1,678,137 patents analyzed across 183 companies. Free tier: 1 TB/month.
Company Universe
200 companies were queried: top 100 tech by market cap plus major employers in Washington, Texas, California, and other regions. 169 returned meaningful patent data (more than 10 patents). 91 of those 169 also have workforce data, which is the subset scored on the composite.
Gender Estimation
Gender is estimated from inventor first names using the same methodology USPTO uses for its annual reports. Gender-neutral and non-Western names may be misclassified. Approximately 5-10 percent of inventor names are unclassifiable and excluded.
Workforce Data
Government-mandated EEO-1 filings released by OFCCP via FOIA in 2026, covering 2016-2020. STEM proxy is EEO-1 Categories 2 (Professionals) plus 3 (Technicians). Where EEO-1 was unavailable, company-published diversity reports were used and tagged Tier B in the source CSV.
Composite Score Choices
Weights of 25/35/25/10/5 reflect a deliberate emphasis on actual patent representation (Output 35%) and workforce foundation (Foundation 25%) as the load-bearing measures. Trend gets only 10% because it is noisier and easier to game. These weights are documented choices, not derived from data.
Key Limitations
This is a representation index, not a causal measure. We cannot tell from the data whether women are not contributing inventions, not being listed on patents they contributed to, not being put forward for patent review, or some combination. The composite measures the outcome, not the cause.
What changed in v2
Version 1 of the index used a single metric: the gap between women in STEM workforce share and women inventor share. We discovered during a public review that the gap metric had a structural flaw. A small gap could mean a company was doing well, or it could mean the company was male-heavy across the board with two low numbers sitting close together. Tesla ranked among the smallest gaps on v1 (7th-smallest of 91) despite having one of the lowest women inventor rates in the dataset (1.9%, in the bottom 12).
Version 2 replaces the single gap metric with the five-sub-index composite described above. Weights and methodology are documented. The composite score is on a 0-100 scale and no longer mechanically rewards companies for being male-heavy across both workforce and inventor pools.
The mechanism behind the index
The composite score above measures what each company looks like at the end of the patent funnel. Understanding why those numbers come out the way they do means tracing the gap through filing, prosecution, citation, and field composition. Our patent gender gap series walks each mechanism:
Patent Gender Gap (hub)
The full series: 8 child pages covering different mechanisms of the gap.
Where women's patents disappear
The prosecution funnel, stage by stage. Each row of our index is a downstream effect of these stages.
The breakthrough penalty
PNAS 2026: the gap is concentrated on unconventional inventions. The mechanism is institutional.
Engineering vs biotech
Field-by-field gap. Why some industries in our index score so differently from others.
Where does your company stand?
We can analyze your patent portfolio against your workforce data and show you exactly where the gap is. Which technology areas are unprotected. Which teams are filing and which are not.
The patent data is public. The insight is knowing where to look and what to do about it.
We also build tools that close the gap. Our scanner surfaces strategic concepts in your codebase so the engineers who would never self-nominate can still have their innovations recognized.
Talk to Us About Your ScoreOr scan your own codebase to see what strategic concepts your engineers are building right now.
Results are strategic concepts, not legal conclusions. Review with a patent attorney.