USCIS H-1B Employer Data: Approvals and Denials by Company

Official USCIS H-1B employer files as rows: fiscal year, employer, initial and continuing approvals and denials, totals and denial rate, NAICS sector with its name, tax-ID last four, city, state and ZIP. Filter by employer, state, city, industry, ZIP or petition volume. FY2009-FY2023, UTC dates.

This actor reads the H-1B Employer Data Hub files published by U.S. Citizenship and Immigration Services — one CSV per fiscal year, no key and no account — and turns them into filtered, sorted dataset rows. One published row is one employer in one fiscal year with the counts of workers USCIS first decided on: initial employment and continuing employment, approvals and denials. On top of the published columns every row gets the totals, the number of decisions and the denial rate, the plain name of the NAICS sector and the full name of the state, so a dataset reads without a code table. Fiscal years 2009 through 2023 are published as files; a fiscal year runs 1 October to 30 September. No API key, no proxy, no browser.

Need it tailored to your workflow? Missing a field, a filter, another source or a different output format? Open an issue and describe what you need — we read every request and adapt the actor.

Use cases

• One company's H-1B record — employers: ["microsoft"] with a fiscal year: every entity whose filed name contains that text, with approvals, denials and the denial rate. Case, punctuation and spaces are ignored, so walmart finds WAL MART ASSOCIATES INC. For journalists and analysts after a news story.
• The biggest sponsors in a state — states: ["CA"], minTotalApprovals: 50, sort: "approvalsDesc": the long alphabetical tail of one-petition employers disappears and the employers that sponsor at scale come first. For recruiters, relocation services and sales teams.
• Where decisions went against the employer — naics: ["54"], minTotalDenials: 1, sort: "denialsDesc", or maxDenialRate: 0.05 for the other direction. For immigration lawyers and in-house mobility teams.
• A five-year sponsor profile — mode: "employerTotals" with five fiscal years: one row per employer with yearCount, firstFiscalYear, lastFiscalYear and the summed counts. For investor and policy research.
• Sponsors in one city — cities: ["Seattle"]: who files H-1B petitions from a city, largest first. For job seekers and local press.
• A watch on a state or an industry — onlyNew: true: each run writes only rows that were not written before, so you never pay twice for the same employer when you re-run a watch with wider filters, take the next maxItems rows of a long list, or add a fiscal year to it. The published files do not change, so a schedule on the same input writes rows until the list is exhausted; a fiscal year USCIS publishes later has to be chosen in fiscalYears once this actor lists it.

Input

Every field is optional. A run with nothing but fiscalYears returns the top maxItems employers of that year by approvals.

Reference

Fiscal years

2023, 2022, 2021, 2020, 2019, 2018, 2017, 2016, 2015, 2014, 2013, 2012, 2011, 2010, 2009.

A fiscal year runs from 1 October of the previous calendar year to 30 September (FY2023 = 1 October 2022 to 30 September 2023). Each year is a separate file and a separate download. FY2023 is the newest year published as a file; a value outside this list is refused by the platform before the run starts, with the list of allowed values.

NAICS sectors

naicsLabel in every row carries the sector name from this table. Code 99 covers a large share of the smaller employers: the file itself does not know their industry.

States and territories

AL Alabama · AK Alaska · AZ Arizona · AR Arkansas · CA California · CO Colorado · CT Connecticut · DE Delaware · DC District of Columbia · FL Florida · GA Georgia · HI Hawaii · ID Idaho · IL Illinois · IN Indiana · IA Iowa · KS Kansas · KY Kentucky · LA Louisiana · ME Maine · MD Maryland · MA Massachusetts · MI Michigan · MN Minnesota · MS Mississippi · MO Missouri · MT Montana · NE Nebraska · NV Nevada · NH New Hampshire · NJ New Jersey · NM New Mexico · NY New York · NC North Carolina · ND North Dakota · OH Ohio · OK Oklahoma · OR Oregon · PA Pennsylvania · RI Rhode Island · SC South Carolina · SD South Dakota · TN Tennessee · TX Texas · UT Utah · VT Vermont · VA Virginia · WA Washington · WV West Virginia · WI Wisconsin · WY Wyoming · PR Puerto Rico · GU Guam · VI U.S. Virgin Islands · AS American Samoa · MP Northern Mariana Islands.

The state is the employer's mailing address on the petition, not the worksite. Some rows of the oldest files carry no state at all; those rows can never match a state filter.

Sort orders

• approvalsDesc — most approved workers first (the default; the published files are alphabetical).
• denialsDesc — most denied workers first.
• denialRateDesc — highest share of denials first; rows without a single decision go last.
• employer — employer name, A to Z.
• fiscalYear — newest fiscal year first, then most approvals.

Examples

One company in the newest published year

{ "mode": "byYear", "fiscalYears": ["2023"], "employers": ["microsoft"], "sort": "approvalsDesc", "maxItems": 25 }

Every Microsoft entity in FY2023 with its approvals, denials and denial rate.

The biggest sponsors in California, with chosen columns

{
  "fiscalYears": ["2023"],
  "states": ["CA"],
  "minTotalApprovals": 50,
  "sort": "approvalsDesc",
  "maxItems": 25,
  "fields": ["employer", "state", "totalApprovals", "denialRate"]
}

Californian employers with 50 or more approved workers, largest first, narrowed to four columns (plus the always-kept ones).

Denials in technical services

{ "fiscalYears": ["2023"], "naics": ["54"], "minTotalDenials": 1, "sort": "denialsDesc", "maxItems": 25 }

IT consulting, engineering and research employers with at least one denied worker, most denials first.

Five fiscal years of one employer group, summed

{ "mode": "employerTotals", "fiscalYears": ["2023", "2022", "2021", "2020", "2019"], "employers": ["amazon"], "maxItems": 20 }

One row per Amazon entity with the five years added up, yearCount and the first and last year with decisions.

Hospitals and clinics above a size threshold

{ "fiscalYears": ["2023"], "naics": ["62"], "minTotalApprovals": 10, "sort": "approvalsDesc", "maxItems": 25 }

Universities in one state, 20 rows at a time

{ "fiscalYears": ["2023"], "naics": ["61"], "states": ["MA"], "onlyNew": true, "maxItems": 20 }

Run it again and the next 20 Massachusetts universities arrive, without paying for the first 20 twice.

Output

A real row from a cloud run of 2026-10-08 (byYear mode, FY2023, employers: ["microsoft"]):

{
  "fiscalYear": 2023,
  "sourceRow": 18720,
  "employer": "MICROSOFT CORPORATION",
  "matchedEmployer": "microsoft",
  "taxIdLast4": "4442",
  "naics": "51",
  "naicsLabel": "Information: software publishing, telecoms, media",
  "city": "REDMOND",
  "state": "WA",
  "stateName": "Washington",
  "zip": "98052",
  "initialApprovals": 39,
  "initialDenials": 0,
  "continuingApprovals": 1950,
  "continuingDenials": 45,
  "totalApprovals": 1989,
  "totalDenials": 45,
  "totalDecisions": 2034,
  "denialRate": 0.0221,
  "sourceMode": "byYear",
  "url": "https://www.uscis.gov/sites/default/files/document/data/h1b_datahubexport-2023.csv",
  "fetchedAt": "2026-10-08T20:02:55.704Z"
}

Dataset views: H-1B employers (employer, place, industry name, the approval split, denials, rate), Decisions in detail (the four published counts plus the computed ones), Employer totals over the selected years.

Use it from code / agents

curl -X POST "https://api.apify.com/v2/acts/yadroo~uscis-h1b-employers/run-sync-get-dataset-items?token=$APIFY_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{"fiscalYears":["2023"],"states":["CA"],"minTotalApprovals":50,"sort":"approvalsDesc","maxItems":25}'

import { ApifyClient } from 'apify-client';
const client = new ApifyClient({ token: process.env.APIFY_TOKEN });
const run = await client.actor('yadroo/uscis-h1b-employers').call({
    fiscalYears: ['2023'],
    employers: ['microsoft'],
    maxItems: 25,
});
const { items } = await client.dataset(run.defaultDatasetId).listItems();

import os

from apify_client import ApifyClient
client = ApifyClient(os.environ["APIFY_TOKEN"])
run = client.actor("yadroo/uscis-h1b-employers").call(run_input={
    "mode": "employerTotals",
    "fiscalYears": ["2023", "2022", "2021"],
    "employers": ["amazon"],
    "maxItems": 20,
})
items = client.dataset(run["defaultDatasetId"]).list_items().items

MCP: add https://mcp.apify.com to Claude / Cursor / any MCP client and call the yadroo/uscis-h1b-employers tool with the same JSON input. Two things help an agent: fields keeps the answer small (["employer","totalApprovals","denialRate"]), and mode, fiscalYears, states, naics and sort are validated by the platform before the run starts, so a wrong value comes back as an error with the allowed values instead of a wrong dataset.

Pricing

Pay per event: $0.001 per run start + $0.002 per dataset row. The start event is charged on every run, including a run that ends with no rows. Tier discounts apply to the row event (Bronze −10 %, Silver −20 %, Gold and above −30 %); the start event is flat.

Typical runs at the full price:

• a company lookup in one fiscal year, 5 rows: 0.001 + 5 × 0.002 = $0.011;
• a state shortlist, 25 rows: 0.001 + 25 × 0.002 = $0.051;
• the default 50 rows: 0.001 + 50 × 0.002 = $0.101;
• a five-year totals run, 20 rows: 0.001 + 20 × 0.002 = $0.041 (reading five files costs time, not rows).

Every row written to the dataset is charged, and only rows: an employer filter that matched nothing writes no marker row, so you pay the start event alone and the status message tells you what did not match. Reading a file always costs the same few seconds whatever maxItems is, so filter first, sort second and keep maxItems small. The run stops as soon as your Maximum cost per run is reached, keeps every row it already wrote and says so in its status message ("Stopped at your spending limit: N rows delivered").

Limits & FAQ

FY2024 and later are not in here. Only fiscal years 2009 to 2023 are published as files. Newer years exist inside the interactive data hub, whose pages and export buttons the site's robots.txt closes to automated clients, so this actor does not read them and offers no workaround.

Speed and the site's pace. robots.txt of uscis.gov asks for a crawl delay of 10 seconds, and this actor keeps it: one request per 10 seconds to the site, retries included. Everything else is fast. Measured in the Apify cloud on 2026-10-08: a one-fiscal-year run takes 4-7 seconds end to end (the FY2023 file is 2.1 MB with 33,332 published rows, FY2022 3.9 MB with 59,983 rows, and each one downloads and parses in well under a second); a run over all 15 published fiscal years read 15 files and wrote 913 rows in 145 seconds — about 6 files and 380 rows a minute, nearly all of it waiting politely between requests. Nothing was throttled or blocked in that run. The largest input the schema allows therefore fits comfortably into the 900 s run timeout, and a run that still would not fit stops by itself, keeps the rows it has written and says so ("Stopped before the run timeout: N rows saved" — verified in the cloud with a 60 s run timeout: 40 rows of 4 fiscal years kept).

What the numbers are and are not. They are first decisions on petitions — appeals, revocations, reopened cases and pending petitions are not in them, so a case later won on appeal still counts as a denial here. They count workers, not petitions in flight. The address is the employer's mailing address, not the worksite, so a national employer filing from one headquarters appears in one state only. 99 means the file does not know the industry; some rows of the oldest files have no state and no ZIP, and some rows of the newest files have no employer name at all — those rows are real decisions and are kept, with empty values. Data-entry spellings in the source are kept as filed (SAGINAW VLY ST UNIV), which is why the employer filter matches a part of the name. The files themselves are not uniform — the newest ones name the columns "Initial Approval", the older ones "Initial Approvals", and the quoting differs too; the actor reads both and stops with the header it saw if a column is ever missing.

Repeated lines and employer identity. A few lines of the published files repeat an employer, an address and the same counts exactly; every published line becomes its own row and sourceRow tells them apart. In employerTotals mode employers are grouped by their name (ignoring case and punctuation) plus taxIdLast4, so

AMAZON.COM SERVICES LLC

Not personal data. The rows carry an employer name plus the last four digits of a tax id — an employer identifier. No beneficiaries, no salaries, no contact details.

H-1B is not a green card. This is the record of H-1B petition decisions. It says nothing about permanent residence, PERM or the green-card lottery.

Salaries are elsewhere. Wages and job titles come from the Department of Labor's own disclosure data, a different dataset that is not in this actor.

Empty results. A filter that matched nothing is a successful run with no rows and a status message naming what did not match (an employer value, a city) — never a silent empty dataset. With onlyNew a run that finds nothing new says so and how many matching rows earlier runs already delivered. A bad value (a ZIP that is not digits, a denial rate above 1) stops the run at once with a message saying what to change, and an off-list value in one of the enum fields is refused by the platform before the run even starts.

Questions or problems: open an issue on the Issues tab of this actor.

Made by Yadroo. Sibling actors: federal-register-documents · sec-edgar-filings · data-gov-datasets · bls-release-calendar · himalayas-remote-jobs