H-1B Filings in Private Equity Research: A 2026 Protocol

Executive Summary

A Labor Condition Application (LCA) record can corroborate three things about a services company: the legal name under which it files, the specialist roles it asked to fill, and the places where it said the work would happen. It cannot establish headcount, hiring growth, revenue or ownership. This report sets out an evidence protocol for private equity origination analysts who screen US technology-enabled and professional-services companies. It is not immigration advice or a salary comparison.

The reason is structural. The Department of Labor (DOL) reviews each LCA within seven working days for completeness ^[1], and certification is only a precursor to a separate petition adjudicated by U.S. Citizenship and Immigration Services (USCIS). An LCA also counts requested positions: in fiscal year (FY) 2025 DOL received 575,063 applications but 934,671 requested positions ^[2]. Researchers using LCA micro-records caution that approvals do not translate one-for-one into H-1B grants ^[3], and the Federal Reserve Bank of Boston notes that an LCA cannot be tracked through the application process ^[4].

The second source, the USCIS H-1B Employer Data Hub, counts petition first decisions ^[5], excludes pending petitions ^[6] and shows employers by only the last four digits of a tax identification number. For FY2025, USCIS reports 406,348 approved beneficiaries in one report ^[7] and 328,185 approvals in a quarterly table of another ^[8]. The two are on different bases, and the passages read do not reconcile them; the petitions report itself cautions that the total number of approved petitions may exceed the number of individuals who receive H-1B status.

The protocol therefore treats a filing as a clue, not a fact. A match counts as a corroborated legal-name clue only when the OFLC employer name, a state registry record and the company's own site agree. Worksites are intended work geography, and DOL's own examples include computer engineers sent to customer locations ^[9], so a listed address does not prove an owned office. Four inferences are rejected: filings equal employees, filing wages equal payroll, a worksite equals a branch, and a filing equals ownership. Where headcount is needed, sources built to measure it exist, such as the employee count that public registrants must disclose ^[10], Census Bureau business statistics ^[11] and Bureau of Labor Statistics (BLS) employment counts covering more than 95 percent of US jobs ^[12].

A reproducible worksheet closes the protocol: source file, fiscal year, case number, extraction date, identity key matched, use state and the remaining question for each row. Only visa-sponsoring firms appear in these records at all, so absence is uninformative.

Introduction and Background

Private equity origination teams screen long lists of small and mid-sized services companies, and many publish little. A Labor Condition Application is one of the few government-held records that names a private employer, a job title, an occupational code and a physical address in a downloadable file. That makes it tempting to read as a proxy for size, hiring momentum or capability. This report examines when that temptation is justified and when it is not.

Search coverage for the topic mostly serves other readers. One finance-career guide, updated in August 2026, presents H-1B wage disclosures as a salary source for private equity, venture capital and banking roles ^[13], and concedes that the filings do not cover every employee and skew toward certain roles ^[13]. Those readers want pay benchmarks. A mandate screen asks a different question: does the record help confirm who the company is, what specialist roles it recruits and where it says it operates, and what must stay unproven?

The report is published by Docket, which provides sourced target research for private equity origination teams. The protocol is written so that any team can run it from free public files, and it makes no claim about any real company. All named examples are fictional and labelled.

Five terms that must stay separate

• Legal employer: the entity that files the LCA and, later, the petition. Under DOL rules the entity that files the H-1B petition is deemed the employer of that nonimmigrant ^[14]. The OFLC layout calls the field the legal business name of the employer submitting the LCA ^[15].
• Brand: the trading name a customer recognises. A separate "Doing Business As" field exists, so a brand may not appear in the legal-name field.
• Client worksite: a customer location where the filer places a specialist. The layout flags placement with a secondary entity ^[15].
• Primary place of employment: the physical location where the work is actually performed ^[14].
• Corporate parent: the entity that owns the filer. No LCA field establishes this. Parent relationships come from registrant subsidiary lists ^[16] or Legal Entity Identifier (LEI) data, which publishes direct and ultimate parents when they hold LEIs ^[17].

A fictional running example (Hypothetical Example)

Consider Northbridge Data Services, Inc., a fictional 150-person analytics services firm that trades as "Northbridge Analytics". The Office of Foreign Labor Certification (OFLC) file shows employer names "NORTHBRIDGE DATA SERVICES INC" in one state and "NORTH BRIDGE DATA SERVICES LLC" in another. The screen question is whether either record corroborates the target. The honest answer is "possibly one, after registry and site checks".

The Filing Sequence: LCA, Petition and First Decision

Three artifacts matter, and each has its own unit, date and owner.

Step 1: the certified LCA

The employer first submits an LCA to OFLC, attesting that it will comply with the material facts and labor condition statements ^[18]. Under 20 CFR 655.730 the LCA states the number of nonimmigrants sought ^[19], the wage rate on an hourly, weekly, biweekly, monthly or annual basis ^[19], and all intended places of employment ^[19]. An LCA may be submitted no earlier than six months before the beginning date of the employment ^[19], so a filing date can precede any hire by up to half a year, and an initial period of authorized employment cannot exceed three years ^[20]. DOL's review is light by design: an earlier Government Accountability Office (GAO) report described it as limited to searching for missing information or obvious inaccuracies ^[21].

Step 2: the USCIS petition

USCIS lists the first step as submitting the LCA to DOL for certification ^[22]. DOL's H-1B Advisor adds that USCIS approval is a separate document ^[18]. Cap-subject petitions also depend on a prior registration, and selection pertains only to eligibility to file the petition ^[23]. The LCA and the petition are separate documents, so the two records are linked but their counts need not match.

Step 3: the first decision

The Employer Data Hub reports USCIS first decisions on petitions for initial and continuing employment, a third unit distinct from LCA positions and from petitions filed.

Why the distinction matters

A single LCA may request multiple positions ^[24], and one request can cover several positions to be filled by H-1B workers ^[4]. A Congressional Research Service (CRS) report states that certified-position totals do not reflect the actual number of workers that employers petition for or admit ^[24]. A National Bureau of Economic Research (NBER) chapter notes that companies can request more workers than they intend to hire ^[25]. A recent working paper treats LCAs as an early view of intended sponsorship, because certification precedes any USCIS petition ^[26], and a Brookings analysis notes that employer-application counts did not indicate how many visas were approved ^[27]. A Boston Fed analysis found that only 137 employers averaged more than 10 positions per request over its study period ^[4], so request size varies widely by employer, and Kerr and Lincoln describe LCAs as measuring demand regardless of whether a visa is realized ^[3]. An analyst who counts rows and calls the result "employees" is combining three units.

Locating and Reading the OFLC Disclosure Files

Where the files are

OFLC publishes disclosure data on its performance page as Excel workbooks organised by federal fiscal year, October 1 through September 30. The FY2026 files retrieved on 2 November 2026 cover determinations issued between October 1, 2025 and June 30, 2026 ^[28]. Each data set is cumulative for the fiscal year ^[28], with unique records identified by OFLC case number.

Access note. OFLC statements here reflect the performance page as read on 2 November 2026. Analysts should download files in a browser, record the exact filename and release quarter, and re-verify before relying on a link, because file names and quarterly labels change.

Layout drift and the year that matters

Beginning in FY2020 the record layouts were substantially different from prior fiscal years ^[28]. A community documentation page for an open-source sponsor database, which cites the DOL source, says column layouts drift between years ^[29]. The same page covers FY2020 through FY2026-Q3 ^[29]. It is a community source, used here only to confirm that downstream builders meet the same problem.

What is and is not in the file

OFLC states that cases may be absent when still being processed with no final determination ^[28], and that a small percentage of determinations may change in later quarterly releases because of appeal or redetermination ^[28]. Typographical errors and anomalies may originate with employers or data entry, and a Boston Fed review of LCA data flagged incorrect worker counts, occupational codes and wage values ^[4].

A conflict worth preserving. The performance page names Federal Employer Identification Numbers (FEINs) among personal information that can keep data out of the files ^[28]. Yet the FY2026 LCA layout lists a field described as the employer's FEIN ^[15], and DOL's FY2025 statistics aggregate employers by the unique FEIN submitted ^[2]. The community page says its employer FEIN field exists only from FY2024 on ^[29]. These statements do not agree on what a downloaded file contains. The protocol resolves the conflict empirically: open the header row of the file actually downloaded, record whether the FEIN column exists and is populated, and never assume either state. The community page says the missing FEIN in earlier years is why its builders need an elaborate employer-identity resolution step ^[29].

Worksites are in a second file

The main LCA file describes the first worksite. Additional locations sit in a separate worksites file whose layout accepts up to ten physical locations with wage information per application ^[30]. A researcher interested in geography must join the two files on case number and treat the result as one-to-many.

Field-Level Claims Table

Table 1 below lists the LCA fields most useful to a screen. For each it separates what the field literally says from what a screen may infer and what it may not.

Two points deserve emphasis. First, no field supports a count of employees; the only count in the record is a count of requested workers. Second, the wage field is a statement about what the filing promises for one position, so it must not be averaged into a payroll estimate. The next section moves from DOL to USCIS data.

Cross-Checking With USCIS Data and Registry Records

What the Employer Data Hub offers

The Hub can be queried by fiscal year, employer name, city, state, ZIP code and North American Industry Classification System (NAICS) code ^[5]. As retrieved on 2 November 2026 it covered FY2009 through FY2026 (quarter 4) ^[5]; coverage advances as new quarters are added, which is why the retrieval date belongs on the worksheet. Its design differs from OFLC's in ways that matter:

• Unit: first decisions on petitions, with counts of workers approved ^[6].
• Exclusions: subsequent decisions, such as on an appeal or revocation ^[6].
• Date basis: the date USCIS adjudicated the filing, not the date received ^[6].
• Identity key: only the last four digits of the tax identification number appear ^[5], and the displayed name is the most common spelling per unique tax ID ^[6].
• Roll-up needed: one employer with multiple addresses in a fiscal year has multiple rows ^[6].
• Geography: the state is the one in the employer's mailing address, not necessarily where beneficiaries work ^[6].

USCIS warns that data entry and petitioner errors may produce a missing or erroneous employer name, tax ID, state, city or ZIP code ^[6]. Its Characteristics report adds that electronic records do not give the cities or states where H-1B workers are employed ^[7].

Registry and identifier checks

Employers must also make each certified LCA available for public examination within one working day of filing ^[33], which makes the employer's own public access file a source of record for a given LCA. A state registry confirms that a legal entity exists, but registries vary in what they show. Delaware's public search describes its intent as providing general entity details ^[34] and notes that additional information can be obtained for a fee ^[34]; the retrieved page appears dated, so analysts should read the current terms of whichever registry they use. For public registrants, Regulation S-K requires a list of subsidiaries and their jurisdictions of incorporation ^[16]. The LEI is a unique 20-character alphanumeric code ^[31] and each LEI represents only one entity ^[31], which helps separate similarly named entities when an LEI exists.

Comparing the three record families

Table 2 below compares the evidence families by unit, date, scope and identity key.

The table shows why no single family should carry a conclusion. OFLC offers the richest role and geography fields but no unit that equals people. USCIS offers decision outcomes but weak identity. Registries and company records supply identity, but only for entities that publish. A match is credible when at least two families agree on identity and none contradicts it.

Worksites, Third-Party Placement and Services Models

What DOL means by a worksite

DOL's worksite fact sheet defines the place of employment as the worksite or physical location where an H-1B nonimmigrant worker actually performs the work ^[9]. The geographic area of intended employment is the area within normal commuting distance ^[14]. So an LCA can describe a commuting area while the file lists one address. The fact sheet is marked as revised in July 2008 and states that it lacks the force and effect of law, so the regulations control.

Exceptions that make a listed location unsafe to read as an office

• Short-term placement: up to 30 workdays in a one-year period ^[35], extendable to 60 in defined cases ^[35].
• Non-worksite locations: certain temporary locations are not considered a worksite and need no new LCA ^[9].
• Client-site work: DOL's own regulation treats a computer engineer working on projects or accounts at different locations as having worksites at those locations ^[14].
• Third-party staffing: the Department of Homeland Security describes a worker staffed to a third party as contracted to fill a position in the third party's organization ^[36]. The 2024 final rule, effective January 17, 2025 ^[36], eliminated the itinerary requirement ^[36], requires a bona fide position in a specialty occupation ^[36] and requires an Internal Revenue Service tax identification number for the US employer ^[36].

A client-site address on an LCA says where the filer intended to place someone, and in a services model that is often a customer's building. It is not evidence of an owned branch.

IT services and the H-1B dependence category

Searchers who look for "IT services company H-1B dependence" are usually asking about the regulatory category of an H-1B-dependent employer. The category is defined by tests that vary with size: for 25 or fewer full-time-equivalent employees, more than seven H-1B nonimmigrants ^[37]; for 26 to 50, more than 12 ^[37]; and for larger employers, H-1B nonimmigrants equal to at least 15 percent of full-time equivalents ^[37]. The LCA records the employer's own answer on dependence ^[19].

For a screen this field is narrow. It records a status the filer reported at one point in time, based on a ratio the screen cannot rebuild because the denominator is full-time-equivalent employees. It is not a quality, risk or valuation label, and no dependence ratio can be computed from public LCA counts. GAO's earlier analysis noted that DOL data do not show what share of each employer's workforce is H-1B ^[38], and that the top H-1B-hiring employers included firms using staffing arrangements ^[38].

Use States and Prohibited Inferences

Five use states

Table 3 below defines the states into which each matched row should be classified.

The last row is the main selection limit. Only employers that sponsor visa workers produce LCA records, so absence is uninformative. Listed companies, the easiest to match, account for only about 13 percent of approved petitions in one NBER study ^[39].

Four rejected inferences

• LCA count equals employees. LCA files count requested workers: one 2026 working paper reports over 8 million requested workers across about 4.5 million certified LCAs ^[26]. Requests are not hires.
• LCA wage equals company payroll. The wage field concerns the first worksite for the position described, and the individual wage data behind a prevailing wage are not a public record ^[33].
• Worksite equals company-owned branch. See the exceptions above.
• Filing equals ownership. The filer is the employer for H-1B purposes; parentage requires registry or LEI evidence. After a merger or spin-off, a successor can keep transferred H-1B workers without new LCAs, so a name change may signal a corporate event, not a new business ^[19].

The Reproducible Worksheet Protocol

A screen that cannot be rerun is not evidence. Record each row as follows:

1. Source file and release: the exact OFLC filename, fiscal year and quarter, plus the Hub retrieval date.
2. Row key: the OFLC case number, the unique identifier assigned to each application.
3. Extraction date and analyst initials, with no names of workers or attorneys.
4. Fields captured: employer string, city and state, job title, SOC code, worksite state, wage unit, case status, received and decision dates.
5. Identity key matched: registry file number, LEI or FEIN state observed in the file header.
6. Use state: one of the five states in Table 3.
7. Remaining question: what evidence would move the row to a higher state, such as "confirm legal name in registry" or "check whether the name changed after a merger".

Volume trends can be shown only as filings per fiscal year for the matched legal employer, labelled as filings and never as hiring growth. Petitions filed in a quarter are not necessarily adjudicated in that quarter, so series across record families misalign by design.

Data Analysis and Evidence

National scale of the LCA record

DOL's FY2025 Selected Statistics give these program-wide figures, covering H-1B, H-1B1 and E-3 visa classes:

• Applications certified: 561,505 ^[2].
• Applications denied: 3,921, and withdrawn: 29,393 (same table).

Against the 575,063 applications received and 934,671 positions requested cited earlier, certified applications are fewer than applications received, and positions requested exceed applications. The FY2026 year-to-date sheet, with data as of June 30, 2026, shows 426,659 applications received through the third quarter; the fourth-quarter column was blank in that release, so FY2026 cannot be compared with FY2025 as a full year. These numbers describe the national filing flow, not any company.

National scale of the petition record

USCIS's FY2025 Characteristics report shows median annual compensation of $133,000 for approved beneficiaries ^[7]. For FY2026 registration, USCIS received 343,981 eligible registrations and selected 118,660 unique beneficiaries ^[7].

Two USCIS publications for FY2025 give different approval totals: 406,348 in the Characteristics report and 328,185 in Table 1 of the Petitions report, against 456,725 petitions filed. The reports carry different extraction dates, and USCIS warns that the total number of approved petitions may exceed the number of aliens provided H-1B status ^[8]. Analysts should cite the exact table and never splice figures across the two reports.

What the numbers say about screening

LCA positions requested nationally (934,671) exceed LCA applications received (575,063), and both exceed the petition approvals shown in either USCIS source. Counts at the three stages are not interchangeable, which is the quantitative reason the protocol forbids filing-based headcount ratios. Research on public firms finds H-1B users larger on average with higher growth rates ^[39], but that is a population-level correlation in approved-petition data for listed companies. It does not let an analyst infer a particular firm's growth from its filings. Name matching adds noise: one study found 52 separate firm names for one large employer's location ^[39] and noted that a single firm often files separate petitions ^[39].

Case Studies and Real-World Examples

Hypothetical Example: screening Northbridge Data Services

All names here are fictional. An analyst downloads the FY2026 LCA file and finds a record with employer string "NORTHBRIDGE DATA SERVICES INC", job title "Data Engineer", a computer-occupation SOC code, a first worksite in a metropolitan area, a requested start date and wage unit "Year". A second record, "NORTH BRIDGE DATA SERVICES LLC", appears in a different state.

1. Registry check: the state registry shows the first string as an active corporation with a principal office matching the target's website. The second is an unrelated limited liability company. The first moves to corroborated legal-name clue; the second is ambiguous shared name and is excluded.
2. Role check: the careers page lists data engineering. The job title becomes a role-presence clue, and the analyst notes that the posting date and the LCA received date need not coincide.
3. Worksite check: the first worksite is a client's building in another city. The analyst records intended geography and the secondary-entity flag, and does not list a branch.
4. Hub check: the Employer Data Hub shows rows for the name across two fiscal years at two addresses. The analyst rolls the rows up, records the last four tax ID digits, and notes that these approvals count workers approved, not employees.
5. Headcount: the analyst writes "no headcount evidence from H-1B sources" and turns to sources built for it.

The output is a few worksheet lines and one open question, not a size estimate.

Implications and Future Directions

Sources that measure headcount

Public registrants must describe human capital resources including the number of persons employed ^[10]; one large IT services registrant's annual report states approximately 351,600 employees as of December 31, 2025 ^[40]. The Census Bureau's Statistics of US Businesses provides annual statistics for establishments with paid employees by geography, industry and enterprise size ^[11], and BLS publishes employer-reported employment and wages ^[12] for the pay period including the 12th day of the month ^[12]. Form 5500 filings are a disclosure document that also serves as a data source for other agencies and the private sector ^[41], but they count plan participants, so they are a partial proxy at best.

Alternative data in origination

Alternative data in private equity is a young practice. An arXiv paper lists key new hires and web traffic among alternative data types ^[42] and observes that private equity and venture capital adopted alternative data more recently than hedge funds ^[42]. A trade interview describes using alternative data at the beginning of deal sourcing ^[43], and a Bain survey of mergers and acquisitions practitioners reported that early adopters used generative artificial intelligence for sourcing and screening ^[44]. A 2019 PitchBook survey of 391 venture capital investors, not representative of private equity, found that 38 percent used data to source all investments ^[45]. No quantified benchmark of LCA data's predictive value for private equity targets was found. Any claim of predictive value should be treated as unproven.

What may change

• Rules: USCIS uses a weighted selection process starting with FY2027 ^[46]. It also describes an additional $100,000 payment condition on certain petitions under a September 2025 proclamation ^[22]. Rules like these can change filing behaviour, so a year-over-year filing change for any employer should not be read as a company-health signal; confirm the current status of each rule on USCIS pages.
• Data formats: layouts and file lists change, so the worksheet records the layout used.
• Method: research teams can log these checks as structured evidence so that a later analyst can reproduce them.

Frequently Asked Questions (FAQs)

How do I use H-1B data for due diligence on a target?

Use it as one identity and role clue among several. Pull the OFLC file for the relevant fiscal years, match the employer string to a registry record, check the company site, and classify each row with the five use states. Do not translate filings into headcount.

How do I analyze H-1B filings for a company?

Follow the worksheet: record file and release, case number, extraction date, fields, identity key, use state and the remaining question. Join the main LCA file to the worksites file on case number and roll up Employer Data Hub rows by employer and fiscal year.

What is labor condition application data useful for in company research?

It supports legal-name candidates, specialist-role presence and intended work geography. It does not give hires, payroll, ownership or branches.

Is H-1B sponsorship a signal of company growth?

Not reliably. Research on listed firms finds H-1B users larger and faster growing on average ^[39], but LCAs measure demand regardless of whether a visa is realized ^[3], and approvals do not translate one-for-one into grants ^[3].

What does H-1B dependence mean for an IT services company?

It is a regulatory category with size-dependent thresholds ^[37]. It is not a measure of company quality and cannot be rebuilt from public counts.

Can public data support private equity target screening?

Yes, for identity and market mapping, when each source is used for what it measures. Headcount comes from filings and statistical programmes, and H-1B records supply role and location clues.

Where do I find H-1B petition data by employer?

The USCIS Employer Data Hub, with the limits listed above. LCA data come from OFLC files.

Is H-1B data an alternative data source for deal sourcing?

It can be one input to a sourcing universe, with the caveats above and no benchmark of predictive value.

Conclusion

A Labor Condition Application record is a government-held statement by a named legal employer about a position, a wage and a place. Read literally, it can corroborate a legal name, a role and an intended geography. Read loosely, it invents headcount, hiring momentum and branches that the record never asserted. The filing sequence, the unit differences between OFLC and USCIS data, the worksite exceptions and the identity limits of both datasets all point to the same discipline: classify each row into a use state, keep the source, fiscal year, row, extraction date and open question on a worksheet, and send size and ownership questions to sources built to answer them.

For origination teams, the value of the protocol is not that it adds a signal. It prevents a misleading one. Only sponsoring firms appear, the numbers describe requests and not people, and some official statements about the files conflict. A screen that records those limits is easier to defend and easier to rerun.