Xcout IntelSign inSubscribe
Methodology · revised

How Xcout Intel collects its data

Xcout Intel reads job postings and posted pay directly from employers' own careers sites — the applicant-tracking systems (Greenhouse, Lever, Workday, Ashby, SmartRecruiters and others) that power them — not from job boards, recruiters or aggregators. Funding comes from public filings and news reports; US visa figures from Department of Labor disclosure files. This page explains each step and what the figures do not show.

Where hiring data comes from

  • Each employer's jobs are read from its own careers site, through the applicant-tracking system behind it. Job boards, recruiters, search results and aggregator sites are not used.
  • Recruitment agencies and job aggregators, which post other employers' roles, are excluded from the hiring reports.
  • Open postings are re-read on a rolling schedule, so the catalog reflects what each employer has open now, not only what it once posted.
  • Company facts used to group hiring (size, sector, headquarters country, news events) come from Xcout Intel's company profiles.
  • Company websites themselves are read by our crawler, XcoutBot, which reads robots.txt before its first request and leaves at least half a second between requests to a site.

How a posting is counted and dated

  • Open, then closed. A posting stays in the catalog while an employer's feed lists it (up to 90 days old). It is treated as closed once no feed has listed it for five days.
  • First-seen date. A posting is dated when it first appeared; a later re-read never moves that date forward. Where a system publishes a relative date (“Posted 30+ days ago”), it is converted to a date rather than replaced by the day we read it.
  • New means new when we saw it. A weekly count of new postings counts a posting only if it was stored while it was new, so postings discovered late do not back-fill past weeks and make every trend look like a slowdown.
  • Duplicates. The large-employer trackers (Fortune 500, S&P 500, Big Pharma) count each open job once per requisition. For pay, an employer posting the same title at the same range in many locations counts once. In the Hiring Pulse, a role reposted under a new ID counts again.
  • Role and seniority are read from the job title.

The Hiring Pulse panel

The Hiring Pulse measures a fixed set of companies, so a change reflects hiring rather than changes in our coverage.

  • A company is in the panel if it opened new roles in at least 10 of the 12 weeks.
  • The headline compares the last four weeks with the first four. Role families are compared as a share of all postings at the same companies across two eight-week windows; when one company moves a family's figure by more than a quarter (or five points), the figure is published without it and the company is named.
  • Not counted: companies whose feed stopped or restarted during the period (a change of more than 95% either way), and any week above three times a company's typical week, which is a re-listing rather than new hiring.
  • Figures from fewer than 50 companies or 100 postings carry a small-sample tag. The report is recomputed every Monday, and each week is kept as a frozen edition.

Posted pay

  • Posted pay is the midpoint of the pay range an employer posts, from postings that state one. It is not a wage paid.
  • Only annual salaries between $20,000 and $1.5 million a year are used (other currencies converted for that test only). Hourly postings and postings with a monthly, weekly or unstated interval are excluded. Bonuses and equity are not in posted ranges.
  • A figure needs at least 30 postings; a month-over-month change needs at least 100 postings (50 for a seniority cell) from employers present in both months.
  • A posting belongs to the month it was first seen. Each Pay Pulse edition is frozen on the 11th of the following month, after late-arriving descriptions have been read.
  • Country figures use each country's own currency; postings in another currency (often US-remote roles listed abroad) are left out. US states come from the first location a posting lists.

AI skills and AI tools

  • AI skills (Pay Pulse, Hiring Pulse) are read by a classifier from the posting's description and title. Postings that only mention AI in passing — usually the company describing itself — are left out of both sides of a comparison.
  • AI tools (AI adoption) are detected by explicit patterns run against job descriptions, each checked against real postings by hand before it is included. A company counts once per tool. Ambiguous names (“Claude”, “Cursor”, “Gemini”) count only beside another AI tool or an unambiguous product name, and a vendor's own postings do not count toward its own tools. Where a mention cannot be told from ordinary language it is dropped, so the counts are floors, not totals.
  • The AI Rank and Green Rank score companies 0–100 from their own job descriptions — what they hire for, not what they announce — monthly.

Funding, exits and fund closes

  • Funding rounds come from public filings (including US SEC Form D notices of exempt offerings) and from news reports.
  • Events, not articles. Reports about the same round or deal are joined when they concern the same company within 45 days, and the amount used is the one most sources agree on.
  • Disclosed value only. Totals sum the events that publish a figure, and the share that do is printed beside each total.
  • Held out of totals, and counted in the open: venture fund closes filed as company rounds, debt facilities, regulatory filing ceilings from entities with no other trace, and very large figures resting on a single report.
  • VC fund closes (VC fundraising tracker) are announced closes found in news across fifteen markets and twelve languages since January 2024. Several articles about one close count once. Quiet closes with no press coverage are missing.
  • Revenue figures (past $100M ARR) are listed only with their source, labelled for what they are (ARR, run-rate or annualised revenue) and dated.

US visa data

  • From the Department of Labor's Office of Foreign Labor Certification disclosure files: every Labor Condition Application (H-1B, H-1B1 and E-3) and every PERM application since fiscal year 2020. Each case is counted once, under its latest status.
  • An H-1B application is filed before a petition and measures demand, not visas issued — the annual lottery decides those. A PERM certification is the first step of an employment-based green card.
  • Wages are the offered annual wage on certified applications, with hourly and weekly rates annualised.
  • A filing is credited to a Fortune 500 or S&P 500 company only through entities on its own records or names confirmed to be its own; a subsidiary that cannot be connected is not counted, so group figures are, if anything, low. See US visa sponsors.

What these figures are not

  • Not wages: posted pay is the midpoint of an advertised range, and employers that publish pay are not all employers.
  • Not employment: a count of postings is a count of roles advertised, not people hired or employed.
  • Not job-board data: nothing is taken from job boards or aggregators, and agencies posting for other employers are excluded.
  • Not complete: an employer whose careers site we do not read is not in the hiring data, a round with no filing or press report is not in the funding data, and an undisclosed amount is not in a dollar total.
  • The AI-skills pay premium across employers is not the value of an AI skill; it mixes the skill with which employers ask for it. Read it beside the same-employer figure.

How to cite

Name the report, the edition and the page:

Xcout Intel, <report>, <edition>, intel.xcout.app/<path>

For example: Xcout Intel, Pay Pulse, <month year>, intel.xcout.app/insights/pay-pulse/<YYYY-MM>. Cite a frozen edition page where one exists, so the figure still matches after the next edition. For this page: Xcout Intel, Methodology, revised 4 November 2026, intel.xcout.app/research/methodology. Full terms are on the attribution page; every report is listed on Research.