How ApplySift scores an applicant file
You export applicants from your ATS as a CSV. ApplySift reads the name, email, phone and applied-date columns (it recognises Paylocity, BambooHR, JazzHR, Workable, Greenhouse and Lever exports, and asks when it cannot tell), checks each row, and gives it a tier with reasons.
The phone check
Every North American phone number belongs to a block of a thousand numbers, and every block is assigned to one carrier. The assignments are public: the North American Numbering Plan Administrator publishes them daily. ApplySift keeps a copy and looks up the block for each number.
Most fraudulent applications use numbers from carriers such as Bandwidth, Onvoy, Telnyx and Twilio, which provision numbers by API to VOIP platforms. Almost nobody's personal phone is in one of those blocks. That is the single strongest signal in the file, and it costs nothing to check. Numbers in a block assigned to Google (Google Voice) are a weak signal, because real people use Google Voice too. A number that has been ported from one carrier to another keeps its original block, so a ported number can be misread either way; it is rare in practice.
The email checks
- Shared mailbox. Gmail delivers
a.b.c@gmail.com,abc@gmail.comandabc+job@gmail.comto the same inbox. ApplySift normalises addresses before comparing them. - Disposable domain. A short list of throwaway providers.
- Fresh-account provider. Outlook, Hotmail and Live addresses are a weak signal on their own: a new account takes a minute and nothing else. In one real pool of 1,455 applicants, nine in ten Outlook addresses also had a VOIP wholesaler phone.
- Digits and names. Five or more digits in the address, or no four-letter run of the applicant's first or last name, are the shapes generated addresses tend to have.
- One name, several mailboxes. The same first and last name applying under different addresses.
Bursts
If the file has applied dates, ApplySift finds hours that held at least twenty applications and ten times the posting's usual hourly rate. A burst by itself can be a job board syndicating your posting, so it never flags a row on its own; it only raises rows that already have another reason.
Tiers
Each reason has a weight. Likely fraud is a total of five or more (a wholesaler number plus almost anything else), Review is two to four, Clear is below two. The scored file lists the reasons in words next to each row, so you can disagree with any of them. ApplySift never rejects an applicant and never contacts anyone.
What it does not do
It does not check devices, IP addresses, locations or identities, read resumes, verify employment, or keep a list of people. It scores the file you give it and forgets it.