Why these numbers deserve a closer look
Lever's three fraud figures describe a forecast, an incident cost and a pilot flag rate, not your team's confirmed fraud. You may suspect that some applicants are not who they say they are, but you have no clear count, no line in the budget, and little time to build a process from scratch. A fraud check sounds sensible until someone asks what it costs and what it actually catches.
Lever, the hiring software company, has published a post about its Spring 2026 release that uses three numbers to make the case that fraud is a serious issue for recruiters. Those numbers are useful starting points. They are also easy to over-read. Each one measures something different, and only one of them comes from a count of applicants.
This post goes through the three figures in the order Lever presents them. For each, it explains what the number measures, what it does not show, and what question to ask before the number shapes a purchase decision. It is not a review of Lever's product, and it does not tell you how common fraud is in your market. For that, you need your own data.
Figure 1: one in four applicants by 2028
The first figure is a forecast. Lever's post states:
> "By 2028, Gartner predicts that 1 in 4 applicants could be fake."
Source: Lever's Spring 2026 release post
What it measures. This is a prediction, attributed to Gartner, about a future share of job applicants. It is not a count of fake applicants at any company today, and it is not a rate measured in your industry or your region. The word "could" matters. The figure describes a possible future, not a condition you can read off your own pipeline this month.
What it leaves out. The post does not explain how Gartner reached the estimate. It also does not define what counts as a "fake" applicant. For a staffing team, that definition matters a great deal. A fake applicant might be someone using a stolen name, a made-up employment history, a credential that cannot be verified, or a person who sends someone else to take an interview. Each of those calls for a different response, and the single number does not separate them.
How to use it. Treat the forecast as a reason to ask a question, not as a reason to buy. The question is simple: over the last six months, how many applicants had details we could not verify, or turned out to be someone other than who they claimed? If you cannot answer that, your first problem is visibility, and a tool may or may not solve it. The forecast also names one target year, 2028, and says nothing about the years in between, so check the date and the context before you repeat the number to anyone else.
A more useful habit is to track your own numbers over time. Count the suspicious or unverifiable applicants each quarter and watch the trend. If the count rises, the forecast becomes relevant to your team. A flat count describes your recent experience; it does not rule out rising risk by 2028. That trend line tells you more about your own risk than any industry figure can.
Figure 2: $28,000 for one incident
The second figure is a cost. Lever's post argues that the cost of one fraud incident goes well beyond money:
Lever puts the financial cost at $28,000 for one incident and describes an efficiency loss of up to 20–30%, alongside lost time, productivity, and team impact.
Source: Lever's Spring 2026 release post
What it measures. These figures describe financial cost and lost team capacity. Read the wording closely. "Up to" sets a ceiling, not a typical result. The post does not say whether $28,000 is an average, a median, or the figure from a single case, and it does not name a study or a method behind either number.
What it leaves out. The cost of an incident depends on where in the process it is caught. An applicant screened out early costs much less than one who gets through interviews, receives an offer, starts work, and is discovered weeks later. Your cost will also depend on your placement model. A staffing firm that puts a worker at a client site may carry costs that a single internal employer does not, such as client relationships and replacement placements. The post does not address any of this, so the figure is a starting point for your own estimate, not a substitute for it.
Turning it into budget math. Start with the number of incidents you think you face in a year. Be honest, and include the ones you only discovered later. Multiply that count by your own cost for one incident, which should include recruiter hours, manager time, onboarding you had to undo, and any re-hiring. Then compare the total with what a fraud check would cost you over the same year. If you estimate one incident a year, the $28,000 figure gives you a yardstick for that year. If you estimate several, the yardstick multiplies. In both cases, the result is only as good as your count.
It is worth working through this for a couple of scenarios. In one, you catch every incident at the screening stage and your cost stays low. In another, incidents slip through to the offer or the first week on the job and the cost climbs. A fraud check is most likely to earn its place in the second scenario, and the only way to know which scenario you are in is to look at your own history.
The team-efficiency point. The efficiency loss may weigh more heavily on a small team than on a large employer. When a team has three recruiters, one person losing a week to a bad hire takes a bigger bite out of total capacity than it would in a department of thirty. That is a general point about small teams, and it is a reason to take the efficiency figure seriously. Ask what your own team loses when a fraud case takes a manager's week, and write that number down.
Figure 3: 30 flagged out of 121 in one week
The third figure comes from a test, not a forecast. Lever's post reports that Fraud Signals flagged 30 suspicious applicants out of 121 during one week in a pilot with its own TA team.
Source: Lever's Spring 2026 release post
What it measures. The pilot counted how many applicants one feature, called Fraud Signals, marked as suspicious during one week, for one team, using a pilot setup. Thirty flags out of 121 applicants is about one in four. That is a concrete picture of how a flagging feature behaved in one real pipeline, which is more than the forecast gives you.
What it leaves out. The post says the applicants were flagged as suspicious, not confirmed fraudulent. That difference is the most important thing to notice. The source does not say how many of the 30 flags turned out to be real fraud, how many were false alarms, what signals triggered each flag, or what the team did with the flagged applicants. A flag is a prompt for review. Without the outcome of those reviews, you cannot tell whether the feature found fraud or found applicants who simply looked unusual on paper.
The sample is also narrow. It covers one week, one hiring team, and Lever's own recruiting group. Your applicant mix, role types, and volume may be very different. A staffing team that places people across many roles and client sites may see a pattern that looks nothing like a single internal team's week. Before you compare any flag rate with your own, ask whether the two groups of applicants are similar in role mix, seniority, and how they arrive. If they are not, the pilot number tells you little about what your team would see.
What the flag rate means for your workload. Even if you ignore the accuracy question, the flag rate tells you about review time. If a tool flagged about one in four applicants at your volume, someone on your team would have to look at those applicants. For a small team, the real cost of a flagging feature may be the hours spent reviewing flags, not the license itself. Estimate how many applicants you see in a typical week, apply a rate like the pilot's to see what that would mean in review hours, and decide whether your team can absorb it.
How to use it. Ask any vendor, Lever included, for the outcome data that the headline leaves out. You want to know how many applicants were flagged, how many flags were later confirmed as fraud, how many were false alarms, and how long each review took. If a vendor cannot or will not answer those questions, the pilot number cannot establish accuracy.
Putting the three together
Each figure answers a different question, and none of them answers the one a small team actually faces.
- The Gartner forecast asks whether fraud is likely to grow.
- The $28,000 figure asks what one incident costs, if you can count it.
- The pilot asks how often a flagging feature marks applicants, and how much review that creates.
The question you need answered is whether a fraud check pays for itself in your hiring. That takes your own incident history, your own cost estimate, and your own review capacity. Lever's three numbers can shape the questions, but they cannot supply the answers.
Be wary of stacking the figures into a single argument. A forecast plus a cost plus a flag rate can sound like proof, but each number rests on different assumptions and none of them was measured on your team. Use them to frame the decision, then replace each one with your own figure as soon as you have it.
Related guides: Accessible job application forms: 5 checks · Applicant data requests: 4 privacy office guides · Applicant drug testing: 6 mistakes to fix before hiring.
Questions to bring to a vendor call
Whether you are looking at Lever or any other fraud check, these questions will tell you more than any single headline:
- What exactly is flagged, and what signals trigger a flag?
- Who reviews a flag, and how long does that review usually take?
- What share of flags were later confirmed as fraud? What share were false alarms?
- Can a flag lead to automatic rejection, or does a person always make the final call?
- How does the check handle applicants who are genuine but unusual, such as people with gaps in their work history or people who recently changed their name?
- What does the check do to the applicant experience, and how long does it add to the hiring timeline?
- What does the price cover, and what extra costs should we expect as our volume grows?
- How does the vendor handle applicant data, and who inside our company can see the flags?
Pay attention to vague answers. A vendor that describes its outcomes in general terms may be hiding weak results, or may simply not track the numbers you need. Either way, you should know before you sign.
Decide in advance how your team will talk to an applicant whose record is flagged. A written script for that conversation keeps recruiters consistent and reduces the chance that a false alarm damages a good applicant relationship.
A simple way to decide
For a small staffing team, a sensible process runs in six steps. None of them requires a large budget or a dedicated fraud desk.
- Count what you already know. Look back over the last six months of placements and near-misses. Note any applicant whose identity, work history, or references did not hold up, and the stage at which you found out.
- Price one incident for your team. Add recruiter hours, manager time, undone onboarding, and any re-hiring. Use your own numbers, then compare them with $28,000 to see how far apart they are.
- Map the review path. Decide who looks at a flagged applicant, what they check, and how long they have to decide. A flag that nobody reviews adds cost without adding protection.
- Ask for outcome data. Request confirmed fraud counts, false-alarm rates, and review times from any vendor. The pilot headline alone does not answer these.
- Trial on one role first. Run the check on a single open role for a few weeks. Compare what it flags with what your recruiters would have caught on their own, and log the review time.
- Decide on the budget line. If the trial catches problems your current process misses, and your incident cost is high enough to justify the price, the check fits. If the trial mostly flags people your recruiters would have cleared in minutes, it may not.
A fraud check is also only one layer. It does not replace interviews, reference calls, or a recruiter's judgment, and it will not catch every case. Treat it as one tool among several, not as a guarantee.
If the answer is no, a cheaper change to your process may cover most of the risk. Verifying references directly, confirming identity before any offer is extended, and training recruiters to notice inconsistencies between a resume and an interview are all low-cost steps. They will not catch everything, but they cost little and they build habits that a tool cannot replace.
What to take away
Lever's three figures are three different kinds of evidence. Used together, they make a reasonable case that fraud deserves attention. Used alone, any one of them is weak support for a purchase.
For a small staffing team, the useful question is not whether fraud exists in the abstract. It is whether your own hiring has carried costs you can count, and whether a check would reduce them by more than it costs. Start with your own history, price your own incidents, and test any tool on a real role before you commit the budget. The numbers on Lever's pages can help you ask the right questions. Your own records will tell you whether a fraud check fits.