Pipeline
Should You Skip the Resume Screen and Go Straight to a Work Sample?
Don't drop the resume screen entirely; demote it. Send a work sample to every applicant only while applicants times review minutes fits the hours available: fifteen hours and a twenty-minute review reaches forty-five people, so forty on an engineering req clears the line and four hundred on a marketing one does not. Above the line, keep the resume as a documented check against the posted requirements, stop using it to rank, and put the sample ahead of the interviews. Cap the task and keep selection rates for whichever gate decides.
The takeThe resume screen survives because it is the one gate nobody has to write anything down about. That is not thrift, it is a cost kept off the books: the step carrying the most measured disparity is also the step that leaves you nothing to show when someone asks how it was applied. A universal work sample looks like the way out until the review hours arrive. My read is that most of the volume making it unaffordable sits upstream, in duplicate applications and ones that never matched the location or the authorization you posted, and that clearing those buys back more hours than any screening rule you could write.
Where Olive fits
Open a role and see what the work shows
Olive is priced per attempt rather than per seat, so the cost of a work sample tracks the number of people you actually send it to, which is the figure this arithmetic turns on. An attempt returns six evidenced findings on one candidate as an input to your decision, never a ranking, and ten attempts a month are free, so a pilot can run beside your current screen and be compared against it.
Rank your shortlistShould you drop the resume screen entirely?
No. Demote it. A resume screen does two separate jobs: it checks whether someone meets the requirements you posted, and it sorts people by how impressive the document reads. The second job is the one that has stopped working and the one you cannot defend in writing. Keep the check, drop the sorting, and move the work sample ahead of the interview loop.
That distinction is doing all the work in this decision. Checking a stated minimum (the license, the years in a regulated function, the shift availability) is fast, verifiable, and the same for every applicant. Forming an impression from a document is none of those things, and it now runs on text that a model helped write. When every resume in the stack reads well, the impression half of the screen is measuring polish and nothing else.
Dropping the step outright is a different proposition, and it usually fails on arithmetic rather than on principle. Every applicant who receives a work sample costs you review minutes that a resume never did, and those minutes come out of a fixed recruiting week. The teams that make a universal work sample work have low applicant counts, a short instrument, or both.
So the answer has a threshold in it rather than a yes or a no. Below the threshold, send the sample to everyone and read the resume afterward as a record of dates and employers. Above it, you still need something that reduces four hundred applicants to a reviewable number, and the resume, used as a documented check rather than a ranking, is the cheapest instrument you already own.
What does the resume screen cost you in compliance terms?
More than most teams assume, precisely because nobody scores it. The Uniform Guidelines define a selection procedure as any measure used as a basis for any employment decision, and name "informal or casual interviews and unscored application forms" inside that definition 1. A resume screen is a selection procedure. It carries a test's job-relatedness and adverse-impact obligations while producing none of a test's documentation.
Two rules decide how exposed that leaves you. The Guidelines say that when the total selection process shows adverse impact, the individual components should be evaluated for it, and that when the total process does not, enforcement agencies will not usually expect a component-by-component look 2. That is enforcement discretion, not a shield. In private litigation the Supreme Court held in Connecticut v. Teal that a nondiscriminatory bottom line neither prevents a prima facie case nor defends one, where a discrete step acted as a pass-fail barrier 4.
The resume screen is also where the measured disparity has actually shown up at scale. In a correspondence experiment sending more than 83,000 fictitious applications to jobs posted by 108 of the largest U.S. employers, distinctively Black names reduced the probability of employer contact by 2.1 percentage points, with the top quintile of firms accounting for close to half of the lost contacts 5. That gap is produced at the stage where a human reads a document and forms an impression.
That does not make a work sample automatically safer. It makes the resume screen the step in your process with the largest gap between how consequential it is and how little of it is written down. If you keep it, write down what it checks; if you replace it, the obligations move with the gate rather than disappearing.
Do the arithmetic: where does the trade flip?
At the point where applicants times review minutes exceeds the hours you can spend on the req. That is the entire equation. Budget fifteen hours a requisition and a twenty-minute review, and a universal work sample reaches forty-five people. The same fifteen hours covers four hundred and fifty resumes at two minutes each, which is why the answer differs by function rather than by conviction.
Run your own numbers rather than these, but the shape holds:
| Req | Applicants | Work-sample review each | Review hours | Verdict |
|---|---|---|---|---|
| Senior engineering | 40 | 20 min | 13 | Send it to everyone |
| Analyst, mid-market | 120 | 20 min | 40 | Only with a shorter instrument |
| Marketing generalist | 400 | 20 min | 133 | Not reachable: three weeks on one opening |
| Marketing generalist | 400 | 6 min | 40 | Reachable if the sample is cut to fit |
The candidate side of the ledger has the same arithmetic and no budget line. A fifty-minute exercise sent to four hundred applicants asks for more than three hundred hours of unpaid time, and the people who finish are disproportionately the people with a free afternoon. That is a selection effect you now own and have to keep records on, which is a real cost even though it never appears in a spreadsheet.
So the lever at high volume is the instrument, not the coverage. Cutting a sixty-minute case to a twenty-minute one with a six-minute review changes the reachable population by an order of magnitude, and a short well-aimed task discriminates better than a long vague one. If you are weighing this against paying for coverage instead, the up-front assessment math for high-volume funnels runs the same comparison with the vendor cost included.
One number to check before any of this: how many of those applicants are real. Applications per opening have risen sharply, and a chunk of the increase is the same person applying to forty reqs. Deduplicating and filtering to genuine geographic and authorization fits can move a req from the unreachable row to the reachable one without touching the resume at all.
What replaces the resume screen when volume is high?
Three things, in this order: knockout questions on requirements you actually posted, a short structured application scored identically for everyone, and a work sample sent to whoever clears both. The resume stays in the file as a record you read later, when you are checking employment dates rather than forming a judgment. What it stops doing is deciding who advances.
Knockouts have to be genuinely job-related and answerable with a fact, not an impression. Authorization to work, a required license, physical location against a stated on-site requirement, the specific regulated experience the role sits inside. Anything a candidate could reasonably answer either way is not a knockout, it is a preference, and preferences belong in the review rather than in an automatic gate.
Be careful with the AI-skills line in particular, because it is the requirement most likely to be written vaguely and applied inconsistently. "Comfortable with AI tools" screens nobody and defends nothing. Writing an AI-skills requirement that is specific enough to act on is the difference between a knockout you can explain and one you will quietly abandon.
The structured application does the work the resume was pretending to do. Four to six questions, a fixed word cap, the same rubric for every applicant, scored before anyone sees a name. It is slower per applicant than skimming a resume and enormously faster than a work sample, which puts it exactly where the arithmetic needs something. If your keyword filter used to do this job, the reason it stopped working is that every application now matches it.
How do you keep the work sample defensible as the first gate?
Write it to closely approximate the work, cap the time, run one case per req, and keep selection-rate records for the gate you just created. Content validity under the Guidelines turns on how closely the procedure's manner, setting, level and complexity resemble the work situation 3. A sample built around a hobby project resembles nothing in the job, and it is now the only thing standing between an applicant and a rejection.
Records are the part teams forget when they move a gate. The four-fifths rule treats a selection rate below eighty percent of the highest group's rate as evidence of adverse impact 2, and it applies to whatever is doing the selecting. Moving the decision from a resume screen to a work sample moves that obligation rather than retiring it, and unlike the resume screen, a work sample gives you a rate you can actually compute.
Three design constraints keep the gate honest. Publish the time cap and hold to it, because an uncapped task selects on available hours. Offer the same case to every applicant on the req, so the comparison means something. Name the accommodations route in the invitation itself rather than in a policy page, since accommodation requests on an AI-based assessment arrive late when the process makes people ask for them.
The honest limit: a work sample tells you what someone produced, not who produced it. With an assistant open, a finished deliverable is weak evidence about the person, which is the whole reason work samples are being redesigned around the process rather than the artifact. If the sample is your only gate, it needs to capture something the artifact does not carry on its own: the questions asked before generating, what was checked, what was turned down. See how Olive measures this
Common questions
Is a work sample legally safer than a resume screen?
Not automatically, but it is easier to defend. Both are selection procedures under the Uniform Guidelines 1, so both carry job-relatedness and adverse-impact obligations. The difference is evidence: a work sample has a written rubric, a fixed task, a selection rate you can compute, and a content-validity argument resting on how closely it resembles the job 3. A resume screen usually has none of that, which is why it is the weakest link in most defensibility stories rather than the riskiest instrument in principle.
What applicant volume makes a universal work sample unaffordable?
Whatever number makes applicants times review minutes exceed your review hours. At a twenty-minute review and fifteen hours per requisition, the ceiling is about forty-five applicants. At a six-minute review it is about one hundred and fifty. Compute yours before deciding anything, because the same instrument that is comfortable on a senior engineering req is three weeks of one person's time on a high-volume marketing one. The lever at high volume is shortening the sample, not skipping people.
Should the work sample be paid if it replaces the resume screen?
Pay for anything long enough to be work, and keep the first gate short enough that it isn't. A twenty-minute exercise sent to everyone is a selection procedure; a full-day project sent to everyone is unpaid labor plus a completion problem that selects on free time. If the task genuinely needs hours, move it later in the process, send it to a small number of finalists, and pay for it. That sequencing is cheaper for you and fairer to applicants than paying at the top of the funnel.
Do you still need the resume at all?
Yes, as a record rather than a filter. You need employment dates, titles, and gaps for reference checks, offer decisions, and in regulated functions for the file itself. What changes is when you read it: after the work sample, when you already have evidence about the work, so the document informs questions instead of setting an impression that colors everything after it. Reading it later also removes the ranking step that carries the most exposure and produces the least writing.
Does a work-sample-first process create its own adverse impact?
It can, and you will only know if you measure it. Completion is not random. Timed tasks select on available hours, reliable internet, and quiet space, and the drop-off is not distributed evenly across applicants. Track the selection rate at the gate by group the way the four-fifths rule contemplates 2, separate non-completion from failure in your records, and publish the time cap so people can plan. A gate you can compute a rate for is more fixable than one you cannot.
Can we run both, resume screen and work sample, without lengthening the process?
Yes, if the resume step becomes a check rather than a review. A documented pass-fail against posted requirements takes under a minute per applicant and adds no calendar time, because it runs the day applications arrive. The delay in most processes is not the screen, it is the wait between stages. Sending the work-sample invitation automatically on the day someone clears the check usually shortens the overall process even though a step was added.
References
- 1. Uniform Guidelines on Employee Selection Procedures, 29 CFR 1607.16 (Definitions) ✓ ecfr.gov Paragraph Q defines a selection procedure as any measure, combination of measures, or procedure used as a basis for any employment decision, explicitly including informal or casual interviews and unscored application forms.
- 2. Uniform Guidelines on Employee Selection Procedures, 29 CFR 1607.4 (Information on impact) ✓ ecfr.gov Section 4C: where the total selection process shows adverse impact, individual components should be evaluated; where it does not, agencies will not usually expect component evaluation. Section 4D: a selection rate under four-fifths of the highest group's rate is generally regarded as evidence of adverse impact.
- 3. Uniform Guidelines on Employee Selection Procedures, 29 CFR 1607.14 (Technical standards) ✓ ecfr.gov Section 14C(4): where a procedure samples a work behavior, its manner, setting, level and complexity should closely approximate the work situation, and the closer the resemblance the stronger the content-validity basis.
- 4. Connecticut v. Teal, 457 U.S. 440 (1982) ✓ law.cornell.edu A nondiscriminatory bottom line neither precludes employees from establishing a prima facie disparate-impact case nor provides the employer a defense, where a discrete component operated as a pass-fail barrier.
- 5. Systemic Discrimination Among Large U.S. Employers ✓ nber.org More than 83,000 fictitious applications to jobs posted by 108 of the largest U.S. employers: distinctively Black names reduced the probability of employer contact by 2.1 percentage points, with the top quintile of firms responsible for nearly half of the lost contacts.
5 sources, numbered by first appearance. Every one was opened and checked against the claim it carries. How Olive sources claims
General guidance for hiring teams. What works at one company and one volume may not transfer to yours.
Olive assesses how a person works with AI. It does not detect AI-written documents, and it never produces a score, a ranking, or a match percentage for a person. Candidates read the same report the employer reads.