Pipeline
Stop Paying an Agency for Access and Start Paying for Evidence
A recruiting agency earns its fee when the constraint is reach into a population you genuinely cannot address and nobody in-house has the hours to try. It does not earn it when the constraint is that you cannot tell your current applicants apart, which is a screening problem no placement fee touches. Before signing, change what the money buys: ask what was verified about each submission, and require every agency candidate to run the same evidence stage as an inbound one.
The takeThe guarantee period is the clause everybody negotiates and the one that protects you least. A replacement costs the agency a resubmission and costs you a quarter of an empty role, an onboarding that went nowhere and whatever the team did not ship meanwhile. Negotiate the evidence instead. One clause saying every submitted candidate sits the same assessment as everyone else is worth more than the guarantee, and an agency worth hiring will agree to it.
Where Olive fits
Open a role and see what the work shows
Olive is bought by the employer and priced per attempt rather than per seat: an attempt returns six findings a human reviewer wrote, each carrying the excerpt from the session it rests on. An agency-submitted candidate and an inbound one sit the same assignment at the same cost, and both are granted the report.
Rank your shortlistWhen does an agency fee earn itself?
When you cannot reach the people who can do the work, and nobody on your side has the hours to try. That is a real constraint on scarce clinical, licensed, executive and specialist roles, and an agency with a live network in that population is genuinely faster than building one from nothing. Most employers do not have that constraint, and a thin funnel is rarely evidence of it.
Put the fee next to what hiring already costs you. SHRM's 2021 benchmarking of its member organizations puts median cost-per-hire at $1,244 for nonexecutive roles and $8,750 for executive ones, and the distributions are heavily skewed: the nonexecutive mean is $4,683 against that $1,244 median, with a 75th percentile of $4,375 1. Two things about that definition matter here. It already folds third-party agency fees into the total, so an agency hire is not an addition to cost-per-hire, it is most of it. And it excludes the hiring manager's and interviewers' time, which is usually the largest real cost of a loop, so treat it as a recruiting-spend figure.
The useful exercise takes five minutes. Convert the fee on your own agreement into currency, divide it by your own cost per hire, and write the multiple down. Then ask what the multiple is buying. Where the answer is reach into a population your postings cannot touch, it is defensible. Where the answer is a shortlist you could have assembled from your own applicants if anyone had time to read them, the money is solving the wrong problem, and the diagnosis of which case you are in is separating a thin pipeline from a scarce market.
What did the agency actually verify?
Usually a phone conversation and a reformatted CV. The reason is structural: money moves on placement, so what a contingency agreement funds is a submission that clears your screen, and the screen is the part that stopped separating people. Ask in writing anyway, because the answer varies between firms and between recruiters inside one firm, and the agreement itself rarely says which you are getting.
Four questions get you most of the way, and none of them are adversarial:
- Did anyone at the agency meet or speak with this person live, and who?
- Which claims on the profile were checked against a record, and which are the candidate's own account?
- Who wrote the profile text, and did the candidate see it before it was sent?
- What would cause the agency to decline to submit someone?
The third question has aged differently from the rest. A submitted profile is drafted by someone working with the same tools the candidate has, so a pre-qualified summary now sits in the same class of claim as a well-written resume, and it carries about as much information. The last question is the tell. An agency that has never declined to submit anybody is a channel, which is fine as long as you price it that way.
The general lesson about verification is that mechanism beats exhortation. When one orthopaedic surgery residency program checked every publication its applicants listed, misrepresented citations came to 13 out of 1,100, or 1.18%, against 18.0% in the same program's 1999 study and 20.6% in its 2007 study 2. That is one program, one specialty and a uniquely checkable kind of claim, and the three cycles do not share a denominator. The authors offer the PubMed identifier added to the application form as a hypothesis they did not test. The transferable part is still the shape of it: what moved was making the claim machine-checkable, not warning people to be honest. So sort the agency's claims into checkable and impression, and treat the two differently. Where identity itself is the doubt, that has its own procedure, in what to do when the person on the final call may not be the person who did the take-home.
Change what the money buys before you sign
Add one clause: every submitted candidate completes the same evidence stage as an inbound applicant, and the fee attaches to a candidate who completed it. That single line converts the agreement from paying for access into paying for a shortlist you can actually read, and it costs the agency nothing except the candidates it was never willing to stand behind.
Whatever that stage is, make it the same one for everybody. On the corrected meta-analytic table, structured interviews come out top at .42, ahead of job knowledge tests at .40, empirically keyed biodata at .38, work samples at .33 and cognitive ability at .31, while unstructured interviews come in at .19 3. Those are corrected correlations with supervisor ratings, not accuracy rates, and the .42 carries a wide credibility interval, so read the table as a rank order. The rank order is enough for this decision: if an agency submission goes straight into an unstructured conversation with a hiring manager, the fee has bought the weakest pairing on that list.
Three other clauses are worth more attention than they usually get:
1. Ownership of the candidate record. Who owns the relationship if you do not hire, and can you approach the person directly in twelve months. 2. Duplicate submissions. Write down what happens when two agencies submit the same person, and who is deemed to have introduced them, before there is money on it. 3. Off-limits. Whether the agency will place your employees elsewhere, and for how long.
All three are ordinary clauses, and all of them are cheaper to agree before a candidate is in play. Off-limits language in particular restrains who the agency may approach, so run the final wording past counsel before it goes into the agreement.
Should you pay contingency or retained?
Retained for a scarce role, contingency where optionality and speed matter more, and neither for a common role you can fill from your own applicants. Contingency pays on placement, so it rewards volume of submissions, while retained pays for the work, so it rewards attention. A scarce role needs search hours spent on a small population, which is what a retainer funds. On a common role, ask whether the same money spent on outreach would have found the person.
The structure choice mostly determines whose risk it is, so decide what you are trying to move. Contingency puts the risk on the agency and buys you optionality, at the cost of getting whatever surfaces fastest, often the same candidates from three firms at once. Retained puts it on you and buys a defined process, at the cost of paying whether or not it works. Neither structure verifies anything, which is why the evidence clause matters more than the structure does.
The guarantee period is worth one honest sentence. It protects against an agency that submits people carelessly, and it does nothing to improve a submission before you hire, because it pays out only once the hire has already failed. Keep it, negotiate it, and do not let it stand in for evidence.
On Monday, do two things. Pull your last three agency hires and write down, from the record, what the agency verified about each one before submission. Then send the agreement back with the evidence clause in it. If the answer to the first exercise is nothing in all three cases, the second one is not a negotiation, it is the whole value of renewing.
Common questions
What is a normal recruiting agency fee?
Fees are quoted as a percentage of first-year salary and the range varies by market, seniority and structure, so the number that matters is the one on your own agreement rather than a benchmark. Convert it into currency, divide it by your own median cost per hire, and look at the multiple. That figure is the actual question. A large multiple is defensible when it buys reach into a population your own postings and outreach cannot touch, and hard to defend when the same candidates were already in your applicant pool.
Do agencies verify anything about the candidates they submit?
Some do, many do less than employers assume, and the agreement rarely says. Contingency structurally pays on placement, so the product it funds is a submission rather than a verified candidate. Ask directly: who spoke to this person live, which claims were checked against a record, and who wrote the profile. The answers vary enormously between firms and even between recruiters at one firm. Whatever comes back, run your own evidence stage anyway, because verification you did not see is not verification you can rely on.
Should agency candidates skip our assessment?
No, and the exemption is the single most expensive habit in agency hiring. Skipping the stage removes the only comparison that would show whether the agency's submissions are better than your inbound applicants, so the fee can never be evaluated. It also means the candidate with the largest cost attached is the one you know least about. Write the requirement into the agreement rather than leaving it to the recruiter's discretion, and apply it to referred and inbound candidates too, so nobody has a route around it.
Is the guarantee period real protection?
It protects against careless submission and very little else. If a hire does not work out inside the guarantee window, the agency owes you a resubmission while you have lost a quarter of an empty role, the onboarding, and whatever the team did not do meanwhile. That asymmetry is why a guarantee should be kept but never treated as quality assurance. The clause that changes outcomes is the one requiring evidence before placement, because it puts the check before the hire.
When should we build in-house recruiting instead?
When the same roles keep recurring, because the fee is per placement and the in-house cost is per year. One scarce executive search is agency-shaped work. Six similar engineering hires a year is a seat. The other trigger is knowledge: a recruiter who works inside the company learns the product, the levels and the managers, and that accumulates in a way an external submission never does. The mixed answer is common and fine, with agencies used for genuinely scarce roles and the recurring ones brought inside.
References
- 1. SHRM Benchmarking: Talent Access (Selection Criteria, Overall) shrm.org Supports the median cost-per-hire of $1,244 nonexecutive and $8,750 executive with the skew stated, and the definition point that agency fees are inside the figure while interviewer time is not.
- 2. Update on Misrepresentation of Research Publications Among Orthopaedic Surgery Residency Applicants ebi.ac.uk Supports the claim that making a credential machine-checkable moved measured misrepresentation from 18.0% and 20.6% to 1.18%, with the single-program and single-specialty limits stated, the non-identical denominators across the three cycles noted, and the PubMed explanation attributed to the authors as an untested hypothesis.
- 3. Revisiting Meta-Analytic Estimates of Validity in Personnel Selection: Addressing Systematic Overcorrection for Restriction of Range gwern.net Supports the corrected rank order in Table 1 placing structured interviews top at .42 and unstructured interviews at .19, quoted as a rank order among the six procedures named rather than as accuracy rates.
3 sources, numbered by first appearance. 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.