Pipeline
Hiring Funnel Conversion Rates, and Why Yours Dropped
There is no benchmark conversion rate a hiring funnel stage should hit. The published ratios carry applications received in the denominator, and that number moved: when applications per opening double, every rate measured from applications halves with no change in your process or your candidates, while the stage-to-stage rates below the top don't move at all. Report the counts beside every ratio, index the funnel backward from hires as interviews per hire and onsites per hire, and compare your funnel against your own last year, not a cross-industry median.
The takeA falling conversion rate is the most reliably misread number in recruiting. It reads as a warning about quality, and most of the time it is a report about volume, which is why teams answer it by tightening a screen that was never the problem. Hunting for a better benchmark only moves the misreading somewhere else. What ends it is a dashboard where no ratio appears without the two counts that produced it, so anybody reading it can see at a glance which half of the fraction moved.
Where Olive fits
Open a role and see what the work shows
What Olive adds to a dashboard is a count rather than a rate: attempts opened, and six evidenced findings returned on each candidate who finishes one. Ten attempts a month are free, and an attempt is counted when the workspace opens rather than when it finishes.
Rank your shortlistWhy Did Your Conversion Rates Fall?
Almost certainly because the denominator grew. A conversion rate is two counts, and the one at the top is the one that moved: application volume per opening has risen steeply while the number of people any team can interview in a week has not. Divide a roughly fixed numerator by a growing denominator and every rate in the funnel falls, in precisely the pattern a team reads as decline.
Work it through on a funnel with round numbers. A req takes 1,000 applications, screens 40, runs 12 onsites, makes 3 offers and hires 2. Application-to-screen is 4%. Next year the same req takes 3,000 applications and the team has the same hours, so it screens 40, runs 12 onsites, makes 3 offers and hires 2. Application-to-screen is now 1.3%. Every stage below the top is byte-for-byte identical, the hires are the same people doing the same work, and the headline metric fell by two thirds.
A team reading only the ratio concludes the pipeline degraded and responds by raising the bar at the screen, which is the one intervention guaranteed to make the number worse. A team reading the counts sees immediately that nothing below the top changed. Why applications per opening rose so sharply, and how many of them are real is the separate question worth answering, and it belongs upstream of the dashboard rather than inside it.
Which Benchmark Are You Actually Reading?
Usually one that measures something other than what you assumed. Published recruiting numbers are built on definitions that differ by source, so two figures for the same concept are routinely not comparable, and the gap between them is larger than the difference you are trying to detect. Read the definition and the sample before the number, every time, or the comparison is decorative.
Time to hire is the clearest case. SHRM's benchmarking, on data collected in 2021, puts median time-to-fill for nonexecutive roles at 44 days, counted in calendar days from the requisition opening to the offer being accepted, with a 25th percentile of 28 days and a 75th of 73 1. The DHI-DFH vacancy duration measure, built on the JOLTS survey, reported a mean of 28.3 working days in June 2016, and establishments with more than 5,000 employees ran at 59.5 2. Both stop counting at offer acceptance and nothing else about them lines up: one counts calendar days on a single requisition, the other working days across every establishment in a national survey, and that second series is from 2016 and has since been discontinued. Neither is wrong. They are answers to different questions, and averaging them means nothing.
The second trap is the shape of the distribution. In that same 2021 SHRM data, median cost-per-hire for nonexecutive roles was $1,244 against a mean of $4,683 3. When the median and the mean sit that far apart, a single benchmark number describes almost nobody, and your own position inside that spread is the only fact that matters. Ask any published figure two questions before quoting it: what exactly was counted, and how wide was the range around it?
Measure the Funnel Backward From the Hire
Anchor every metric to a hire rather than to an application. Screens per hire, interviews per hire, onsites per hire, offers per hire. These say the same thing a conversion rate says, in a form that does not move when the top of the funnel does, and they translate directly into the resource question a recruiting team is actually managing, which is hours.
The reason they hold still is structural. Applications received is set by the labor market, the posting, the job board's distribution and how many agents are applying on candidates' behalf. Interviews conducted is set by your capacity and your bar. Anchoring to the hire measures the part of the system you control, and a change in it is a change worth investigating.
A workable dashboard for one role family fits in eight numbers: applications received, screens conducted, onsites conducted, offers extended, offers accepted, hires, screens per hire, and onsites per hire. Six counts and two ratios. Publish the counts first and put every ratio directly beneath the two numbers that produced it, so nobody has to reconstruct the fraction to interpret the movement.
One stage deserves separate treatment. If a keyword filter or a matching engine sits between applications and screens, its output is not a conversion rate at all, it is a setting. What replaces a keyword filter once every resume matches is the question underneath a screen rate that suddenly stopped discriminating between anybody.
Compare Your Funnel Against Your Own Last Year
Your own history is the only comparison where the definitions hold still. Same role family, same level band, same sourcing channels, same quarter of the year where seasonality bites. That comparison can carry a conclusion, because the things a cross-industry median cannot hold constant are held constant by the fact that it is the same company hiring the same kind of person.
Split by channel before you read anything into a total. Referrals, inbound applications, sourced outbound and agency candidates convert at very different rates, so a blended rate mostly reports the channel mix. A shift toward inbound will drag the blended number down while every individual channel holds steady, which is the same denominator illusion one level in.
When a rate does move for real, treat it as the start of an investigation rather than the finding. Look at the counts on both sides, then at the channel mix, then at whether the stage's own definition changed, which happens more often than anyone expects when a new applicant tracking system arrives mid-year. Which funnel metrics still mean anything now that candidates use AI covers the stage-by-stage version of that audit.
And if the movement is the result you were hoping for, be careful with it. A rate that improved in the same quarter you changed the process is not evidence the change worked, for the same reason the benchmark comparison failed: too many things moved at once. Proving a hiring change actually worked needs a control, and it is worth setting up before the change rather than after the chart.
Common questions
What is a normal application-to-interview rate?
There is no usable number, and any figure quoted as one is a median over a sample that does not include you. The rate depends on the role, the channel mix, how visible the posting is, whether the job board pushes it, and how much of the volume is automated. A team whose posting is syndicated everywhere will show a far lower rate than an identical team hiring mostly from referrals, with no difference in quality.
Should offer acceptance rate be benchmarked?
It is the one stage ratio worth watching closely, and still only against yourself. Offer acceptance has a small denominator you fully control, so it moves for real reasons: compensation, competing offers, how long the loop took, what the candidate experienced. Track it by role family and by quarter, and read a drop as a prompt to ask the declining candidates why, which is one of the few recruiting questions with a direct answer available.
How many applications per opening is too many?
The number stops being informative long before it stops growing. Past the point where the team cannot read everything received, extra volume adds no signal and costs review time, so the useful measure becomes what share arrives through channels you can act on. Ask how many applications get a real read rather than how many arrive, and manage the first number, since the second is largely out of your hands.
Do these metrics still work if an AI step sits in the funnel?
The counts do; the ratios need a note attached. Any automated stage has a threshold somebody set, so its pass rate reports that setting as much as the candidates. Record the threshold and the version alongside the rate, and treat any change to either as a break in the series rather than a movement in it. Otherwise a vendor's model update shows up on the dashboard as a change in candidate quality.
What single metric would you put on one slide?
Onsites per hire, with the two counts printed beside it. It is stable when application volume moves, it maps directly onto interviewer hours, and it goes up when the bar is being applied late in the loop instead of early. Add hires against plan for context. Anything anchored to applications received belongs in the appendix, where the definition can sit next to it.
References
- 1. SHRM Benchmarking: Talent Access (Selection Criteria, Overall) shrm.org Supports the definitional-mismatch claim for time to hire: median 44 calendar days from req open to offer accepted, 25th percentile 28 days, 75th percentile 73 days.
- 2. DHI Hiring Indicators, Report 28: Mean Vacancy Duration Fell to 28.3 Working Days in June dice.com Supports the claim that a second published time-to-hire figure counts working days on a different endpoint and is not comparable, and that large establishments run far above the national mean.
- 3. SHRM Benchmarking: Talent Access (Selection Criteria, Overall) shrm.org Supports the skew claim: median cost-per-hire of $1,244 for nonexecutive roles against a mean of $4,683 in the same table.
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.