Nobody designs a process to reject good people. It happens anyway, and it happens invisibly, because the engineers you lose never appear in any report. They decline the first call, withdraw in week three, or take the other offer, and your dashboard records a closed requisition either way.
The useful thing about this failure is that it leaves fingerprints in data you already have. You do not need a consultant to find them.
Here are eight signals that your process is filtering out strong engineers, what to measure for each, and the benchmark to compare against. Rise92 sees these patterns constantly across global engineering hiring.
Read this before the benchmarks
Published recruiting benchmarks disagree badly, because they do not count the same things. One 2026 report puts interview-to-offer at 7.0% for small business and 72.2% for enterprise, from a single dataset under a single definition. Both numbers are real.
So use these figures as orientation, not as a target. Your own trend over six months is worth more than any industry median, because it is the only comparison where the definitions are consistent.
The eight signals
| # | Signal | Benchmark to compare against |
| 1 | Offer acceptance slipping | 73% for technical roles, 82% overall |
| 2 | Offers sitting open | Two days to close versus six is the warning |
| 3 | Interview count creeping up | 17.6 technical interviews per hire, already high |
| 4 | Time to offer stretching | 48 days technical, 62 days engineering time-to-fill |
| 5 | Pipeline is almost all inbound | Sourced applicants are eight times likelier to be hired |
| 6 | No stage-level visibility | Application-to-interview 3.6% technical |
| 7 | Interviewers reaching opposite conclusions | Structured loops roughly double predictive validity |
| 8 | Your best engineers would fail your loop | No benchmark, and none needed |
1. Your offer acceptance rate has quietly slipped
The single clearest signal, and the easiest to check.
The benchmark: overall offer acceptance averaged 82% in 2026 per Gem, the highest since 2021. Technical roles run materially lower, at roughly 73% against 84% for business roles, per Ashby’s 2025 data. General sector ranges run 77% to 92%.
What it means: below roughly 75% on technical roles, something upstream is broken rather than something wrong with your offer. Usually it is compensation misalignment discovered late, or momentum lost between the final round and the offer landing.
What to do:
- Discuss compensation ranges in the first conversation, not the last
- Track declines by stated reason, and take the stated reason seriously
- Treat a decline as process feedback rather than a market condition
The trap: attributing a falling offer acceptance rate to “a competitive market” without checking whether your own cycle time moved.
2. Your offers sit open instead of closing
This one almost nobody tracks, and it is among the strongest early indicators available.
The benchmark: how long offers stay open, not just whether they close. One 2026 analysis identified two days versus six as the strongest early warning signal in its dataset.
What it means: an offer that sits for a week is an offer being shopped. The person is waiting on someone else, which usually means you were not their first choice, which usually traces back to how the process felt rather than the number on the page.
What to do:
- Record days-to-decision on every offer
- Watch the trend, not individual cases
- If the average is climbing, look at the four weeks before the offer, not the offer
3. Your loop grew and nobody removed anything
Every process accumulates. A bad hire produces a new stage, and no stage ever gets deleted.
The benchmark: Ashby’s 2026 data shows business interviews per hire rose 36% since 2021 to 11.7, and technical interviews per hire rose 52% to 17.6. Gem puts the average process at 13 interviews, up 42% in three years.
What it means: those figures are already the industry norm, and the norm is the problem. If you are at or above them, you are losing senior people who have other options and less patience.
What to do:
- Count your actual stages, including informal ones
- For each, name what it measures that no other stage measures
- Delete any stage that fails that test
- Three rounds is generally right for senior engineering roles
Worth noting: more rounds does not mean more signal. Four unstructured conversations produce four correlated impressions rather than four independent data points.
4. Your time to offer is drifting past six weeks
Speed is the most controllable cause of losing people, and the most consistently ignored.
The benchmark: technical roles average 48 days from application to offer against 38 for business roles. SHRM 2025 puts overall time-to-fill at 44 days and engineering specifically at 62. Talent Board found 57% of people abandon a process because it took too long.
What it means: at 62 days, you are not competing for anyone who has options. You are competing for whoever is still available after everyone faster has finished.
What to do:
- Measure stage-to-schedule and schedule lead time separately, since total time-to-hire tells you the number without telling you where to push
- Check your reschedule rate, which one dataset put at 14% of interviews, with 18% of scheduling signals needing a person to intervene. Both are invisible in standard reporting, and both consume days
- Decide within a week of the final conversation
5. Almost everyone in your pipeline applied
If your engineering recruitment funnel is fed almost entirely by inbound applications, you are only reaching people who are actively searching. That is a small and non-representative slice of the engineering population.
The benchmark: an applicant sourced directly is roughly eight times more likely to be hired than one who applied through a job board, per Gem. Referrals are hired at a similar multiple and move substantially faster. Personalized outreach gets 18% to 25% response rates against 5% to 8% for templates.
What it means: the strongest senior engineers are employed, performing well, and not reading job boards. They are not in your pipeline because your pipeline cannot reach them, not because they rejected you.
What to do:
- Track hires by source and look at the mix, not just the volume
- Check your existing database, since 46% of hires came from re-engaged people already in the system per Gem 2026
- Build referral flow deliberately, with bonuses paid at 90 days rather than on signature
- For genuinely scarce roles, use off-market sourcing through closed networks
The uncomfortable math: technology roles require roughly 191 applicants per hire against 47 in healthcare. Adding application volume to solve an access problem is expensive and does not work.
6. You cannot say which stage loses the most people
Most teams track the top of the funnel and the bottom, with a black box in between. That black box is where you lose hires.
The benchmark: in 2026, roughly 3.6% of applications reach an interview for technical roles against 4.7% for business, about half the 2021 rate. Interview-to-offer sits at 7.3% for technical roles against 10.4% for business roles, per Ashby’s 109 million application dataset.
What it means: without stage-level numbers, you cannot tell a hard requisition from a broken stage. Both look identical from the ends of the funnel.
What to do:
- Break interview drop-off out by stage and by source
- Compare your own stages against each other before comparing against any industry figure
- Look specifically at the stage where withdrawal, as opposed to rejection, concentrates
The distinction that matters: rejections tell you about your standards. Withdrawals tell you about your process. Most teams only review the first.
7. Two interviewers reach opposite conclusions about the same person
Disagreement is healthy. Wild, unexplainable disagreement means nobody is measuring the same thing.
The benchmark: structured interviews show roughly twice the predictive validity of unstructured ones across meta-analyses, with Sackett and colleagues in 2022 putting the figures at 0.42 against 0.19.
What it means: when interviewers apply private standards, outcomes depend on who happened to be on the panel. Strong people get rejected by one combination and weak people advance through another, and both look like considered decisions.
What to do:
- Write the scorecard before the first conversation
- Ask every person the same core questions
- Collect written scores before any debrief, so the room cannot anchor to whoever speaks first
- Treat disagreement as a prompt to examine the evidence rather than a tie to break by seniority
The failure that undoes everything: overriding a structured assessment late, based on an impression formed in the final conversation. That single move reintroduces the bias the structure existed to remove.
8. Your strongest current engineers would not pass your loop
The cheapest diagnostic available, and the most uncomfortable.
How to run it: take your three best engineers. Walk them through your current process honestly, stage by stage. Ask whether they would have passed, and whether they would have stayed in the process at all.
What it means: if the answer is no, your process is not measuring the qualities that actually produce good engineers at your company. It is measuring something adjacent that correlates with interview performance.
What to do:
- Ask your best hires what nearly made them drop out
- Ask your recent declines what the process felt like, and mean it
- Rebuild the scorecard from what your strongest people actually do well
Why this works: you already have the ground truth. You know who is good, because you have watched them work for years. That is better evidence than any benchmark.
What this actually costs
The cost of filtering out strong engineers rarely appears as a line item, which is why it persists.
| Hidden cost | Where it shows up |
| Extended vacancy | Delivery slips, existing team absorbs the load |
| Settling for the available option | Replacement cost at roughly 200% of salary |
| Interviewer time | Roughly 21 interviewer-hours per technical hire at small companies |
| Reputation | A small market talks, and a poor process travels |
The interviewer time figure deserves attention. A bloated loop is not free rigor. It consumes most of a working week per role from the people least able to spare it, and the additional stages are usually the ones producing the least independent signal.
When you compare sourcing options, compare total landed cost including vacancy and internal time rather than headline fee. A one-time fee that looks expensive in month one is frequently cheaper than three extra months of vacancy. The comparison sits in transparent pricing.
How to run the diagnostic this week
Four numbers, pulled from your existing system:
- Offer acceptance for engineering roles, last 12 months
- Median days from first contact to offer
- Percentage of hires by source: inbound, referred, sourced
- Withdrawal count by stage
Then one conversation with the last three people who declined you.
That is enough to locate the problem in most cases. The pattern is usually obvious once the numbers sit next to each other, which is the entire value of doing it.
Frequently asked questions
Look at withdrawals rather than rejections, since rejections reflect your standards while withdrawals reflect your process. If people are leaving mid-process, if your offer acceptance rate sits below roughly 75% for technical roles, or if your strongest current engineers say they would not have passed, the process is the problem. Rejection volume alone tells you nothing either way.
Roughly 73% is the reported technical benchmark against 82% overall, so engineering running below business roles is normal rather than alarming. Below 75% typically indicates late compensation misalignment or lost momentum between final round and offer. Your own six-month trend matters more than the absolute figure, since published benchmarks use inconsistent definitions.
Industry averages have risen to roughly 17.6 technical interviews per hire, and that average is itself the problem rather than a target. Three substantive rounds is generally right for senior engineering roles, with each testing something the others do not. If you cannot state what a stage uniquely measures, remove it.
Break withdrawal out by stage and compare stages against each other before comparing against any external benchmark. Concentrated withdrawal at one stage usually points at either an unpaid assignment that is too large or a scheduling gap where momentum died. Both are fixable in a week.
What to take from this
The choice is narrower than the marketing suggests. Establish whether the person is an employee or a genuine contractor, establish whether you own a local entity, and five of the six options usually eliminate themselves.
Then model total cost per country rather than comparing provider fees, because statutory variation dwarfs fee variation.
And keep the sourcing decision separate from the employment decision, because conflating them is how companies end up paying an ongoing percentage for something a one-time arrangement would have covered.
To work through which structure fits your situation, get in touch.
The pattern underneath all eight
Every signal here describes the same thing: a process optimised to reject rather than to attract.
Rejection-optimised processes are easy to build by accident. Every stage feels like rigor, every additional interview feels like diligence, and the cost lands entirely on people who never become your data. The engineers who withdraw do not fill in a form explaining why.
Fixing it is mostly subtraction. Fewer stages, faster decisions, compensation discussed early, and a written standard everyone scores against. None of that requires new tooling.
If your numbers come back healthy and you still cannot hire, the constraint is access rather than process, and that needs a different solution entirely. To work out which one you are facing, feel free to get in touch.
