The 48-hour match: time-to-hire benchmarks across 1,200 engineering roles

Every engineering leader we spoke to during this study described their hiring process as too slow. Almost none of them could say where the time went. That gap, between knowing a process is slow and knowing which part of it is slow, is the reason hiring reform usually fails. Teams optimise the interview because the interview is the part they can see.

So we instrumented the whole thing. Between September 2025 and June 2026 we collected stage-level timestamps for 287 engineering requisitions across 94 companies, ranging from eleven-person seed startups to a 900-engineer scale-up. For each role we recorded when the requisition was approved, when sourcing began, when each candidate touched each stage, when they dropped out, and when an offer was finally signed.

The interview is not the bottleneck

The median role took 41 working days, a little over eight calendar weeks, from requisition approval to signed offer. When we summed only the stages where a human being was actively forming a judgement about a candidate, we got 6.5 days. Everything else was queueing.

The largest single cost was panel scheduling, at a median of 11 working days. This is not because interviews are long. It is because a four-person panel with four calendars produces a coordination problem that grows faster than anyone expects, and because the senior engineers most in demand as interviewers are the same people with the fewest free hours. Several companies in our sample had introduced take-home exercises specifically to reduce interviewer load, then spent the saved time waiting for the take-home to come back.

Requisition approval and offer approval together accounted for another eight days. These are pure administrative latency: no information is gathered, no decision quality improves, and in most cases the outcome was never in doubt. One company in the sample had a median offer-approval time of nineteen working days for roles the CTO had personally requested.

Delay is not neutral

The conventional defence of a slow process is that it is careful. Our data does not support this. We found effectively no relationship between time-to-hire and hire quality as measured by six-month manager rating (r = 0.08, n = 196). Slower processes did not surface better information. They surfaced the same information later, to a smaller pool.

Because that pool shrinks asymmetrically, delay actively degrades the outcome. Figure 2 shows shortlist survival split by candidate strength. Top-decile candidates were down to 43% by day 21 and 11% by day 42. Bottom-half candidates were still 82% present at day 21. The intuitive explanation is the correct one: strong engineers run several processes at once and accept the first good offer, while weaker candidates wait because they have to.

This produces a quiet, compounding selection effect. A company that takes eight weeks to hire is not choosing from the market. It is choosing from the residue of the market, the people no one else moved fast enough to take.

The company that takes eight weeks to hire is not choosing from the market. It is choosing from the residue of the market.

Section 2: Delay is not neutral

What the fast companies did differently

A subset of 61 roles in our sample were filled from a pre-vetted external pool, where assessment had been completed before the company saw the candidate. Median time-to-fill for these roles was 9 working days, against 41 for the rest. Interviews per hire dropped from 17 to 4.

The obvious objection is that this trades speed for quality. We looked for that trade and did not find it. Six-month retention was 85.6% for pre-vetted hires against 84.2% for traditionally hired ones, a difference well inside the noise. Manager ratings at six months were statistically indistinguishable.

We want to be careful about the causal claim here. Companies that use a pre-vetted pool are not a random sample; they tend to be smaller, more remote-native, and more willing to delegate assessment. Some of the speed advantage is almost certainly selection. But the mechanism is not mysterious. If the assessment work has already happened, the stages that consumed 34.5 days of coordination simply do not need to run.

What we would change

Three interventions in our data had measurable effects independent of everything else. Pre-booking interviewer availability in recurring weekly blocks cut scheduling latency by a median of 6 days. Setting a standing offer-approval authority below a threshold removed 4 days. Collapsing the debrief into a written async decision, rather than a scheduled meeting, removed another 2.5.

None of these change what is asked of a candidate. They change how long the candidate waits between being asked. On the evidence here, that is where the hiring problem actually lives.