I fixed the funnel everywhere. A 40% pay gap beat me at home.
Time-to-hire fell from 45 days to 30 everywhere the blocker was process. Israel, my home site, moved zero, because the blocker was a salary band. You can't interview your way through a 40% comp gap. The pipe-or-wall test I should have run in week one.
I found out in month four, with a beautiful funnel, a happy scoreboard in every other region, and a wall at home I'd spent a quarter treating like a pipe.
Time-to-hire fell from 45 days to 30. And Israel, the site I'm from and the one I most wanted to win, moved nothing.
The overhaul worked. Where the problem was process, we won. At home the problem was money. I burned four months finding that out. This post is so you burn one week.
600 applied. 5 signed.
The Data Engineering Group's funnel, top to bottom:
600 applied
150 attempted the home assignment
30 reached interviews
10 final round
7 offers
5 accepted
The shape was brutal: 600 in the top, 5 signatures at the bottom. Time-to-hire was 45 days, long enough that the strong candidates signed a faster offer while we were still booking round three.
(A 2026 check, added when I reworked this post: SHRM's benchmarking puts the all-roles average time-to-hire at 44 days. We were dead average. Dead average is how you lose everyone holding a second offer.)
Every leak had a name.
- The home assignment was a wall. Good Israeli candidates with better-known logos in their inbox wouldn't give up a weekend for it. Nothing about a weekend of free work says we respect your time.
- The first screen ran in English. That quietly told local candidates how seriously we took their site. You send signals you never decided to send. Candidates read all of them.
- Every hire needed a sign-off from France. Delay, plus a thumb on the scale for French candidates. Both compounding. Both invisible on any dashboard.
None of these were secrets. Everyone in the loop could recite them. They were just nobody's job to fix, which is the natural state of every broken process: visible to all, owned by none.
Four boring fixes
Impressive fixes are what you reach for before you've diagnosed anything. We did four boring things, and one thing that wasn't about hiring at all.
A manager conversation moved to the front. Right after the recruiter screen, someone who could talk shop reached candidates while they were still deciding whether we were serious. Not in round four, after the home assignment had already scared them off. Candidates decide early whether you're real. We'd been showing up late with the only people who could prove it.
We wrote an actual engineering brochure. Our interviewers were each improvising the same pitch fifteen times a week, grinding their morale to powder. One good artifact did that job once. It gave them their evenings back.
Candidates got two doors. They could choose a take-home or an on-site coding session. Same bar, two doors. That one change roughly doubled the strong people who finished the assessment stage. The candidates the take-home filtered out weren't weaker. They were busier. In this market, busier usually means better.
We built feedback guides. We stopped ghosting rejected candidates. Back then I sold it with the folk line about one of them telling ten friends.
A 2026 check corrected that pitch: the line is customer-service folklore from 1980s consumer-complaint studies, not hiring research. The measured version is stronger. Talent Board's 2024 candidate-experience benchmark found resentment at a record high, 25 percent in tech in North America, nearly double the all-industry average. Candidates given specific feedback were over 50 percent more willing to refer other people in.
The number on the slide
To get any budget, I quit arguing in adjectives and put a number on it.
I sat with the product managers. We worked out the revenue lost every month a seat stayed empty. I walked into the room with that figure instead of a feeling.
A CFO who shrugs at "we're understaffed and morale is suffering" moves fast when the cost of doing nothing is the number on the slide. Translating engineering pain into money was the single highest-leverage thing I did in the whole program.
It had nothing to do with hiring. Nobody teaches it to engineers. Every director needs it weekly.
Every site climbed. Home moved zero.
The scoreboard looked good.
- Time-to-hire fell from 45 days to 30.
- Spain turned into a genuine success: hired well, grew more diverse, outgrew its office.
- The fully-remote team grew past my forecast.
- France added 20 percent headcount.
- Candidate satisfaction climbed into the high sevens. Soft number, I know. The direction was real.
Everywhere the problem was the process, fixing the process worked.
Israel went nowhere. We streamlined every step. Zero delta.
The blocker was never the process. It was the salary band. The gap between what headquarters would approve and what a senior engineer actually costs in Tel Aviv ran 30 to 45 percent.
(A 2026 check: the market says I was being polite. levels.fyi's current medians put a senior engineer's total comp in Tel Aviv at roughly double the Paris number: ₪538,621, about €155K, against €78,412. Even the man complaining about the wall underestimated the wall.)
No funnel tweak closes a 40 percent comp gap. The candidate who loves your team still reads the offer, and the offer is 40 percent light.
I'd spent months perfecting the shape of the pipe while the actual problem was the pressure inside it.
The pipe-or-wall test
This diagnostic belongs in week one of every turnaround. I call it the pipe-or-wall test.
Every blocker is a pipe or a wall.
A pipe is process: something you own and can reshape. Move the screen earlier, rewrite the brochure, give candidates a second door. Pipes bend.
A wall is money or politics set above your pay grade: a salary band, a sign-off chain, a commitment the executive table never actually made. Process work moves pipes. Process work has never once moved a wall.
Two tells are cheap to check:
- The delta tell. Every improvement lands, and one site still reads zero while the others climb. That's not a process problem. Stop optimizing.
- The signature tell. The fix you actually need requires an approval you can't give yourself. That's a wall with a name on it. Your job there is escalation and evidence, not another funnel tweak.
I missed both.
A process person attacks every problem as a process problem, because that's the tool in his hand. He'll burn a quarter optimizing a pipeline whose real defect is a number set two pay grades above him.
The work is pleasant, even good. It changes nothing.
The fixing was the cheap part. The diagnosis cost me four months: sorting the problems that are yours to solve from the ones made of money or politics that no interview loop will ever touch. I got most of the way through the program before I admitted which kind Israel was.
The same diagnosis-first rule holds before you scale anything: fix the org before you grow it. Structure decides what hiring can even do.
Walls have builders
The wall didn't build itself.
Every hire needed a French signature. Every regional call ran on Paris defaults, made by people who'd never sat in the sites they were deciding for. I have watched headquarters and a local site grade the same team with different rulers before. This was the hiring edition of the same disease.
A company only half-committed to a region gets the worst of both: the band stays wrong, the site stays starved, and the recruiters take the blame for all of it. A nicer careers page just wastes everyone's time. The candidates' most of all.
Diagnose before you optimize. Run the pipe-or-wall test in week one. If the change needs a signature you cannot give, stop polishing the funnel. Bring evidence to whoever owns the wall.
The overhaul was good work. I'd run all of it again.
What stings is the sequencing. The band was the same height in month four as in month one. Walls don't mind waiting.
The next stalled site I meet, my first meeting won't be with the recruiters. It'll be with whoever owns the band.