Years ago; around 2008, at ASTL, when we were maybe 20 people; I sat through a quarterly hiring review where Marcia, wearing one of the many hats you wear at a startup that size, walked us through a funnel chart. Eleven hundred applications at the top. Forty-something phone screens. One accepted offer at the bottom. Conversion was up over the prior quarter, so the slide had green arrows on it, which sadly remembers how much we cared about accessibility in the choice of colours back then.
When it got to me I asked who the strongest engineer we’d talked to all quarter was. The room thought about it and named the person we’d hired. That was the only name anyone could still remember.
Nobody at that table could have told you what excellent looked like for the role. Which meant nobody could say whether we’d found it. We’d filtered eleven hundred people to get there and had no idea.
I should say what I was doing in that room. I ran engineering. For a stretch, hiring for my own roles was another of those hats, so alongside Marcia’s funnel I was writing the specs and sitting on the closing calls myself. This is not a story about a recruiting team doing something to me.
I defended the funnel for a long time. I helped build some of those dashboards. I used to open hiring reviews by quoting top-of-funnel volume like it meant something. It took me embarrassingly long to see the obvious thing, which is that a filter tells you who survived it, and that’s a fact about the filter.
The funnel came to hiring from sales, where it earns its keep. A sales prospect is fungible and is already trying to buy something. More at the top, more at the bottom. Fine.
Hiring has neither property. Run a wide net at volume and you converge on average, because the people most available to a cold outreach campaign are the ones with the least going on where they sit today. Inbound skews toward people who need a role on your timeline. The people you actually want aren’t on anyone’s timeline. They’re busy.
Every company already knows the other way to do this. It’s how they hire a VP. An exec search starts with a written spec and a mapped list of the small number of people who fit, and getting from that list to three conversations can eat half a year of somebody’s attention.
How many of those months does an ordinary req get? Zero is the usual answer. A search is organized around people. A req is organized around a req. Most of what goes wrong later follows from that, and nobody notices it was a decision.
What I do now starts with scoping, pushed further than is comfortable. Write down what excellent looks like for this role, on this team, over the next year. Then rank it. A flat list of qualifications is a wish. Ranking makes you admit which one you’d trade the others away for, and most hiring managers would rather not.
I’ve sat in a kickoff where the hiring manager, a senior manager, wanted someone from IBM and had no answer when I asked which part of that person’s current job maps to the work this hire would walk into. I’ve also been the one with no answer, so I’m not throwing stones from very far away. Both times the logo was standing in for work nobody had bothered to describe.
The phrase to watch for in yourself is “I’ll know it when I see it.” What it means is that scoping is postponed until you’re already looking at humans. The bar gets set by whoever showed up that month. You end up comparing candidates to each other instead of to the job.
Most leaders also waste their networks, mostly by asking them a useless question. “Do you know any good engineers?” returns nothing valuable 99% of the time. You’ve just handed your scoping to someone who never saw the scope, and they will answer with whoever they had lunch with.
Narrow it until it feels too specific to ask. Who’s the strongest person you’ve worked with at taking a half-specified data contract and getting three teams to agree on it? Who did you watch turn a research prototype into something a product team could run without a PhD on call? People can answer a question shaped like that, and they answer it with names.
Then notice which names come back from unrelated corners of your network. That handful is the shortlist. It’s usually shorter than you’d like.
One name that kept coming back for us took over a year to land. Let's call her Johanna. Not her real name, obviously. Johanna told me no twice. Neither no had anything to do with us. The first time, she'd just been handed a team that was replacing the payment and invoicing system, the one that processed millions of dollars a day, and she wasn't going to walk out on them. The second time there was a rewrite of a legacy sales system written in ASP that she wanted to see shipped before leaving anything behind.
I told her it was the right call and that I'd have thought less of her for leaving in the middle of it. Then I asked when the rewrite would land. Spring, she guessed. I put a reminder in my calendar for April. In between I sent her an architecture writeup on migrations using the strangler fig pattern that I figured she'd want to argue with (she did), and an intro for a friend of hers who was job hunting. Neither came with an ask attached. Almost all of it happened over food at Wurst in Union Station, not in video slots with an agenda. One of the best hot dogs in Toronto, if you're interested.
Somewhere in those months I stopped hearing “not now” as a no. It’s usually the start of a longer conversation, and treating it that way has produced better hires than any sourcing campaign I ever ran, at a fraction of the cost, with the minor downside that you can’t put it on a dashboard.
Interviews are the part I got wrong the longest. Across a table, an interview mostly measures interview performance. That’s a real skill. It’s rarely the one the job needs.
Working next to someone gets you closer to the actual work. A paid half day on a real problem. An afternoon pairing on an open bug in the codebase they’d inherit. Every time I’ve run a process both ways, the side-by-side version changed at least one decision, and not always toward no; more than once it rescued somebody the panel had gone lukewarm on. One of them bombed the system design round, fixed a bug in the pairing session that had been open for a month, and now manages the entire team there. It runs the other way too. The candidate is evaluating you the whole time, and a scripted panel tells them nothing except that you have a scripted panel.
The close starts at first contact. By the time an offer goes out, either the person already wants to work with you or the number has to carry everything, and if the number is the whole conversation a better-funded competitor wins it. Every time. Put a fair number on the table early and spend your energy on the parts of the decision that aren’t about money.
For finalists I run a short daily sync with everyone who’s touched the candidate. It costs calendar. It catches the thing they mentioned that nobody followed up on, and following up is most of the game. What people carry out of a process is whether anyone was listening to them.
There’s also a stretch almost everyone forgets: the weeks between signature and start date, when the person is still sitting inside a company that now has every reason to fight for them. Counteroffers after acceptance have become normal enough that I plan for them. Staying in contact through that gap is cheap. A dinner with the future team, or onboarding material sent well before day one, buys more goodwill there than a signing bonus.
Caring more than the other company costs nothing. It’s the one advantage available to a team with no brand and no recruiting budget, and it holds up against a hyperscaler working from a comp band you’ll never match.
One more thing I didn’t start out believing: talent density belongs to teams, not individuals. Drop a strong engineer onto a weak team and you get weak-team output within months, then a resignation letter. Maybe smaller AI-era teams change that math. I haven’t seen it happen yet.
So the shortlist worth building is a list of people who’d make each other better, which is a harder thing to search for than any single name.
If you want to test this, run it on one role while the funnel keeps running next to it. Write the ranked scope before you talk to anyone. Ask your network a question narrow enough to be answered with names. Go meet whoever keeps coming up, before you have an offer in your pocket. The pace has never stopped feeling uncomfortable to me. I’ve also never once regretted a hire that came out of it.