Most hiring postmortems look at the wrong end of the process. A role sits open five months, and the review circles interview loops, comp bands, hiring manager alignment. Usually the damage was done in week one, when the shortlist got assembled out of whoever happened to be visible on one professional network.
The candidates you never saw don’t show up anywhere in your reporting. That’s the gap the newer sourcing tools are aiming at. Instead of another search box on the same platform, something like Lessie AI treats sourcing as one continuous task: describe the person you need in plain English, and it searches across a wide spread of sources, ranks what comes back, and drafts the first message. For a team with no dedicated sourcer, that changes what’s realistically possible in a week.
Recruiting has a sampling problem
The people easiest to find are the people doing the most to be found. Complete profiles, current titles, open-to-work badges. That group overlaps with strong candidates. It is not the same group.
Think about who’s missing. The engineer maintaining a library half your stack depends on, whose profile runs three lines and hasn’t been touched since 2021. The researcher whose paper your team quotes in every planning doc, who never made a profile at all. The designer whose portfolio would win the role in ten minutes, findable on their own site and nowhere else.
All three are invisible to a title-and-location filter. All three are exactly who the hiring manager described in the intake call.
The workaround has always been human. A good sourcer cross-references four or five platforms, follows citations, reads commit history, works backwards from conference speaker lists. It works. It also eats hours per role, which is why it gets reserved for VP searches and quietly dropped everywhere else.
What predicts fit doesn’t live in a profile field
Go through what actually tells you someone can do the job:
- what they’ve built and maintained, rather than what they’ve listed under Skills
- what they’ve published or presented, which shows both depth and whether they can explain it
- portfolio and shipped work, which for design and product roles is the only real evidence there is
- the stage of company they’ve operated at, since scaling from 5 to 50 is a different skill than running 500
- where they show up in a community and who they work near
Each of those sits on a different platform with a different structure. Nothing indexes all of them. So sourcing from one place produces a systematically narrower slate than the market contains, and you never find out, because the misses are invisible by definition.
An agent-based approach goes after this at the retrieval step instead of asking a recruiter to run five searches and reconcile them in a spreadsheet. Lessie quotes 100+ sources, which reads like marketing copy until you count how many places the evidence has scattered to.
Say the requirement out loud
Filter-based sourcing makes you translate a human requirement into whatever fields the platform supports, and the requirement gets flattened in transit. “Someone who’s scaled a design system at a B2B company” becomes “Senior Product Designer, 5-10 years, SaaS.” The part that made it useful is gone before the search even runs.
With a natural-language interface it survives:
Find senior product designers in Berlin who worked at B2B SaaS companies and have a strong portfolio.
Find AI researchers who published papers on multi-agent systems.
Find UX researchers in San Francisco who worked on enterprise products and published usability studies.
These are sentences recruiters already say out loud in intake meetings. The point isn’t that typing them beats clicking dropdowns on speed. It’s that they carry conditions the dropdowns couldn’t hold in the first place. Vendors quote match accuracy above 95% against roughly 60-70% for broad filter matching, and most of that gap comes down to criteria a filter has no way to express.
Getting a reply is the other half of the job
Finding the right person is only worth something if they answer. Good candidates get a lot of recruiter mail and bin the generic stuff on sight.
Personalization fixes that, and personalization has always been expensive. To reference someone’s work you have to read their work. At fifty candidates a role, most people give up and send the template. Everybody knows this. Everybody does it anyway.
Running discovery and outreach in one system is what makes it affordable. The tool already read the repo, the portfolio, the paper. That context is still sitting there when the message gets written. You end up with outreach that mentions specifics without a human spending twenty minutes per candidate digging them out. Around triple the reply rate is the figure that gets quoted, and it isn’t surprising. Personalization always worked. It just never scaled.
Skills-based hiring needs a skills-based search
The last few years produced real agreement that hiring should weigh what someone can do over where they went. Degree requirements came off job descriptions. Work samples replaced résumé screens. The intent was genuine.
Sourcing never caught up. You can commit fully to skills-based evaluation and still build your slate the old way, because the tools at the top of the funnel index credentials: title, employer, tenure, school. You end up assessing for skills a population you selected for pedigree, which quietly reinstates the filter you spent a year removing.
Searching across sources changes the shape of the funnel here, not just its speed. Demonstrated capability leaves traces. A repo, a paper, a case study, a talk, a product that shipped. Those traces are the evidence skills-based hiring assumed it would have on hand.
It also widens the pool in ways that matter if you care who’s in it. Career breaks, non-linear paths, self-taught backgrounds, employers nobody in your office recognizes: all noise to a pedigree filter, all unremarkable to a search built around what a person has built. That isn’t a diversity feature bolted onto the side of a product. It’s what happens when you stop using prestige as shorthand for competence at the retrieval step.
What it costs to run a real sourcing motion
Systematic sourcing has been priced as an enterprise capability. Contingency agencies charge upward of 20,000�ℎ���.������������������������������170 a month per recruiter before anything else in the stack. Both assume enough hiring volume to spread the cost across.
Most companies aren’t that. They hire in bursts, one or two people cover the entire function, and they fall back to job ads because the tooling was never justifiable for four open roles.
Tools starting around $35 a month change that arithmetic, mostly by dropping the headcount assumption baked into per-seat pricing. Seats exist because the software assumes a human driving every search. If discovery, ranking and first contact run as one pass, you don’t need a seat per role.
Two things it won’t fix
Better sourcing doesn’t rescue a bad intake. If the hiring manager can’t articulate what they want, a wider search hands you a bigger pile of wrong people, faster. Natural language actually raises the stakes on a clear brief, since the brief is now the entire query.
And an agent surfaces and ranks. It doesn’t assess. Whether someone will do well on your specific team, under your specific manager, is a human call and should stay one.
What moves is where the hours go. Time currently spent assembling a slate goes into the conversations that decide the hire instead. For two people covering fifteen open roles, that isn’t a marginal gain. It’s the difference between running a sourcing function and posting an ad and hoping.