Candidate Research in Executive Search: How the Practice Works

“Candidate research” gets used two ways in executive search, and the difference is worth pinning down before anything else. Sometimes it means the whole engagement, brief to placement. This page uses it more precisely: the specific work that sits between “we know what we’re looking for” and “here is who we found, and here is why you should believe what we’re telling you about them.” Finding people. Establishing what’s actually true about them. Deciding, one requirement at a time, who continues.

That middle section is also the part of a search that varies most without anyone outside the firm being able to tell. A brief is a document a client can read. A placement is an outcome a client can judge. Candidate research sits between the two, mostly invisible, and a thin pass through it looks identical to a rigorous one right up until a client asks where a specific claim came from.

What this covers, and what it doesn’t

This page is about the research function specifically — not fee structures, not the client-management side of a retained engagement, not how a search gets scoped in the first place. Those live in how a search runs end to end, which is its own subject.

What belongs here is narrower and more mechanical: the work of turning a set of requirements into a small number of people whose fit can be defended, one criterion at a time, with a source behind each claim. Four things have to happen for that work to hold up, and most of what separates a good research process from a weak one is how seriously each of the four gets taken.

Why it’s harder than it looks

The difficulty isn’t finding names. At the executive level there is rarely a shortage of people who hold the right title. The difficulty is that a title is a weak proxy for what someone actually did, and the people worth finding are disproportionately the ones a keyword search won’t surface at all.

Two things drive that. First, the most senior, least visible candidates are often underrepresented in the exact databases research usually starts from — profiles that are thin, out of date, or simply absent, because the people least motivated to be found are frequently the passive candidates a keyword search is least likely to surface. Second, even when a profile exists, it’s self-reported. “VP Operations” can mean four sites and full P&L ownership, or it can mean a title given at a smaller company with none of that scope. Nothing about the title tells you which.

So a research process is really solving two problems that pull against each other: cast wide enough to not miss the person who never shows up in an obvious search, and look closely enough that “found” doesn’t get confused with “confirmed.” Most process failures come from optimizing one at the expense of the other — a wide net with no verification discipline behind it, or a verification-heavy process applied to a pool too narrow to have found the right people in the first place.

The four things a disciplined process has to do

1. Find people a keyword search would miss

Discovery has to do more than run the obvious search terms against the obvious databases. The candidates most worth surfacing are frequently the ones who aren’t optimizing their own visibility — which means discovery has to work through indirect signals: who a company’s own filings and announcements name, who shows up in adjacent-industry moves, who a network path connects to rather than who a search term matches.

This is also where tooling choices matter most, and most research processes default to whichever tool is already open rather than the one suited to the search. The database most research starts from indexes people who keep their own profile current — close to the opposite of the population most worth finding. Active sourcing reaches past that gap but doesn’t scale the way a database search does, and a sourcing-services engagement buys both at a cost most boutiques can’t spend on every search. Knowing which gap each category of sourcing tool actually closes, not just that gaps exist, is most of what separates a stacked sourcing process from a lucky one.

2. Establish what’s actually true

A name with a plausible-looking title is not the same thing as a verified candidate, and the gap between the two is where most bad shortlists come from. Verification means going past the resume to something a third party said independently: a filing, an announcement, a named press release. Then checking whether it actually supports the specific claim being made — not just one adjacent to it.

The standard worth holding here is simple to state and easy to fail under time pressure: a claim is only as strong as the source behind it, and the source has to be specific enough that someone who wasn’t part of the research could go find it themselves. “Public sources” describes nothing. A named filing, dated and locatable, describes something. Executive CVs in particular are written to survive a fast scan, which means the work is often reconstructing what someone actually ran from sources other than their own description of it.

3. Map evidence to what the brief actually requires

Once a claim is established, it still has to be checked against the specific thing the brief asked for — not a general impression of strength. A candidate can be genuinely accomplished and still fail a specific must-have; a candidate can look under-credentialed on paper and still satisfy every requirement that matters. The only way to tell the difference is to evaluate each requirement on its own, separately, rather than forming one summary judgment and working backward to justify it.

This is also where honesty about gaps has to live. Evidence for a requirement is rarely a clean yes or no — a title matches but the scope is undocumented, or a role is confirmed but the timing is ambiguous. A process that forces every candidate into “qualified” or “not qualified” is quietly discarding information a client would want, and the candidates worth including as imperfect-but-substantial fits are usually the ones an overly binary process cuts first, without writing down why.

4. Decide who continues — and say why

The output of all of this is a decision about who reaches the client, and the decision has a shape that’s easy to get wrong. A single ranked list quietly makes a judgment call that belongs to the client, not the researcher — who is “best” depends on context the research usually can’t see: what the last search taught the client, what the board will and won’t accept, what a hiring executive actually wants versus what they said they wanted.

The alternative that holds up under scrutiny separates candidates who meet the stated requirements from candidates who miss something explicit but bring comparable substance, and says, for each name on the second list, exactly what they miss and why they’re worth a look anyway. Most shortlists fail at this step quietly, not by leaving the second list out but by writing reasoning too generic to act on — a line that reads as filler the first time a client pushes on it. What a shortlist entry actually has to contain to survive that kind of pushback is specific per claim, not per candidate; the mechanics of building one come down to whether the reason for each cut got written down at the moment of the cut, or reconstructed afterward when someone asked.

Where AI actually fits

Every research process now runs into the same question: what changes when a model is doing some of the work. The honest answer is narrower than it first looks, and worth stating plainly rather than oversold or waved off. A model is genuinely useful at pattern-matching evidence against a stated rubric quickly and consistently, and at drafting the first pass of a comparison a person would otherwise do by hand. It is not verifying anything on its own: a model asked to confirm a claim will produce a confident-sounding answer whether or not the underlying evidence actually supports it, with no way to tell the difference between a name that matches on paper and the person the name is actually attached to.

That gap is exactly where the discipline above has to sit, regardless of who or what runs the first pass. Inside a screening step specifically, a model can do the rubric-matching reliably, but it cannot close the verification gap on its own — which means every claim it marks confirmed still needs a human check, not just the ones that look wrong. The more open-ended version of the same question — what a general-purpose assistant can and can’t responsibly do with a research prompt — has a shorter answer: it will produce something, and what happens to that output before it reaches a client is where the actual work still lives.

The practice bends by search type

None of the above is uniform across every kind of search, and treating it as though it were is its own failure mode. A CEO search is less often blocked by a shortage of research and more often blocked by a client, frequently a committee, that hasn’t yet agreed internally on what it’s actually hiring for. Research can’t resolve that; a process that assumes it can produces a shortlist nobody in the room can agree on. A board search evaluates something different in kind from an operating role, and a candidate who reads as an obvious fit by operating-role logic can fit badly on a board for reasons that logic doesn’t surface. An operations search lives or dies on scope detail: site count, P&L ownership, regulatory environment. None of that comes from a title alone, which means the proxies research usually leans on are the least reliable ones for this specific function.

Find, verify, map, decide: the mechanics hold across all of it. What changes is which of the four takes the most weight, and what “verified” actually has to mean for the role in front of you.

How to tell if a research process is any good

Because the research stage is largely invisible to a client, it’s worth having a way to check it — whether it’s your own team’s process or a vendor’s. A few questions do most of the work:

• Ask to see the actual citation behind one claim, not a description of it. A URL, a filing, a dated announcement — something you could open and check yourself — means the process is doing its job. “We checked, it’s accurate” is not a citation, and a process that can only offer reassurance instead of a locatable source hasn’t actually verified anything.
• Ask what happened to the candidates who didn’t make the list. A process that can say which requirement each cut candidate failed, and what was checked before the cut, is a process that was actually evaluating against the brief. A process that can only describe cuts in general terms was filtering on convenience.
• Check whether every requirement was evaluated on its own, or whether candidates got one overall impression. A single “strong candidate” judgment is a summary, not evidence. A specific per-requirement result is something you can act on.
• Look at whether the near-misses were considered at all. A shortlist that only contains people who clear every stated requirement cleanly is either evaluating a very easy market, or its research stopped exactly at the words the brief used and never considered who might satisfy the same underlying need differently. Neither is common, and the honest reason is usually the second one.
• Ask how long the discovery pass actually was before anyone started filtering. A process that narrows immediately to a comfortable number of names has usually already made the decisions it should have made at the end, not the start.

None of these require special tooling to check. They require someone to ask, and a process built with the answers in mind from the start.

Where Cerna fits

Cerna is built around exactly the four-part structure above, run as a standing discipline rather than something reconstructed under deadline for each search. Discovery starts wide — a first pass typically surfaces 150–200 candidates before any filtering begins, because recall, not precision, is the goal at that stage. Every criterion in the brief is then evaluated for every candidate individually: confirmed, unconfirmed, or flagged, never averaged into one impression. Every confirmed claim carries a source specific enough to check, and a separate pass reviews the output before it’s sent, specifically looking for the claims the first pass got wrong. The result is delivered as two sections: candidates who meet the stated requirements, and laterals who miss something explicit but bring comparable substance, each with the reasoning attached. What Cerna does not do is rank the result or tell you who’s right — that decision, and the context it depends on, stays with the recruiter.

If any of the four sections above described a gap in how research gets done today, seeing how a search actually runs end to end is the next step.

Further reading

This page draws on the individual posts below, each covering one part of the practice in more depth: active sourcing in executive recruitment, what candidate sourcing tools are actually built for, what LinkedIn Recruiter covers and misses, what candidate sourcing services include, how executive search researchers find candidates, how to screen executive candidates, what AI candidate screening actually involves, how to shortlist candidates for an executive role, what a good shortlist actually contains, how AI is actually used in candidate research, what ChatGPT can and can’t do for executive recruiting, how a CEO search actually works, how a board search differs from an executive search, what makes an operations executive search different, and executive search best practices for candidate research.