Automating the Executive Search Process

What is the executive search process?

The executive search process is the sequence a retained firm runs from client briefing to placed candidate, usually described as five stages: brief, search, assess, shortlist, appoint. Each stage is defined by the document it produces and the sign-off that releases the next one. The documents are the process, which is why they are also where the time goes.

Most published descriptions of this are written for the company hiring a search firm. This one is written for the firm. If you run a retained search process, the useful question is not what the stages are called. It is how much of each stage is work a person has to think about, and how much is work a person has to type.

Look at the third column. In every stage it is the same shape of task: read what is already on the record, apply the firm's own template, produce the document. A consultant who has just come off a two-hour briefing call is not making a judgement when they write the criteria up. They are transcribing one they already made.

Which parts of an executive search can AI actually do?

The document production, and only that. An AI tool with read access to your ATS can turn a briefing transcript into written criteria, score an assessment against them, and assemble a shortlist comparison from approved records. It cannot decide whether the criteria are right, whether a candidate is worth a call, or what to recommend.

Those three sit on the other side of a line, and most of what is sold as AI for executive search is one of the two halves pretending to be both. Here is the same five stages with the judgement half named, because a system that quietly crosses that line is the one that costs you a client.

Why the boutique firm gains more from this than the big one

A 5 to 25 person firm has no operations function and no research team to absorb the documentation load, so it falls on the consultant who took the call. That consultant is also the person doing business development, which is the work that actually grows the firm. In a large firm the same load falls on researchers and associates who were hired for it, which is what an executive search operations function is for. The smaller the firm, the more expensive the typing.

Which three triggers are worth automating first?

Three events carry most of the documentation load in a search: the briefing call being written up, the assessment being scored, and a candidate reaching shortlist. Each one fires off a note or a stage move your consultants already make, each one reads records that already exist, and each one produces a document somebody currently writes by hand.

What do the documents actually look like?

Two documents from the same search, so you can judge the work rather than the description. The first is what trigger 1 returns after a briefing call. The second is what trigger 2 returns once a candidate has been assessed against those criteria. Both examples are fictional, written to show the shape and the depth.

That last box matters more than it looks. The criterion the client never actually asked for is the one most likely to send a search sideways, so it is surfaced as a question rather than quietly promoted into the framework. That is the difference between a draft and a decision.

Once the client signs those five criteria off, they become the thing every candidate is scored against. Here is the same search two weeks later.

Two details in there are doing the real work. Every score has a quote attached, so the document is arguable rather than authoritative. And the thing that needs a human is called out at the bottom instead of being buried in the middle, which is the opposite of what a tidy summary usually does.

Why does every output need an approval gate?

Because an AI tool writing about people will produce a confident, well-formatted error, and a confident error in front of a client costs more than a slow document. The gate is one named person reading the draft before anybody outside the firm sees it. It takes a minute, and it is what makes the rest safe to build.

I have the error rate from one of my own builds, which is more useful here than a general warning.

1. Unconstrained copy invents things. On a sourcing build for a UK project controls agency, 64% of first-pass outreach openers contained a factual error, including inventing a programme and describing a risk register as a modelling tool. Constrained to quoted facts only, that stops. A deliberately general sentence does no damage; a confidently wrong one costs you the relationship.
2. Evidence drifts between roles. On the same build, 11 of 17 profiles recorded as describing current work were actually describing a previous employer. Still valid candidates. Wrong claim in the document.
3. Anything added after the check is not checked. I appended 18 candidates to a finished list after quality control had passed. Those 18 were rejected at 44%, against 6% for everything that went through in order.

The citation rule does two jobs at once. Your consultant can verify any line against the record in seconds, which is what makes the document defensible in front of a client. And the system cannot quietly invent a reason, because it has to produce the sentence it came from.

What should you not automate?

Anything where being wrong is expensive and hard to spot. That is the test, and it rules out more than people expect. The system drafts, flags and scores. It does not decide who gets shortlisted, it does not send anything to a client, and it does not write to a candidate without a person reading the message first.

• Shortlist decisions. A score is an input to that conversation. Treating it as the outcome is how a firm ends up defending a ranking it does not believe in.
• Anything outbound to a client or candidate. One wrong detail about somebody's career, sent under your name, undoes a lot of relationship-building.
• Reference and feedback wording. The facts can be assembled. How they are put is a judgement about two people who both have to work together afterwards.
• Fee and offer conversations. Nothing about this is repeatable, which is the only thing that makes a task a candidate for automation.

What is genuinely hard about building this?

Not the technology, which is the opposite of what most firms expect. Connecting an AI tool to an ATS that supports it is a configuration job, and generating a branded document is what these tools are best at. The hard parts are making the output sound like your firm, and keeping your ATS triggers clean enough to fire.

I have installed 85 Claude setups across recruitment businesses since January, and the split above is the most consistent thing about them. The connection takes a morning. The tuning takes weeks.

Making it read as though your firm wrote it

This takes iteration, not clever prompting. Build one document, run it against three or four completed searches where you already know what good looked like, and tune until a consultant reads the draft and does not reach for the delete key. That loop is where the build time goes, and it is not skippable. A document that is 80% right reads as more work than no document at all, because somebody now has to find the 20%.

Trigger discipline

An automation fires on a clean stage move or a correctly typed note. If half your consultants log a briefing call as a general note, the trigger does not fire for half your searches and the system looks broken when it is working exactly as configured. Tighten this during the build. Working around it afterwards means writing rules to guess what a note was, which is the kind of complexity that quietly rots.

What it costs to keep running

The per-document usage cost is small enough that it is not the number to plan around, even at full search volume. The cost that matters is the build and the tuning, and that is one-time. After the first document, your consultants own it, and adding the fifth or sixth is a short job rather than a new project, because they share the same triggers and read the same records.

Where should a boutique firm start?

One document, two consultants, three completed searches to tune against. Build the briefing write-up first, pilot it on real assignments with the two people most likely to tell you the truth about it, then widen. Resist building four documents at once, because the thing you are tuning is judgement and you can only hear feedback on one at a time.

1. The criteria document, off the briefing call. Clearest before and after, and it releases the search. If this one works your consultants will ask for the rest, which is worth more than picking the theoretically highest-value document first.
2. The rest of the brief stage. Charter, search approach, briefing roadmap. These share the trigger and most of the same reading, so they come quickly once the first one is right.
3. Assessment and shortlist. The heaviest documentation load in the whole process, and the biggest return in time.
4. Offer and appoint. Interview confirmations, reference capture, the offer checklist.
5. Reporting last. Post-search report and the weekly client update. Valuable, and not what is costing you hours today.

That order walks the executive search process from the front, because the earliest documents are the ones every later stage depends on.

On the sourcing side of the same firm, the equivalent first move is your own database. On one build, the agency held 18,195 candidate records and only 14% of the market for the skill they were hiring was visible to them, while 63% had never been seen by anyone. The full method is in the waterfall enrichment guide.

How do you map your own process before automating any of it?

Write the process down before you automate a line of it, because the mapping tells you which documents are worth the build. Three prompts below do it from the outside in: list what your searches produce, split each one into repeatable and judgement, then test the split on a document you already have.

Paste them into Claude and answer honestly rather than tidily.

I run a boutique executive search firm. We are [N] people and we
specialise in [your vertical].

Help me map our fulfilment process. Ask me one question at a time, and
keep going until you have all of this:
1. Every stage a search goes through, in our words, not textbook ones.
2. Every document or write-up we produce at each stage, including the
   internal ones nobody outside the firm sees.
3. Who writes each one, and roughly how long it takes them.
4. What releases the next stage: a sign-off, a stage move, or nothing
   formal at all.

Then show me the whole thing back as one table and tell me which rows
you had to guess at, because those are the ones we have never written
down.

Here is the map we just built: [paste the table].

For every document on it, split the work into two columns:
- REPEATABLE: reading records that already exist, applying our template,
  assembling, formatting, chasing.
- JUDGEMENT: anything that requires a decision, a relationship, or
  knowing something that is not written down anywhere.

Rules:
- If a document's first draft needs a decision in order to exist, say so
  explicitly. That one is not a draft.
- Rank the documents by how much repeatable work they carry AND how
  easily one of my consultants could tell a good draft from a bad one in
  under a minute.

Give me the top three to build first and say why each one beat the
others.

Here is a real [document name] we produced on a completed search, with
the client and candidate names removed: [paste it].

Here are the notes and records it was written from: [paste or attach].

Two questions:
1. How much of this document is traceable to something in those records,
   and how much came from the consultant's head? Mark it line by line.
2. For every line that came from their head, is that judgement we should
   keep, or is it something we simply never wrote down?

Be hard about the second one. I am trying to work out what we would lose
if the first draft of this was written from the record.

Where to go next

For the wider context, the recruitment operations guide covers how fulfilment systems fit the rest of a firm, the waterfall enrichment guide covers the sourcing layer that feeds a search, and the candidate sourcing guide covers the engine around it. If you would rather have the mapping done with you, that is the Claude ops build.

Frequently asked questions