Recruitment AI for search & selection (2026)

Why S&S is a different AI question

Search & selection runs on fees. A fixed amount or a percentage of the annual salary, paid the moment a candidate signs and stays. That sounds simple, but it changes everything about which recruitment AI works for you and which does not.

Look at the math. A staffing firm earns a few tens of euros margin per hour per temp worker, across hundreds of placements. For you it works the other way round. One placement of a controller at 90,000 euros against a 22% fee is nearly 20,000 euros. But that fee only lands if the match is right. And in case of doubt it does not get paid, or worse, you lose the relationship.

That is the whole point. Your margin is not in throughput. It is in depth. In how well you understand what a hiring manager is really looking for behind the job profile, and how well you read a candidate beyond what the CV says. A tool that helps you screen thirty candidates in a day does not help you. A tool that helps you truly understand five candidates, that is where your money is.

There is a second question underneath that one, and most buying guides walk straight past it: where the depth is supposed to live. The understanding you sell is worth very little sitting in a notetaker while the job, the pipeline, the placement and four years of client history sit in a different system. That is the arrangement a lot of S&S firms have quietly ended up with, and the cost of it is not the licence fee. It is that the shortlist and the file stop being the same artefact.

This is exactly why the broader guide on recruitment AI per agency type treats S&S as a separate segment. Here we go deeper. Which AI layer strengthens your fee economics, which quietly undermines it, and where does the output have to end up to be worth anything in week six?

The fee economics of an S&S firm

Before we get to tools, let us sharpen the economics underneath. Because it determines everything.

An S&S firm lives on a handful of numbers. The fee per placement. The number of assignments you can carry at once. And, most important and most underestimated: the chance a client comes back. That last number is your real profit engine. Winning a new client costs you acquisition, pitching, building trust. An existing client who calls back because the last match was spot on costs you a phone call.

Run the numbers. Say you do 40 placements a year at an average fee of 15,000 euros. That is 600,000 euros in revenue. Now the question that matters: how much of that comes from clients you already knew? At a healthy S&S firm it is more than half. At an excellent firm it is three quarters. That repeat business is no accident. It is earned with matches that stuck and with reporting that showed the hiring manager you understood their business.

This is where “volume” thinking falls apart. The time you save by screening faster is almost never the time that costs you money. The time that costs you money sits in the wrong placement, in the match that walks away after three months, in the hiring manager who stops calling you back. No tool built for speed solves that. In fact, some make it worse.

What you save	What it is actually worth
10 minutes per intake write-up	Marginal. Admin is not a fee.
Pre-screening 30 candidates faster	Risk. Depth disappears, mismatch odds rise.
Catching a mismatch before the shortlist	High. Prevents a lost placement.
A hiring manager report that lands	Direct. This wins the next assignment.
Copying a summary from one tool into another	Negative. Same work twice, and the second copy drifts.

The right-hand column is where AI makes a difference for an S&S firm. The left-hand column is what most tools sell.

The AI stack that strengthens your fee

Good recruitment AI for S&S does one thing: it makes you deeper, not faster. Concretely, that means four things.

Recruitment intelligence beyond CV matching. A CV tells you what someone has done. It does not tell you why they left, what energizes them, whether they want to lead or specifically do not, and whether the career arc points toward this role or away from it. Those signals sit in the conversation. AI that reads the intake conversation for motivation, career direction and soft signals gives you a profile no CV parser can approach. This is the difference between conversation intelligence and simple transcription. The first thinks along. The second types along.

Structured reporting toward the hiring manager. This is your differentiator, and it is badly underrated. A shortlist that only holds names and scores is a commodity. A shortlist where each candidate comes with why they fit, backed by what they said in the conversation, is advice. The hiring manager feels the difference immediately. AI that produces a structured summary per conversation type, candidate intake, vacancy intake, evaluation, gives you the building blocks for reporting that strengthens the relationship instead of merely informing it.

Clickable transparency. Suppose you put a candidate at the top of the shortlist and the hiring manager asks why. “My gut” is not an answer that justifies a fee. “Because in the conversation she said she was looking for exactly this step up, here, listen” is an answer that builds trust. Tools where every sentence in the summary is clickable and links back to the exact moment in the conversation give you that backing. Not for yourself, but for the conversation with the client.

A place for all three to land. This is the layer nobody demos, and it decides whether the other three are worth anything six weeks later. An S&S process is rarely one conversation. You speak to a candidate at intake, after a client meeting, after an assessment, and in between you are updating a job, moving an application, sending a profile. If the conversation output lives in one product and the pipeline lives in another, somebody re-types the useful half and abandons the rest, and the two versions drift within a month. When both sit on the same record, the shortlist and the file are one artefact: the reason a candidate is at the top is stored next to the application it belongs to, and it is still there when the hiring manager asks in week six. That is recruitment intelligence in practice — not summarising per conversation, but having the summaries accumulate somewhere they can be searched and cited.

Four layers. All aimed at depth. None of the four at speed.

What actually works against your fee

Now the part most S&S firms get wrong. Because the market pushes you toward volume tools, and that marketing is convincing.

Mass-screening tools promise you sort hundreds of candidates in minutes. For a staffing firm that is gold. For you it is poison. The reason is exactly the fee economics above: your value sits in the depth such a tool automates away. If you reduce a candidate to a match score of 87%, you throw away what you are paid for. The hiring manager can run an ATS that spits out scores by themselves. They do not need you for that.

Watch the signals in the demo. A tool boasting about “10x more candidates per recruiter” or “autonomous shortlisting” is built for a different agency type than yours. The per-agency-type analysis calls this pitfall two: buying shiny features that do not fit your economics. At S&S that pitfall is costlier than elsewhere, because the tool literally automates the part of the work your fee is paid for.

A second category to be careful with: autonomous rejection agents. Not because they do not work, but because at S&S they are rarely worth it. You work with small numbers and high stakes. The time you gain by automating a rejection pales against the risk of brushing off a candidate who later reaches your hiring manager through a connection. In a market where reputation is your entire asset, that is not a trade you want to make.

The third one is the quietest, and most S&S firms are living with it right now: a stack where the intelligence and the records sit in different products. It rarely arrives as a decision. You buy an ATS in year one, a notetaker in year three because the ATS write-ups were unusable, and by year five your best material about a candidate is in a tool your pipeline knows nothing about. Ask the vendor selling you that second product what happens to its output. If the answer is a two-way sync, be more sceptical rather than less: two systems of record writing to each other gives you two systems of record and no truth, and reconciling them lands on your consultants every week forever. A match score throws depth away on purpose. A split stack throws it away by accident, which is harder to notice and takes longer to fix.

In short: when in doubt, choose the tool that lets you slow down at the right moments, and make sure what it produces ends up where the placement does.

Simply

Simply is an ATS. That is worth stating plainly, because it is the answer to the question the section above leaves open. The job, the applications, the shortlist and the placement live here, on a data model you shape yourself — and so does the conversation that justified the shortlist. We do not sync with another ATS, not Bullhorn, not Carerix, not anything in that category. For an S&S firm it fits on four points.

AI summaries per conversation type. A candidate intake calls for a different structure than a vacancy intake or an evaluation conversation. An intake comes out as a brief, a screening as a screening, a debrief as a recommendation with its reservation attached. The structured fields it picks up — salary expectation, notice period, availability — arrive as proposed updates sitting next to the record. You approve, you edit, or you throw it away. Nothing is written until you do, and the approval and the approver go into the audit log with it.

Verifiable transcripts. Every sentence is anchored to the moment in the audio it came from, and a claim in the summary points back at the line it was derived from. When a hiring manager asks why a candidate is at the top, the evidence is one click away, weeks after the call. That is not a feature for you, it is a feature for the conversation with your client.

Search over what was actually said. Individually a summary saves you ten minutes. Collectively they become the thing the system knows, because summaries are written into the notes on the candidate, the job or the company, and search runs over the same database as the records — keyword, semantic, or both fused into one ranking. So “who told me they would relocate for the right role” becomes a question you can ask of your own conversations rather than of whatever anyone remembered to tick. Retrieval is permission-filtered, so a record you may not open is a record you cannot find.

Dashboards and reporting. Time to fill, stage conversion, where placements come from, built over any object in your model including the fields you added yourself. Worth knowing what it is not: these are aggregates and they stop there — counts, rates and trends, with no drill-through from a chart to the rows behind it. If you need the records, you open them where the permissions apply in full.

Honest about the boundary. Posting vacancies to job boards and multiposting are coming; they are not in the product today, so if your process starts with a multiposting run, that part stays where it is today. We do not host a branded careers site. For a staffing firm running on 90-second intakes we miss the volume features. And coming from another system is a migration rather than a connection: four weeks, and one-way — after the move there is no link back to the old ATS, which is exactly why the mapping is something you review rather than something we do quietly. We would rather say all of that in the first conversation than in month three.

Want more on the agency economics of S&S, read the for-whom page for search & selection.

Compliance: short but not optional

Two regimes run in parallel, and at S&S you can ignore neither.

The GDPR gives every candidate rights around access, deletion and, in automated decision-making, the right to human intervention. At S&S you work by definition with human judgment, so that right is usually well covered, provided your AI tool plays a supporting role and does not reject on its own.

The EU AI Act makes recruitment AI high-risk from 2 August 2026. That means requirements around risk management, data governance, logging, human oversight and accuracy. For you as a firm the practical question is simple: can your vendor show that the tool meets these requirements, with documentation and audit logs? If not, you inherit the legal exposure. The guide on GDPR and the AI Act for recruitment tools walks through this point by point.

One more question worth asking in the demo, because the answer is usually flattened into a single reassuring sentence: where does the data sit, and where does the model run? Those are two questions, not one. A vendor who leaves it at “everything is processed within the EU” is leaving you to work out which part they mean, and a DPIA reviewer will pull it apart anyway. Our own answer, so you have something to compare against: the platform and the candidate data are EU-hosted, on infrastructure we run ourselves, and the text that needs a model is processed inside the EU too — or on your own key with your own provider and region. The models themselves are bought in, which is a different question from where they run. Which model handled which request is in the audit log, with its cost in euros.

The reassurance for S&S: because your work inherently runs on depth and human judgment, it fits well within what the AI Act asks. A tool that supports you instead of replacing you is exactly what the legislator has in mind. The firms that get into trouble are the ones that automated rejection at scale. That is not your model anyway.

Where to start

No platform purchase, no annual contract based on a demo. Three steps.

Step 1. Calculate your repeat revenue. What percentage of your placements comes from clients who already came back to you before? That number is your real profit engine and it immediately makes clear where AI should help: with the quality of the match and the reporting, not with throughput.

Step 2. Take your three biggest time sinks that produce no fee. For most S&S firms those are conversation write-ups, drafting hiring manager reports, and data entry. That is where your first gain sits, and the nice part is that the tools solving this make you deeper rather than faster-and-shallower. While you are listing them, mark which ones exist only because two systems have to be kept in agreement. That number is usually higher than people expect, and it is the one that never shows up in a business case.

Step 3. Run a real assignment through it before you sign anything, with a metric set in advance. Not “feels good”, but “saves me two hours per assignment that I now put into extra candidate research”, and “in week six I could still show the hiring manager where that claim came from”. Our trial is fourteen days on the full product with no card; Starter is 79 euros and Pro 129 per user per month billed yearly, and AI usage runs on your own provider key so your provider bills you for it directly — no credit bundle to buy, no markup from us. Measure the outcome against your fee economics, not against a time-saving dashboard.

Start with your money. Not with the tool.