Playbook

Last reviewed: May 2026.

AI moves the bottleneck. Most teams haven’t moved with it.

The useful question about AI in recruiting isn’t which tools to buy. It’s where the constraint goes once the obvious work is automated. For years the constraint was recruiter capacity: the manual volume of sourcing and screening that capped how many roles a recruiter could carry. AI takes a real bite out of that work. The constraint then moves to the hiring managers and interviewers, who still have the same number of hours in the day. A team that automates sourcing and screening but does nothing about interview load just relocates the traffic jam.

The operational response is the point

The leaders getting value from AI aren’t the ones with the longest tool list. They’re the ones who changed how the work runs to match the new constraint. That means cutting interview stages that no longer earn their place and putting real accountability on scheduling and interviewer responsiveness. It also means using transcript and interview-intelligence tools so hiring managers spend less time writing up debriefs and more time deciding. AI earns its keep when it both removes mechanical work and surfaces the signal that helps a team actually close.

Be careful what you let AI decide

AI screening carries a specific risk worth naming. Research on large language models used for resume screening has found a self-preference effect: models tend to favor resumes that read as though they were written by the same model family, which can hand a meaningful shortlisting advantage to candidates who happened to use a particular tool. That has nothing to do with who’s best for the job. If you’re going to use models in screening, mitigate it deliberately: constrain the criteria with system prompting, and take a majority vote across more than one model instead of trusting a single ranking. Keep a human reading the borderline cases.

A maturity question, not a tool question

The right framing for a leadership team is maturity: where are we using AI to change real outcomes, where are we using it as theater, where are we exposed to it being used against us (candidates optimizing for the screen, vendors overstating capability), and where are we sitting it out and paying for that. The answer drives different decisions than a procurement list does. The tools change every quarter. The operating questions don’t.

I keep this piece current because the specifics move fast. The last review date is at the top.

The most underweighted hire in recruiting is the one that isn’t a recruiter

When a recruiting team gets another headcount, the default move is to add a recruiter. Sometimes that’s right. Often the seat that returns the most capacity is an operations hire, or, cheaper still, a process or a system that removes work the whole team is repeating. The leaders who scale recruiting well ask a sharper question before they post the req: would another full-cycle recruiter actually produce more hires than fixing the thing that’s slowing every recruiter down?

Why operations is the load-bearing layer

A recruiting function is a system before it’s a group of people. The forecast, the data that proves the forecast, the rubrics that hold the bar, the scheduling that keeps loops moving: that infrastructure decides whether the recruiters can do their jobs at all. Strong recruiters on a broken operating layer still miss the plan, because they spend their hours on coordination and cleanup instead of candidates. Operations is what turns individual effort into a function that holds together as it grows.

The data layer is what makes the forecast trustworthy

A forecast is only as good as the instrumentation under it. Plenty of teams can’t pull clean time-to-fill or pass-through by stage, often because they switched ATS in a hurry and never rebuilt the reporting. When the data is shallow, the plan is a guess, and finance is right to distrust it. Building the clean signal under the forecast (consistent stages, reporting that reconciles to the operating plan) is operations work, and it’s the precondition for being believed.

Encode the bar into the system, not the leader

Quality holds at scale when the system creates the friction. Calibration sessions, scorecards with real SLAs, leveling rubrics, and interview kits that live inside the ATS mean ten interviewers produce the same signal whether or not the head of TA is in the room. The goal is to distribute a standard so the interviews run inside it, rather than depending on one person to personally police every loop. That’s an operations build, and it’s what keeps the bar from drifting the moment hiring speeds up.

The pipeline you already paid for

The most underused asset in most recruiting stacks is the CRM. Companies spend heavily to generate candidates, then let silver-medalists and prior applicants go cold while paying again to source net-new for the same roles. Reactivating that pipeline is cheaper than fresh sourcing and faster to convert, and it’s almost always an operations problem (data hygiene and nurture) rather than a sourcing one.

Operations is where AI actually gets implemented

A lot of companies went looking for a “talent engineer” to own AI in recruiting. In practice that work is being absorbed into an upskilled operations function, which is the right home for it. The person who owns the systems and the data is best placed to decide where AI removes work and where it degrades quality without anyone noticing. Treating AI as an operations capability rather than a separate hero role is both more accurate and easier to sustain.

Deciding when to invest in operations instead of another recruiter is one of the highest-leverage calls a recruiting leader makes, and one of the least discussed. If you’re weighing it, I work on exactly this through Foresight Insight.

A headcount number is not a hiring plan

Most companies set an end-of-year headcount target and treat the number as the plan. The number is an aspiration. The plan is the thing that converts it into reqs open by specific dates, each with a named owner and a realistic time-to-fill behind it. Without that conversion, a recruiting team is reacting to roles as they land instead of working a forecast, and the leadership team has no honest read on whether the number is even reachable.

Treat hiring like a forecast

Finance forecasts revenue. Recruiting should forecast hiring with the same discipline. Start from the target, work backward through historical time-to-fill by function and level, account for the roles that have to be sequenced (you can’t hire the team before the manager), and you get a calendar of when each req has to open to land the hire on time. That calendar is the plan. It tells you in March whether the December number is already at risk, while there’s still time to do something about it.

What good looks like

A healthy function can answer these on any given week without scrambling:

• How many hires do we still need this year, by function and level?
• Which reqs need to be open right now to hit those dates, and are they?
• Where is actual time-to-fill drifting from the assumptions in the plan?
• Which roles are gating others, and what’s the cost of the delay?

The numbers behind those answers live in the ATS and the operating plan, not in someone’s head. When the head of finance asks how hiring is tracking, the answer is a number with a date.

The common failure mode

The usual failure is a recruiting team measured on speed to fill whatever lands in the queue, with no work-backwards plan above it. Roles open late because nobody mapped the sequence, then everything compresses into the back half of the year, quality drops under the rush, and the team gets blamed for a planning failure that started upstream. The fix isn’t more recruiters. It’s a forecast that makes the gaps visible early.

If you’re standing up this kind of plan and want a second set of eyes, I do this work through Foresight Insight.

The bar is a system problem, not a willpower problem

Most quality problems in hiring get framed as people problems: this manager hires too fast, that panel is too soft. Caring more doesn’t scale. The bar holds when the operating system makes it hard to lower, and slips the moment it depends on one person being in the room.

Build the friction into the process

The mechanisms are well understood and underused: structured interviews with defined competencies, scorecards interviewers complete before they discuss the candidate, calibration sessions where a group rates the same profiles until they agree on what a “yes” means, and leveling rubrics and comp bands that take the negotiation out of individual heads. Each one moves a judgment from gut to rubric, which is what lets ten interviewers produce one standard.

Calibration is the highest-leverage habit

If I could install one practice, it would be regular calibration. Pull a set of real candidates, have the panel score independently, then surface where they disagree and resolve it. It’s the cheapest way to find out that your interviewers aren’t actually evaluating the same thing, which they usually aren’t until you make them do this. Calibration is also how a standard survives turnover, because it lives in the shared rating rather than in the tenured interviewer who’s about to leave.

Where it breaks

The bar drops fastest under hiring pressure, which is exactly when leaders are tempted to suspend the mechanisms that protect it. “We’ll tighten up once we’re through this crunch” is how a company trains itself to hire below its own standard. The structural fix is to make the friction cheap enough to keep running under pressure, so it doesn’t get switched off when it matters most.

Your binding constraint is probably the hiring manager

When hiring is slow, the reflex is to look at the recruiting team. Often the real bottleneck sits with the hiring managers: the roles they haven’t clearly defined and the debriefs they haven’t given. Recruiting can run a tight process and still stall if the manager is the slow part of the loop, and adding recruiters does nothing to fix that.

Why this gets missed

Recruiting metrics usually measure the recruiting team, so they make recruiter activity visible and hiring-manager drag invisible. Time-to-fill gets attributed to the recruiter even when half of it is a manager who took two weeks to review a slate. Until you instrument the manager’s part of the process, you’ll keep solving the wrong problem.

What to put in place

Make the manager’s responsibilities explicit and measured. An intake that pins down the role and the bar before sourcing starts. Service levels on resume review and debrief turnaround, with the delay visible in reporting. Interview load tracked so you can see when a team is over-committed and protect its calendar. Enablement so managers run interviews that produce real signal instead of a vibe check.

The leadership conversation

This is a place where a recruiting leader has to manage upward, holding managers and executives accountable for their part of hiring. That’s uncomfortable, and it works far better as a system (shared SLAs and visible reporting) than as a personal confrontation in a one-on-one. Build the accountability into the operating rhythm and the data does the arguing for you.

Recruiting is one part of the people function, and the seams are where it breaks

Most TA leaders treat recruiting as a self-contained funnel that ends when the offer is signed. The handoff into onboarding, the shared work with HR business partners, the data that has to move from the ATS into the HRIS, the feedback loop on whether a hire actually worked out: that connective work sits between teams, so it belongs to no one’s metrics and falls apart out of view. I’ve spent years on both sides of that line, and the seam between TA and the rest of People is one of the most underworked sources of leverage in the whole function.

Why it gets missed

TA and HR optimize separately. A recruiting team is measured on time-to-fill and offer acceptance, a People Ops or HRBP team on retention and manager effectiveness. Nobody owns the handoff, so a candidate has a great hiring experience and then a rough first week, or a role gets filled fast and the new hire turns over in six months because the bar and the actual job never reconciled. Each team did its job. The machine seized anyway.

The seams worth engineering

A few places where TA meets the rest of People, and where designing the handoff pays off:

• The offer and leveling line. Comp belongs to People Ops, leveling is co-owned with the hiring manager, and bands are non-negotiable from TA’s side. When those ownership lines are explicit, the “hire above band” pressure runs through a process instead of around it.
• The onboarding handoff. The moment a candidate becomes a new hire is a baton pass, and a dropped baton there undoes the whole hiring effort. A simple shared checklist of what TA knows that onboarding needs (the promises made, the role context the new hire is walking into) closes most of the gap.
• The data flow. The ATS and the HRIS have to reconcile, or your headcount numbers and your forecast both drift. This is operations work, and it’s where TA and People Ops either build one source of truth or maintain two that disagree.
• The feedback loop. HR business partners and managers learn how a hire is actually working out. That signal should flow back into TA’s calibration and scorecards, so the bar sharpens over time instead of staying frozen at the point of hire.

Why a whole-function background matters here

This is the part of the job a TA leader does better after having run the rest of the people function. I was once brought into a company specifically to build this bridge between the HRBP team and recruiting, to make the handoffs and the information flow actually work. Having owned People Ops elsewhere, across programs, analytics, onboarding, compliance, and M&A, I know where the boundary lines belong and where TA tends to over-reach. A TA leader who has only ever run a funnel often doesn’t see the seam at all, because they’ve never stood on the other side of it.

Engineering the connection between recruiting and the rest of People is unglamorous and high-leverage, and most companies leave it to goodwill. If you’re trying to make yours run as one machine, I work on this through Foresight Insight.

Measure recruiting like a function with a P&L, not a scoreboard

A lot of recruiting reporting is vanity: req counts and submittal volume. It looks like activity and tells a leadership team almost nothing about whether hiring is healthy. The metrics that survive scrutiny answer business questions: are we getting good people, fast enough, at a defensible cost, and is the plan on track.

The numbers that hold up

Four measures carry most of the weight. Quality of hire, measured however imperfectly through performance and retention of recent hires, because a fast cheap hire who fails is the most expensive outcome there is. Time-to-fill against the plan rather than in the abstract, so you see drift early. Pass-through by stage, which tells you where the funnel actually leaks. And true cost per hire, including agency and tooling spend, plus the loaded time of everyone in the loop, not just the recruiting budget line.

Speak the language of the people you report to

If your CEO came from finance or risk, frame recruiting as forecasting and exposure. A hiring plan is a forecast with a confidence interval. A slipping req is a risk with a cost attached. Reporting that connects hiring to those terms gets taken seriously in rooms where “we’re working hard on it” does not.

What to stop reporting

Cut the metrics that exist to look busy. If a number doesn’t change a decision, it doesn’t belong in the leadership update. Reporting fewer, better numbers also protects the low-maintenance principle: a dashboard you can actually keep current beats a comprehensive one that goes stale and stops being trusted.

A recruiting function breaks in predictable places

The recruiting function that works at fifty people is the wrong one at five hundred, and the failure is predictable enough to plan for. The mistakes that hurt most are made one stage too late: keeping founder-led hiring after it stops scaling, or adding recruiters without the operations to support them.

Around 50 people

Hiring is founder-led and mostly works on taste and network. The risk is that the founder’s standard lives only in the founder’s head. The job at this stage is to start writing the standard down (a simple scorecard, a consistent interview question set) so it can survive being handed off later.

Around 150 people

The founder can no longer touch every hire, and the first real recruiters come in. This is where founder taste has to become a shared decision model, or quality starts drifting the moment the founder steps back. The common error is delegating the function (handing off the process) without distributing the standard (making the rubric legible enough that others produce the same signal).

Around 500 people

Volume forces specialization. Sourcing and operations split out from full-cycle recruiting, and the function needs a real operating plan and reporting. The break here is adding bodies faster than the operating layer can hold them, so the team grows but velocity and quality don’t.

Around 1,500 people

The function is a system with its own internal services, and the leader’s job is mostly design and capacity allocation rather than individual reqs. The risk shifts to bureaucracy: process that was protective at 500 becomes drag at 1,500 if nobody is pruning it.

The throughline

At every stage the same thing determines success: whether the standard lives in the system or in a few people’s heads. That’s why I treat operations and bar protection as the spine of the whole function, not as back-office support.