Guides

Recruiter Compensation in the AI Era: Should BD KPIs Change When an AI Does the Research?

Most recruitment agency comp plans still pay for call volume and email volume. When AI absorbs the research and drafting, those activity KPIs stop measuring what matters. Here is how to redesign recruiter compensation for the AI era.

TB Team Boilr

· July 28, 2026 · 15 min read

Abstract dark liquid-metal texture with two converging currents, symbolising activity and outcome merging into one measure

TL;DR

Most recruitment agency comp plans still reward calls made, emails sent, and CVs submitted [1]. That made sense when a consultant's own hours were the scarce resource. It stops making sense once AI absorbs the research, the sourcing, and the first draft: AI-adopting agencies cut manual sourcing time by up to 90% and time-to-fill from 40-60 days to roughly 14 [2], so the activity numbers a comp plan was built to reward can now be produced by software, not judgment. 75% of recruitment leaders already admit their commission setup has problems, and half were planning changes [3]. The fix is not to abandon commission - it is to move the KPIs it is built on from volume produced to decisions made: which signals a consultant chose to act on, which drafts they rewrote and why, and which AI-surfaced opportunities they closed. boilr's Tasks model is built to make that shift measurable, because it logs exactly where the AI's work ends and the consultant's judgment begins.

Why Activity KPIs Are Running Out of Road

Commission plans in recruitment were never really about calls and emails. They were a proxy. Agencies could not directly measure "good judgment" or "found the right opportunity at the right time", so they measured the thing that correlated with it: effort. More dials, more emails, more submittals meant more shots on goal, and more shots on goal meant more placements. That correlation is breaking down.

The activity itself is now automatable. AI monitors thousands of sources around the clock, drafts the first-touch message, and can cut manual sourcing time by roughly 90% [2] - the exact hours a legacy comp plan was designed to reward can now be produced without a consultant lifting a phone.
Volume metrics were already vanity metrics in disguise. A true KPI has to be measurable, actionable, and tied to revenue; "anything else is a vanity metric" [4]. Emails sent and calls dialled fail the actionable test once an AI agent is the one dialling and sending the drafts.
AI-drafted outreach converts worse than human-reviewed outreach. Across more than 4 million completed sends, AI-drafted cold recruiting email replied at 4.97%, while a recruiter's own hand-typed first-touch email replied at 12.6% [5] - proof that the volume an AI can generate is not the same thing as the outcome a comp plan is meant to reward.
Agencies already know the current setup is broken. 75% of recruitment leaders report issues with their existing commission structure, and half were planning changes going into 2025-2026 [3]. Only 33% of the fastest-growing firms were actually investing in the tech to manage it properly [3].
Threshold and tiered models are already moving toward outcomes. 77% of agencies now run tiered commission schemes, up 10 points on the year before, rising to 90% among the most profitable agencies [1] - tiering is a step toward outcome-based pay, but the tiers themselves are still usually built on the same volume inputs.
AI adoption itself now correlates with hitting quota. Sales organisations in the highest AI-engagement tercile report 57% of reps at quota, against 39% in the lowest tercile [6] - a gap large enough that a comp plan ignoring AI usage is measuring the wrong variable entirely.

None of this means BD effort stops mattering. It means the effort worth paying for has moved up a level: from "did the work happen" to "was the judgment good."

What Actually Changes in a BD Comp Plan When AI Does the Research

An AI research and drafting layer does not remove the consultant from the commission equation. It removes one specific category of work from the equation: the category that used to be measured by activity counts. Three things shift as a result.

1. The unit of value moves from "touch" to "call"

When a consultant had to find the company, find the contact, and write the first draft themselves, "did they make the touch" was a reasonable stand-in for "did they do the work." When AI finds the company, drafts the contact list, and writes the first draft, the touch itself is nearly free. What is not free is the decision to act on signal A over signal B, to rewrite the draft's opener because the AI missed a relationship detail, or to walk away from a lead the AI scored highly because the consultant knows something the data does not. That decision is the unit worth paying for now.

2. "Quality of judgment" becomes measurable, not just aspirational

Agencies have always said they want to reward "quality over quantity", but without a system logging what a consultant chose to act on versus what they ignored, there was no data to pay against - so plans defaulted back to counting activity because that was the only thing the CRM actually recorded. An AI layer that surfaces every signal and logs every consultant decision against it (accept, edit, reject, ignore) creates exactly the dataset a quality-based plan needs: edit rate, override rate, and close rate on AI-surfaced opportunities specifically.

3. The commission base narrows toward outcomes AI cannot manufacture

AI can manufacture volume. It cannot manufacture a placement, a signed mandate, or a candidate's decision to accept an offer. As the volume side of the funnel gets cheaper to produce, the honest move is to weight commission more heavily toward the outputs only a human relationship can close - not because activity stops mattering operationally, but because paying extra for something software now does for free just rewards the wrong behaviour.

Activity KPIs vs Outcome KPIs: What to Pay For Now

The table below maps the legacy activity metrics most recruitment comp plans still lean on against the outcome-and-judgment metrics an AI-augmented desk should actually be paid against.

Legacy activity KPI	What it measured pre-AI	Why it breaks down now	Better AI-era replacement
Calls made / dials per day	Effort and time invested	AI can trigger and log call tasks faster than a human can dial	Call-to-meeting conversion rate on signal-sourced calls
Emails sent per week	Outreach breadth	AI-drafted volume is nearly infinite and converts worse unedited [5]	Edit rate + reply rate on AI-drafted sends the consultant approved
CVs submitted per recruiter	Sourcing throughput	Sourcing time can drop ~90% with AI, so raw count stops signalling effort [2]	Submittal-to-interview ratio (industry benchmark ~3:1) [7]
Leads worked / prospects contacted	Pipeline building activity	AI can generate and score leads continuously without consultant hours	Close rate on AI-surfaced signals specifically vs self-sourced leads
Time logged in CRM	Engagement / diligence proxy	Says nothing about whether the time produced a good decision	Override rate: how often the consultant correctly overrode an AI recommendation
Number of new accounts opened	New business hunting	AI can flag account-worthy signals faster than a consultant can research them	New accounts converted to a signed mandate within 90 days

Building the New BD Comp Plan: A Practical Sequence

Redesigning a commission plan mid-flight is risky - recruiters plan their lives around their OTE. Move in stages, not a single cutover.

Audit what your current plan actually pays for. List every metric in the plan today and mark each one "activity" or "outcome." Most legacy plans are 70-80% activity-weighted; that ratio is the starting diagnostic.
Instrument the judgment layer before you change a single number. You cannot pay for edit rate, override rate, or AI-signal close rate if nothing logs them. Get the AI/BD tooling reporting those numbers for 2-3 months before they touch commission.
Separate "AI-assisted" pipeline from "self-sourced" pipeline in the CRM. This is the single most important structural change - without this split you cannot tell whether a placement came from consultant hustle or from a signal the AI surfaced, and you will misattribute commission either way.
Run the new metrics as a shadow scorecard first. Publish the new KPIs alongside the old comp plan for one full quarter without changing pay. Let consultants see how they'd score before their income depends on it.
Rebalance the base-to-variable split before you rebalance the KPI mix. If activity is being de-weighted, consultants need a stable floor while the new outcome-weighted variable proves itself - protect the base, don't just cut the old bonus.
Phase the new plan in with a guaranteed floor for one quarter. Guarantee the higher of "old plan result" or "new plan result" for the transition quarter so nobody is punished for a measurement change they didn't choose.
Review quarterly, not annually. Static annual comp plans are being replaced by continuous, quarterly-reviewed models precisely because what AI automates keeps shifting [8].

The KPIs a Human-in-the-Loop BD Desk Should Actually Track

Metric	What it tells you	Healthy range
Edit rate on AI-drafted outreach	Whether the consultant is genuinely reviewing, not rubber-stamping	Track the trend; falling over time signals better AI drafting, not laziness
Reply rate: reviewed vs unedited AI sends	Whether human review is actually adding value	Reviewed sends should outperform unedited AI sends
Close rate on AI-surfaced signals	Whether the consultant is converting the opportunities AI found	Benchmark against self-sourced close rate; parity or better is healthy
Override rate (correct)	Whether the consultant's judgment is adding accuracy over the AI score	Should stay positive - overrides that beat the AI score, not just disagree with it
Submittal-to-interview ratio	Sourcing quality, not just sourcing volume	~3:1 industry benchmark [7]
Time from signal to sent outreach	Whether review is fast enough to keep the AI's speed advantage	Same day, ideally under an hour
Placements per consultant per quarter	The metric that ultimately funds every other KPI	Track trend against headcount, not in isolation

How boilr Makes the Judgment Layer Measurable

boilr is built as an AI sales employee per consultant, and its Tasks model is designed to draft, not send - which is exactly the seam a redesigned comp plan needs to instrument:

Signals - monitors funding rounds, executive moves, expansions, and job-posting velocity around the clock, often surfacing hiring intent 48-72 hours before it hits job boards, so signal timing stops being a manual research task.
Companies - identifies and enriches target clients against the agency's ICP automatically, removing the prospecting-hours variable a legacy comp plan was built to reward.
Candidates - sources shortlists against live requirements so sourcing throughput is no longer the scarce resource on the desk.
Company Brain - the agency's shared memory of winning angles and ICP patterns, which is what makes "good judgment" repeatable and attributable across a team rather than locked in one consultant's head.
Tasks - every drafted opportunity lands as a task with the signal, the contact, and the drafted approach attached. The consultant's accept, edit, or reject action on that task is the exact data a judgment-based comp plan needs and a volume-based one never captured.
Verify and send - boilr never sends unsupervised. The consultant decides what goes out, which keeps the sending decision - and the accountability for it - with the person a comp plan should reward.

What stays entirely human, and should stay compensated as such:

Deciding which AI-surfaced signal is actually worth acting on today
Rewriting an outreach draft with relationship context the AI cannot know
Discovery calls, negotiation, and closing
Client and candidate relationship management over multiple mandates

5 Mistakes Agencies Make Redesigning BD Comp for the AI Era

Mistake #1: Cutting commission because AI "did the work"

Why it fails: AI did the research, not the deal. Cutting pay because a signal was AI-surfaced punishes the consultant for using the tool correctly and teaches the desk to hide AI usage rather than lean on it.

Fix: Pay for the outcome regardless of which channel surfaced the opportunity; measure AI-usage separately as an efficiency metric, not a commission penalty.

Mistake #2: Keeping the old activity KPIs and just adding new ones on top

Why it fails: Teams that track 20-40 metrics but act on only 3-4 real KPIs are the norm, not the exception [4]; bolting judgment metrics onto an unchanged activity-heavy plan just adds noise consultants learn to ignore.

Fix: Retire at least one legacy activity metric for every new outcome metric you add.

Mistake #3: Changing the comp plan without instrumenting the data first

Why it fails: You cannot pay for edit rate or override rate if no system is logging them; changing pay ahead of the data invites disputes about numbers nobody can actually verify.

Fix: Run the new metrics as reporting-only for a full quarter before they touch anyone's pay.

Mistake #4: Treating this as a one-time redesign

Why it fails: What AI automates keeps shifting quarter to quarter; a comp plan frozen for a year will be measuring last year's automation boundary by the time it is reviewed [8].

Fix: Review the KPI mix quarterly, not annually, alongside whatever the AI layer has newly taken over.

Mistake #5: Ignoring that consultants price certainty into their OTE

Why it fails: A comp plan that changes constantly is a named frustration for recruiters, and uncertainty about how they'll be paid drives attrition regardless of how sound the new metrics are on paper.

Fix: Phase changes in with a guaranteed floor for at least one transition quarter, and communicate the "why" before the "what."

A One-Quarter Plan to Redesign Recruiter Comp for the AI Era

Weeks 1-2: Audit

List every metric currently tied to commission. Tag each "activity" or "outcome." Calculate the current split.

Weeks 3-6: Instrument

Turn on edit-rate, override-rate, and AI-signal-close-rate tracking in your AI/BD tooling and CRM. Report it, but don't pay against it yet.

Weeks 7-10: Shadow scorecard

Publish the proposed new KPI mix alongside real commission statements so consultants can see how they'd score under the new plan before it's live.

Weeks 11-12: Launch with a floor

Go live with the new plan, guaranteeing the higher of old-plan or new-plan pay for the launch quarter. Set the first full quarterly review date before you launch, not after.

See what a judgment-first BD desk looks like in practice. Try boilr free or book a demo to see how a task moves from signal to draft to a consultant's decision to send.

Frequently Asked Questions

Should recruitment agencies cut commission when AI does more of the BD work?

No. Cutting commission because AI surfaced a signal or drafted an outreach message punishes consultants for using the tool correctly and encourages them to under-report or avoid AI usage. Pay should track the outcome - the placement, the mandate, the closed deal - regardless of which tool helped surface the opportunity. AI usage is worth measuring as an efficiency metric, not treating as a reason to reduce pay.

What KPIs should replace calls made and emails sent in a BD comp plan?

Replace pure volume counts with metrics that capture judgment and outcome: edit rate on AI-drafted outreach, reply and close rate on AI-surfaced signals specifically, override rate (how often a consultant correctly disagreed with an AI recommendation), submittal-to-interview ratio, and placements per consultant per quarter. These require an AI/BD system that logs consultant decisions, not just activity counts.

Is it true that most recruitment agencies already have problems with their commission structure?

Yes - 75% of recruitment leaders report issues with their current commission setup, and half were planning changes going into the current cycle [3]. AI adoption is accelerating that reckoning because it changes which activities are actually scarce, but the underlying dissatisfaction predates AI.

How do I measure "quality of judgment" instead of just activity?

You need a system that logs what a consultant chose to act on, edit, or reject against every AI-surfaced opportunity - not just what they eventually sent. From that data you can calculate edit rate, override accuracy, and close rate specifically on AI-sourced pipeline versus self-sourced pipeline, which together approximate judgment quality far better than a raw activity count ever could.

Won't consultants just stop using the AI tools if commission is tied to how they use them?

Only if the metrics punish AI usage itself. The design goal is to measure what the consultant did with the AI's output, not whether they used AI at all - reward good editing and good acceptance decisions, not manual re-work for its own sake. Done correctly, this incentivises heavier, smarter AI usage, not avoidance.

Should base salary go up if variable commission is being tied to fewer, harder-to-hit outcomes?

Most agencies should protect or raise the base-to-variable ratio during a transition, at least temporarily. Consultants price certainty into their expected OTE, and a comp plan that changes commission mechanics while also cutting the guaranteed floor drives attrition regardless of how sound the new KPIs are on paper.

How often should a BD comp plan be reviewed once AI is part of the workflow?

Quarterly, not annually. Static annual comp plans are increasingly being replaced by continuously reviewed models because what AI automates keeps shifting [8]; a plan reviewed once a year will be measuring last year's automation boundary by the time anyone looks at it again.

What role does boilr play in redesigning BD compensation?

boilr does not set your comp plan, but its Tasks model creates the data a judgment-based plan needs: every AI-surfaced signal, drafted outreach, and consultant decision (accept, edit, reject) is logged in one place. That turns "quality of judgment" from an aspiration into something an agency owner can actually measure and pay against.

Sources

Information sourced from public industry reports, benchmarks, and research publications as of July 2026.

Back to all articles