Every team hiring more than a handful of people a year makes this decision, usually implicitly and usually by drifting. The four models have wildly different cost structures, and the right answer depends less on your headcount than on a property of your roles that most teams never state: how precisely can you specify what you're looking for?
The four models and what they really cost
| Model | Est. annual cost | Ramp time | Best for | Fails when |
|---|---|---|---|---|
| Recruiters do everything | Marginal — but 15–25 hrs/role of existing capacity | None | Under ~15 hires/year | Sourcing is dropped whenever interviews get busy |
| Dedicated sourcers | $80K–$120K fully loaded each | 2–4 months to hire, 1–2 to ramp | Sustained volume; hard, ambiguous searches | Volume is spiky; handoffs leak candidates |
| Agency / RPO | 15–25% of salary per placement, or retainer | Weeks | Outlier roles, urgent gaps, new geographies | Used as the default channel — costs compound |
| Automation (agents) | $1K–$15K plus 3–6 recruiter hrs/role | Days to weeks | Definable, repeatable roles at volume | Briefs are ambiguous; no interview capacity |
Software cost estimates come from our pricing benchmark, and labour figures assume a fully loaded recruiter cost of ~$55/hour. All are directional planning numbers, not quotes.
The variable that actually decides it
Company size is the usual framing and it's a weak predictor. The better question: can you write down what a qualified candidate looks like, precisely enough that someone else could apply it?
Highly specifiable roles — most engineering, sales, support, high-volume healthcare, skilled trades — have a bar you can articulate and repeat. These are where automation performs, because the calibration loop converges quickly and the same brief serves many requisitions.
Poorly specifiable roles — first executive hires, a function nobody in the company has done, roles where the requirements emerge during the search — need someone who will argue with the hiring manager. That is a human skill, and it is what a good sourcer or a retained search is actually for. See sourcing hard-to-fill roles.
Most companies have both, in roughly an 80/20 split by volume. That is why the endpoint for most teams is a hybrid rather than a single model — and why "which one should we pick" is usually the wrong question.
The break-even points
Recruiters-do-everything → automation. Practically immediate. On 20 roles a year, a $2.5K agent needs to save roughly 2.3 hours per role to pay for itself; manual sourcing runs 15–25. The barrier is trust, not arithmetic — which is why a piloted rollout matters more than a business case. Full model in our ROI math.
Automation → dedicated sourcer. A sourcer costs $80K–$120K fully loaded. For that money you could run an agent plus a substantial database plus contact data and still have $60K left. The case for the sourcer therefore rests on capability rather than cost: ambiguous briefs, relationship-heavy markets, senior searches where outreach must come from a person. If your open roles are mostly definable, this is the hire to defer.
In-house → agency. One placement at 18% of a $120K salary is ~$21,600 — roughly what a full in-house stack costs for a year. Agencies are rational for outliers, urgency, confidentiality, and markets you have no network in. They stop being rational the moment they become the default channel, because nothing accumulates: no data, no talent pool, no institutional knowledge.
Agency → RPO. Above roughly 100 hires a year with sustained volume, an embedded RPO team usually beats per-placement fees. Below that, the fixed cost is hard to justify.
The costs each model hides
Recruiters do everything: the invisible cost is inconsistency. Sourcing is the first activity dropped when interviews spike, producing pipeline that arrives in bursts and roles that stall for reasons nobody can attribute.
Dedicated sourcers: ramp time (three to six months before full productivity), the handoff to recruiters (candidates leak at every handoff), and single-point risk — one person leaving takes the market knowledge with them.
Agency / RPO: nothing compounds. No candidate pool, no reusable brief, no data on what worked. And agency candidates are usually being marketed to your competitors simultaneously, which compresses your decision time.
Automation: two real ones. First, calibration and grading time in the early weeks — real hours, and skipping them is how pilots fail. Second, less hands-on control: you're trusting a screening bar rather than watching each profile, which some hiring managers find uncomfortable until pass-through data reassures them.
How to run a hybrid
The pattern that works for most teams above 25 hires a year:
| Role type | Share of volume | Model |
|---|---|---|
| Repeatable, definable (engineering, sales, support) | ~70% | Automation plus recruiter review |
| Specialist or ambiguous | ~20% | Human sourcer or the most experienced recruiter |
| Executive, confidential, or genuinely scarce | ~10% | Retained search or a specialist agency |
Three operating rules make the hybrid work rather than just coexist. Route roles deliberately at intake — decide the model when the requisition opens, not after it stalls for two months. Keep one system of record so every channel's candidates land in the same ATS with source attribution; without it you cannot compare channels and you will re-source people you already know. Report cost per sourced hire by channel, using the definitions in our metrics guide — it is the only way the routing decision improves over time.
Sequencing the transition
If you're moving from one model to another, the order matters:
1. Instrument first. Two weeks of hours logging and a qualified-candidates-per-week baseline. Changing models without a baseline means you'll never know whether it worked, and you'll be arguing from anecdote at renewal.
2. Pilot on live roles with a control. Whatever you're adding — an agent, an agency, a sourcer's first month — run it against a comparable role handled the old way. The design is in our buyer's framework.
3. Don't cut the old channel until the new one has cleared two cycles. Overlap costs a month of licence or fees; a bad cutover costs a quarter of pipeline.
4. Reassign the freed hours explicitly. Time saved that isn't reassigned disappears. Decide in advance: more roles per recruiter, faster interview loops, better candidate experience, or reduced agency spend. If nobody owns the reallocation, the ROI stays theoretical.
For tool candidates, start with the segment ranking closest to your situation — startups, enterprise, staffing agencies, solo recruiters — or the 2026 overall ranking.