The short version: an agency gives you one firm's recruiter and one firm's database. A marketplace gives you the specialists across many firms who have already filled your exact role, coordinated under one contract. Both charge a percentage of salary on a successful hire. What differs is reach, duplicate control, and what happens between offer and joining.
If you are choosing between them, these are the dimensions that actually matter.
Reach
An agency's reach is the intersection of one recruiter's network and one database. That is often enough for a common role. It is usually not enough for a staff-level backend engineer with multi-region systems experience, or a perception engineer, or a first head of function.
For those, the constraint is not effort. It is whether the person working your search already knows where that skill set lives. A marketplace solves this by matching per role rather than per account: your robotics search goes to someone who has closed robotics roles, and your GTM search goes to someone else entirely.
Empanelling five agencies gets you similar reach and five times the management overhead — five contracts, five sets of candidates arriving in five formats, and no mechanism preventing two of them from emailing the same person on the same day.
Duplicate outreach
This is the failure mode nobody prices in.
When several vendors work the same role, candidates get contacted repeatedly about one job by people who do not know about each other. It reads as chaos, and it reflects on your employer brand rather than on the vendors.
A marketplace solves it structurally with candidate claiming: a recruiter claims a person on a role, the claim is timestamped and visible to everyone, and no one else approaches that person for that role. A duplicate check runs across the whole platform before anyone invests effort. Multi-vendor empanelment has no equivalent, because the vendors have no shared system.
Cost
Indian contingency fees generally sit between 15% and 25% of first-year salary, and marketplaces sit in the same band — it is the same contingency model.
So cost comparisons are rarely about the percentage. They are about:
- What triggers the invoice. On offer, on acceptance, or on joining. The gap between acceptance and joining is where Indian searches most often die, and paying on joining moves that risk off your side of the table.
- What happens if they leave. A replacement guarantee — ninety days is a reasonable bar — means a bad hire costs you a re-run, not a second fee.
- What you pay when nothing happens. Retainers, platform fees, seat licences and posting fees are all money spent before a hire exists. A pure contingency model has none.
Control and exclusivity
Retained search buys dedicated attention and asks for exclusivity plus payment up front. That is a reasonable trade for a CEO search and a poor one for a backlog of engineering roles.
Contingency — whether from an agency or a marketplace — should run non-exclusively beside your own team. If a vendor asks for exclusivity on a contingency fee, they are asking you to absorb their risk without compensating you for it.
Confidential searches
Replacing someone who still works for you, or hiring a first head of function before the team knows, is where multi-vendor breaks down entirely. Every additional firm is another set of people who know.
Handled properly, a confidential brief goes to a restricted, named subset of recruiters under NDA, the role is never published, and the comp bands and hiring plan stay inside that group. Ask any vendor exactly who will see the brief. A vague answer is the answer.
Where AI actually helps
AI is very good at the search half — building a structured spec from a role, working many sources at once, screening against explicit criteria, and writing evidence per candidate instead of a match score. It compresses days of list-building into hours.
It is not good at the part that decides your search. It cannot persuade a senior engineer to leave a job they like, read a hesitation in a call, or hold someone's attention through a notice period while their current employer counter-offers.
So the honest framing is not "AI recruiting versus an agency." It is: does the human working your role have the right background, and is the boring work done by software instead of by them? A marketplace that puts agents behind a specialist gets both. A tool with no specialist attached gets you a longer list and the same problem.
Choosing
Pick an agency when you have a long-standing relationship with a firm that genuinely knows your domain, or when volume in one function makes a dedicated desk worthwhile.
Pick a marketplace when your open roles span functions, when a role has been open long enough that reach is clearly the constraint, when you want multi-firm coverage without multi-vendor management, or when you want the fee tied to joining rather than to offer.
Pick retained search for a genuinely small number of senior, business-critical hires where you want exclusive, dedicated attention.
Whatever you choose, ask the same four questions: who specifically works my role and what have they closed like it; what stops two people contacting my candidate; when exactly does the invoice arrive; and what happens if the person leaves in ninety days.