Is AI sourcing worth it for a small agency

I get this question from small agencies more than any other, and it usually arrives with a worry attached. They have seen the big agencies turn on AI sourcing and pour thousands of profiles into a pipeline, and they cannot see how a three-person desk competes with that, or whether they should try. The honest answer is that the flood is the part they should not copy, and the finding is the part that works better at their size than at anyone else's.

I have spent the last few years building the sourcing side of Recruitly, which means I have watched what the search index actually returns, what people do with it, and what it costs. This post is what I would tell a friend running a small desk, with the mechanics underneath so the advice makes sense.

What AI sourcing actually does under the hood

Strip the marketing off and AI sourcing is two things chained together. The first is a model that turns a role description into a search: title variants, skills, seniority, location, the companies people in this job tend to come from. On Recruitly that step reads the job spec and an ideal candidate profile and produces the filters a good researcher would have typed by hand, in a few seconds instead of twenty minutes. The second is an index of people, in our case a little over 800 million profiles, that the filters run against, returning a ranked list.

That last point is the one to hold onto. The ranked list is ranked by how well a profile matches the query, which is a statement about the document, and the document is the part every candidate has now polished. So the list tells you who looks right on paper, faster and more widely than you could have looked yourself. It does not tell you who will pick up the phone, who is actually open to a move, or who will turn out to be the person in the interview room. That work is still yours, and it is where a small agency wins.

The cost most small agencies get wrong

The obvious cost is credits. On Recruitly a people search costs 15 cents, a live profile refresh half a cent, a verified work email 15 cents and a personal email or phone number 25 cents. Those are small numbers, and the expensive part of sourcing 300 people for a role is what happens to the 300 afterwards.

Every one of those people has to be contacted, or they were found for nothing. Every reply has to be read. Every conversation has to be had. And every one of them who is any good is now sitting in your pipeline waiting for a client who will see two of them this week, which means the good ones are waiting longer than they would have if you had found thirty. The credits are a rounding error next to the consultant hours and the candidates lost to the wait.

AI sourcing pays for itself in credits on any desk. The question for a small agency is whether you use it to find thirty people you will actually call or three hundred you will mostly leave, and the vendors' own dashboards push you towards the second because it is the number that looks like value.

Why a small agency has the advantage here

A three-person desk cannot absorb three hundred sourced profiles per role, and that turns out to be a feature. The limit forces the valve that big agencies have to impose on themselves with policy. You source to the size of pile you can actually work, which is also the size of pile the client can actually decide on, and the two numbers are close on a small desk because nobody has the spare hours to pretend otherwise.

The other advantage is the database you already have. A small agency's own records are the warmest list it will ever source from, because those are people who replied before. The candidate who calls you back is worth fifty cold matches, and AI sourcing is at its best when it runs against your own records first, which is what the dedupe and the per-job AI search on Recruitly are built to do, and against the wider index second, for the roles where your database is thin. Add your team's LinkedIn network, which Team Relationship Intelligence reads so you can see who on the desk already knows a person before anyone sends a cold message, and a small agency's version of AI sourcing is a warm-first search with a cold backstop. That is a better instrument than the flood, and it is cheaper to run.

The three places it is worth it

After watching a lot of small desks turn sourcing on, the uses that pay fall into three buckets, and the ones that do not pay are almost always the flood in a different costume.

A role you have no database for. A new sector, a new city, a client who has moved into a skill set you have never placed. Here the index earns its keep, because your own records have nothing, and the model's version of the search is better than the one a consultant would write for an unfamiliar market. Source thirty and call all thirty.

The specific ten. The searches that convert are the narrow ones. People who held this exact title at one of these four companies in the last three years. AI sourcing is far better at that query than at "senior developers in Manchester", because the narrow query returns people the client will recognise, and the client seeing two a week is the whole constraint.

Enrichment only for the shortlist. Find the email and phone for the people you have decided to call, after you have decided, which on Recruitly is a click on the record rather than a batch on the list. Enriching three hundred profiles you will never contact is the credit cost most owners remember and resent, and it was the flood, again.

How to run it on a two-person desk

None of this needs a policy document. It needs four habits, and they are the habits that fall out naturally when you do not have the hours to waste.

1. Set the valve per role before you search. Ask how many people the client will see this week, double it, and that is the size of the pile. On most roles that is a number between ten and thirty, and it is the number you source to.
2. Build the search from the last placement. The ideal profile that produces good filters is the person you already placed in a role like this, and the model reads that better than it reads a client's wish list.
3. Run it against your own database first, then the network, then the index. Stop at whichever level fills the pile. The database you built is your cheapest and warmest source and the index is the backstop, and most small desks have that order reversed.
4. Measure conversations. The number that tells you whether sourcing is working is how many of the people you found you actually spoke to, and how many of those the client wanted to meet. Profiles found is the vendor's number, and it goes up when you do this wrong.

Worth it, then, and emphatically so for the finding, and for the roles your records cannot cover, and for the twenty minutes per search a consultant gets back. The flood is a big-agency habit that big agencies are now trying to unlearn, and a small desk has the one advantage that matters here, which is that it was never able to afford it in the first place.

Written for owners of small agencies who have been told they need to source like a big one.