Almost everything sold to agencies works the candidate pipeline. The constraint in most agencies is the client pipeline, and no amount of faster sourcing fixes a thin book of roles.
Recruitment agencies run two pipelines, and that is the whole analysis. Candidates in one, clients in the other. The industry's AI tooling is pointed almost entirely at the first.
What the candidate-side tools do well
Sourcing against a brief. CV parsing and matching. Screening questions. Scheduling. Interview notes. Formatting a CV to a client's template.
All of it is mature, most of it works, and the administrative saving is real. A consultant who stops spending Friday reformatting documents gets Friday back.
Why that often does not show up in revenue
| If your constraint is | Faster sourcing |
|---|---|
| Not enough live roles | Does nothing. More candidates for the same placements |
| Roles you cannot fill | Helps, directly |
| Consultants buried in admin | Helps, indirectly, by freeing selling time |
| Clients going elsewhere | Does nothing. That is a relationship problem |
Rows one and four are where most agencies actually sit, and both are client-side. Sourcing speed against a thin book increases the number of candidates competing for the same few fees.
What the client side actually needs
Knowing which companies are hiring before the brief goes out. Hiring signals, funding, headcount growth, a departure in a team you place into. This is findable and most agencies do it ad hoc, if at all.
Remembering what happened last time. Which hiring manager, which role, what the feedback was, why the second placement did not happen. In most agencies this lives in a consultant's head, which means it leaves when they do.
Noticing a client going quiet. The account that placed three roles last year and nothing this year. Nobody notices because nobody is looking, and there is no alert for an absence.
AI helps with the first of those. The second and third are a record-keeping problem that no model solves, because the information was never captured.
The screening caution
Automated sorting against criteria is reasonable. Automated rejection is not, for two reasons that are usually discussed separately and should not be.
The fairness problem is obvious and increasingly regulated. The commercial one is less discussed: the candidate you rejected without a human looking is a hiring manager in three years, and agencies run on exactly that kind of long memory.
Keep a person at the point of no.
The part that compounds
An agency's genuine asset is the relationship record: who you placed, who hired them, what the feedback was, which manager moved where. That asset decays every time a consultant leaves with it in their head, and it is the one thing AI cannot reconstruct, because it was never written down.
Empiraa Signal is a CRM and more, built so the client pipeline has the same discipline the candidate pipeline already has: prospect data to find companies worth approaching, sequences to approach them, the booking link, the proposal, and one record per account that stays when a consultant does not. Unlimited users on every plan, so the consultants, the researchers and the directors all see the same thing rather than one licence holder and everyone else working from email.
Related reading: AI agents for small businesses, and what an AI CRM actually is.
