Most of the AI pitched to recruiters writes things: outreach sequences, job ads, polished resumes. Some of it is useful. None of it is where the fee comes from.
On a boutique desk, revenue moves on a short list of numbers you already watch: how fast the first credible submittal goes out on a new req, your submittal-to-placement ratio, and your fall-offs. If a tool cannot show you a change in one of those, it is a writing aid, not a revenue tool.
Time to first submittal
On a req the client has given to several agencies, the first strong submittal usually sets the bar. The slow part is rarely the writing. It is turning the JD and the intake call into criteria, and then working out which of the people you already know fit them.
This is where software earns its keep: criteria pulled from the JD in seconds, your own pool ranked against them, and plain-English filters for the things the client said that never made it into the JD. You still make the call on who goes. You just get to the shortlist while the req is fresh.
Submittal-to-placement
Clients remember the agencies that send three right candidates instead of eight maybes. Raising the ratio means screening harder before anything goes out, on two sides at once: the client's must-haves, and what the candidate last told you they need on rate, location, remote and start date.
Doing both by hand for every candidate on every req is where good intentions break down. A shortlist that marks each candidate fits, check or conflicts, with the reason, turns it into a two-minute review instead of a morning.
Fall-offs
You know which of last quarter's fall-offs were avoidable. The unglamorous fix is current terms on every candidate before they are submitted. That is a perfect job for automation: scheduled check-ins that ask whether someone is still looking and whether anything changed, with the answer written straight back to their record.
Proof of submission
When two agencies send the same person, the one with a clean record of who was submitted, to which client, for which req and when, is in the better conversation. A submission log that writes itself costs nothing to keep and settles arguments.
What to keep human
The client relationship. The candidate conversation. Judgment on the close calls, which is most of them. The right software takes the retyping, re-reading and chasing off your plate so there is more time for those.
Measure it
Pick one live req this week. Note when it came in, when your first submittal went out, and how many you sent to get an interview. Run the next comparable req with better tooling and compare. If the numbers do not move, the tool is not working for you.