Is AI Worth It for a Small Recruiting Agency? Run the Numbers

Most "AI ROI" content for recruiters comes from vendors selling the AI. This post gives you the inputs to do the arithmetic yourself, with sources, and a short list of signs that AI is not worth it for your desk yet.

What one placement is worth

According to HeroHunt, a vendor, writing in October 2026, contingency fees typically run 20-30% of first-year pay; retained search runs about one-third of first-year cash compensation; and contingency on a $120,000 engineer comes to roughly $24,000-36,000 (HeroHunt, Oct 2026, vendor). On a $25,000 salary at a 20% fee, one placement is $5,000.

For temp and contract desks, the same source reports gross margins from 16.1% (ManpowerGroup, Q2 2026) to 39.1% (Robert Half, Q2 2026). Those are large public firms; your margins will differ.

The useful question for a placement desk: would better, faster submittals plausibly contribute to one more placement this year? If yes, compare one placement fee with what setup costs. If you cannot picture that, keep reading.

What the software already costs

AI spending sits on top of your applicant tracking system (ATS). A third-party read from 100Hires, a competitor, in July 2026 lists Bullhorn Core at $165 per user per month, about $5,940 a year for three recruiters and $19,800 for ten (100Hires, Jul 2026). It lists Bullhorn's AI Assistant as a Pro-plan feature and Amplify as a separate metered product. Confirm with your vendor; see which AI is in your ATS.

So the first decision is often not "should we buy AI" but "should we move up a plan, add a product, or use a business AI account alongside the ATS we have."

What training costs

The American Staffing Association (ASA) sells a six-session "Practical AI for Staffing Agencies" masterclass at $299 for members and $449 for non-members, capped at 30 participants (ASA, 2026). Bullhorn runs AI-literacy sessions for its own users (Bullhorn, undated). Group training is affordable. The gap it leaves is setup in your system, with your format.

What vendors report about results

Bullhorn's GRID 2026 report, a vendor survey of about 2,300 recruitment professionals (Feb 2026), reports:

• 46% say AI cut screening time by half or more
• 55% say AI screening alone improved their key metrics by more than 25%
• top-performing firms are about four times more likely to use AI
• only 10% have implemented agentic AI (AI that runs multi-step tasks on its own) across the full workflow

Treat these as direction, not promise. The survey is run by a company that sells the AI, and "firms that use AI grow faster" does not prove AI caused the growth.

Do your own arithmetic

You need four numbers from your own desk. Fill them in over one normal week.

For a placement desk:

1. Submittals sent per week: ____
2. Minutes a recruiter spends writing and formatting one submittal: ____
3. Minutes it takes after a template, including the check (measure on 10 real ones): ____
4. Your average fee per placement: ____

Hours freed per week = (line 1 x (line 2 - line 3)) / 60.

Then ask what a recruiter does with those hours. If the honest answer is "more calls with candidates and clients," you can judge whether that plausibly adds a placement over a year, and compare one placement fee with the cost of setup.

For a temp desk: skip placements. Price recruiter hours freed per week against what you pay for those hours, and look at speed to fill on urgent orders.

Signs AI is not worth it yet

• Low volume. If your desk sends only a few submittals a week, a good written template and checklist may be enough.
• No agreed format. If every recruiter writes submittals differently and nobody wants to agree, AI will copy the mess faster. Fix the format first.
• No owner. Templates drift as clients and roles change. If nobody will own them, plan for that or wait.
• You want it to decide. If the hope is AI that screens candidates out on its own, read the AI hiring rules post first. Tools that score or screen can bring bias-audit, notice and record-keeping duties in places like New York City, California and, from January 1, 2027, Colorado, according to law-firm summaries. Illinois already requires notice when AI is used in covered hiring decisions. That is general information, not legal advice.

Signs it is worth it now

• Recruiters are spending more time writing than talking.
• Clients have flagged inconsistent or error-filled submittals.
• You are adding recruiters and want new hires producing your format in week one.
• You already pay for AI features nobody uses.

Costs people forget

The software line is the easy part. Count these too:

• Owner time to agree on the submittal format, the review rule and what candidate data may go into which tool.
• Recruiter time for training and a two-week pilot, when edits are slower than usual.
• Upkeep. Templates drift as clients, roles and your ATS change. Someone has to own them.
• Checking. The review step never goes away. If a template saves time only because nobody checks it, it is not saving time.

A decision in three questions

1. Do we have the AI already? Check your plan. If yes, setup and training are the cost. If no, compare an upgrade with a business AI account.
2. Which one task first? Usually submittals. One template, one review rule.
3. How will we know? Spot-check 20 drafts after a month and track client-flagged errors and recruiter time per submittal.

If those answers are clear, start small. If they are not, the free planning work is the useful part. See how to build a submittal template and what is actually working in 2026.

A free second look

Lunan's free conversation walks through these same questions with you and gives you a short written answer, including "not yet" if that is the honest one.