We are a small recruitment agency drowning in admin and follow-up. Where do we use AI first?
Short answer: Start with candidate and client follow-up, because it is repetitive, time-stamped and low risk. Next comes CV intake (pulling details from a CV into your applicant tracking system, the ATS, which is the recruiter's database of candidates and jobs). Leave screening, ranking and rejecting for last, and keep a recruiter making every decision about a person. A one-desk pilot with a business-plan AI assistant costs tens of dollars a month in software, with the real cost being the recruiter's time to review drafts.
You run a small recruitment agency, your recruiters spend more time chasing and typing than placing, and you want to know which admin to hand to AI first.
I'm Ranjit, and I run Blur Studio: 15+ years building software, 120 apps taken live, and clients who have gone on to raise $10M+. I build AI systems and processes that help businesses do a lot more with a lot less time. I also work differently from most of what else is out there, more on that further down this page. Say hello.
Why follow-up comes first
Ask a recruiter where the day goes and the list is always the same: updating the candidate record, emailing candidates who have not heard back, chasing a client for feedback after an interview, rewriting the same "thanks, here is where we are" message. None of that is judgment. It is typing with a date attached. For the scheduling side, see recruiter candidate follow-up and interview scheduling.
That is the best place to start with AI for three reasons.
- A wrong draft is cheap. A recruiter reads it before it goes out, and a typo in a status update is not a hiring decision.
- The payoff is visible fast. Candidates who get a timely reply stay warm, and clients notice a faster agency.
- It does not touch the hard legal ground. Using AI to rank or reject people is regulated in some places. Drafting a polite update is not.
The order I would use
| Order | Job | AI does | Recruiter still decides | Risk |
|---|---|---|---|---|
| 1 | Candidate status updates | Drafts the message from the record | Reads and sends | Low |
| 2 | Client feedback chasing | Drafts the nudge after a set number of days | Chooses tone, sends | Low |
| 3 | Interview notes to the record | Turns call notes into a clean summary | Checks facts, adds judgment | Low to medium |
| 4 | CV intake | Reads a CV and fills fields | Spot-checks the fields | Medium |
| 5 | Job ad drafting | First draft from the brief | Edits for accuracy and fairness | Medium |
| 6 | Screening and ranking | Last, and only as a suggestion | All decisions about people | High |
Follow-up in practice
Here is a simple follow-up rhythm that a small desk can run. The timings are mine, so adjust them to your market.
- Candidate applies or is submitted: an acknowledgment goes out the same day.
- Candidate interviewed with a client: a "where we are" note goes out within two working days, even if the answer is "no news yet".
- Client silent for three working days after an interview: a nudge to the client contact is drafted for the recruiter.
- Offer or rejection: a person writes it, with AI used only to tidy the wording.
The AI reads the record (name, role, stage, date, notes) and drafts the message. The recruiter reads, adjusts and sends. Over time, a tool connected to your ATS can raise a reminder when a record has sat in one stage too long. Connecting the two systems is where a recruitment CRM matters: if your database has clean stages and dates, automation has something to work from. If it does not, fix that first.
CV intake
CVs come in as PDFs, Word files, links and odd layouts. AI is good at reading these and pulling out name, contact details, current title, employers and dates. This is the same family of job as AI document processing.
The failure mode is quiet. A model can misread a date range, merge two jobs or guess a missing detail. So do two things: spot-check a sample of parsed CVs against the originals every week, and never let a parsed field be the reason a candidate is dropped from a search.
The review rule that keeps strong candidates from being lost
The biggest risk is not a rude email. It is a good candidate silently filtered out by a tool that no one checks. My rule:
- AI may sort, summarize and suggest.
- AI may never reject, hide or rank a person without a recruiter looking at the full list.
- Every "no" to a candidate is typed or approved by a person.
- Once a month, a recruiter reads a sample of candidates the tool ranked low and asks whether any should have been higher.
The second reason for this rule is legal. In New York City, Local Law 144 says employers and employment agencies may not use an automated employment decision tool (software that scores or ranks candidates) unless it has had a bias audit within the past year, and certain notices must be given to candidates. The city's consumer protection department (DCWP) enforces it. The European Union's AI Act also lists AI used for recruitment or selection, including filtering applications and evaluating candidates, in its high-risk category. Other places have their own rules and they are changing. If you place candidates in or from these regions, or you use any ranking tool, ask a lawyer before you turn it on. I am not giving legal advice here.
Drafting messages and filling fields does not fall in the same category as scoring people, but a tool vendor may blur the line. Ask any vendor in plain words: does this tool score, rank or filter candidates, and what has been done to test it for bias?
A one-desk pilot
Do not roll out to the whole agency. Pick one recruiter and one desk, and run the pilot for four weeks.
- Week 1: measure. Track time spent on follow-up and data entry for that recruiter, using a rough daily log.
- Week 2: turn on AI drafts for status updates and client nudges only. The recruiter reviews every one.
- Week 3: add interview note summaries.
- Week 4: compare. Time per day on admin, replies sent within two working days, candidates who dropped out, client complaints. Count drafts that needed rewriting.
If the numbers do not improve, stop and find out why before spreading it.
What it costs
Give a pilot a software line and a time line.
- Software: as of November 2026, Claude Team, the business plan of the Claude AI assistant, is listed at $20 per seat a month on annual billing, with a minimum of 2 seats, so a small pilot is about $40 a month. Zapier, a tool that links apps so one event triggers another, lists its Professional plan as starting from $19.99 a month, with the price rising as you pick a higher monthly task allowance and depending on whether you pay monthly or yearly. Together that is roughly $60 a month at the low end, before any other tools. Check each vendor's pricing page for current prices, since they change.
- Time: a recruiter spending an extra 10 minutes a day reviewing drafts is the real cost. Count it in the pilot.
- Setup: connecting AI to your ATS may need help. A direct integration depends on whether your ATS exposes an API (a way for other software to read and write its data).
For a wider look at where automation fits in an agency, see HR automation and AI workflow automation for small business.
What about finding new clients
Business development follow-up is the same pattern: a person decides who to approach, and AI drafts and schedules the touches. My guide to the AI sales development rep (AI SDR) covers what that can and cannot do. For a recruitment agency, I would keep outreach to your own clients and warm leads, in your own voice, before trying mass cold outreach.
Candidate data
CVs are personal data. Use business plans where the vendor says it does not train on your data. Anthropic, the company behind Claude, states that for its commercial products (Team and Enterprise plans and its API) it will not train its models on your chats unless you join its Development Partner Program or explicitly opt in. Check the equivalent for every tool, write down the answer and tell candidates how their data is used in your privacy notice. Rules differ by country, so ask a lawyer where it matters.
How I'd approach this
I would not buy an "AI recruiter" product. I would take one recruiter, one set of status messages and one weekly metric, and see whether admin time drops. Follow-up wins first, screening last, if ever.
I would also keep the human rule absolute. If your agency's brand is that real people look after candidates, then the AI's job is to give your recruiters more time to be those people. A candidate who gets a quick, accurate update from a named recruiter remembers it. A candidate who gets an obviously robotic rejection does not return.
If your database is messy, fix the records first. Automation on bad data is just faster mistakes.
Questions people ask
Can AI screen CVs for a small recruitment agency?
It can read and summarize CVs, but I would not let it rank or reject candidates. In some places, such as New York City, tools that score or filter candidates need a bias audit and notices. The EU AI Act lists recruitment and selection tools as high risk. Use AI for intake and summaries, and let a recruiter decide who moves on. Ask a lawyer before using any ranking tool.
What should a recruitment agency automate first?
Candidate status updates and client feedback chasing. They are repetitive, dated and low risk. After that, interview note summaries and CV intake. Your recruitment CRM needs clean stages and dates for any of this to work well.
Will AI replace recruiters?
Not the part clients and candidates pay for: judgment about people, relationships and negotiation. It can take the typing and chasing off a recruiter's day. That is a good use of it.
How much does a one-desk AI pilot cost?
Software can start around $60 a month for two assistant seats and an automation tool, based on published starting prices as of November 2026, but the larger cost is recruiter time to review drafts and any setup to connect your ATS. See my AI implementation guide for how rollouts are staged.
Is it safe to put candidate CVs into an AI tool?
Use a business plan, check that the vendor says it does not train on your data, tell candidates in your privacy notice and limit who has access. Data rules differ by country, so ask a lawyer where it matters.
Can AI write job ads for me?
Yes, as a first draft from the client brief. A recruiter should edit for accuracy, remove wording that could discourage some groups of applicants and check pay and location details. The ad is your agency's promise to the market, so a person owns it.
More in this topic: Problem solving: AI strategy and adoption (24 guides) →