The problem
Take a small recruitment agency filling about eight roles at a time for its clients, receiving around 600 applications a month. Two recruiters do everything: read CVs, call candidates, coordinate interviews with client managers, and keep the client's tracker up to date.
The first read of each CV is where most hours disappear. Many applicants do not meet the basics (location, notice period, a required skill), and the recruiter only finds out after opening the file. Then scheduling: a single interview can take several messages to agree a time between a candidate and a busy hiring manager.
Where the hours go
| Task | By hand | After |
|---|---|---|
| First read of 600 CVs | 1,800 min (3 min each) | 600 min (1 min per summary) |
| Read shortlisted CVs in full | (part of the above) | 450 min (150 at 3 min each) |
| Book 80 interviews | 960 min (12 min each) | 240 min (3 min each) |
| Update client trackers | 480 min | 60 min (check only) |
| Per month | 3,240 min (54 h) | 1,350 min (22.5 h) |
That is 31.5 hours a month back for two recruiters.
What it costs and what it saves
At ₹500 an hour, 31.5 hours is ₹15,750 a month. We assume ₹5,000 a month for AI usage and a scheduling tool. That leaves about ₹10,750 a month, and recruiters spend the time on calls with candidates and clients, which is where placements actually come from.
How we set it up
- Must-haves go in the form. Location or willingness to relocate, notice period, expected salary range and the one or two skills the client insists on are asked directly. A clear answer beats guessing from a CV.
- Each CV gets a summary against the role. An AI model reads the CV next to the written job requirements and returns a short summary: which requirements it meets, which it does not, and the evidence for each, quoted from the CV.
- Recruiters decide. The recruiter reads every summary, opens any CV that looks borderline, and reads every shortlisted CV in full. Nothing is rejected by the model.
- Interviews book themselves. Shortlisted candidates get a link showing times when the client's interviewers are free, with reminders the day before and an easy way to reschedule.
- Trackers stay current. Each stage change updates the client's sheet or ATS, so the weekly update is a check, not a rewrite.
How we keep it fair
Screening is where automation can do real harm, so the rules are strict:
- The model is not given names, photos, age, gender, marital status or addresses, and it is not asked to score college names.
- It only checks the requirements the client wrote down, and it must quote the CV for every claim.
- No candidate is rejected without a recruiter reading their summary.
- Candidates are told how their data is used, and their data is deleted on request, in line with India's Digital Personal Data Protection Act.
- We review a sample of decisions each month to check that the summaries are not missing good people.
Run the numbers for your agency
Count last month's applications and interviews, and time one first read and one scheduling thread honestly. The savings calculator turns that into hours and rupees, and you can add your own tasks.
Questions people ask
Can the AI make the shortlist on its own?
It could, and we do not build it that way. It saves recruiters the slow part of reading, and the decision stays with a person who can be asked why.
What about candidates with unusual careers?
That is exactly why the summaries quote evidence instead of giving a score. A recruiter can see a career change or a gap and judge it, rather than a number hiding it.
Does this work with our ATS?
Most applicant tracking systems have an API or an export. If yours does not, a well-kept Google Sheet works as the tracker, and we connect to that.
Written by Kishan Makani
Founder of Makani Tech, 14 years in the tech industry. Builds AI and automation systems for students, developers and business owners from Surat.