AI in Human Resources: Automate Recruiting and Internal Queries

If your HR team spends more time answering "how many vacation days do I have left?" than doing actual people work, this article is going to sting a little. AI in human resources automates three concrete areas: screening CVs against the target profile, guiding the onboarding of new employees, and handling repetitive internal queries about leave, payslips, and policies. AI pre-screens and assists; decisions about people stay human. At Deepyze we implement these systems for companies across LATAM, and the numbers are consistent: between 30% and 40% of an HR team's time goes to tasks a machine does better.

The hidden cost: the same questions, every month

Let's start with the least glamorous but most profitable area. In a 100-employee company, HR receives between 150 and 300 internal queries per month. The vast majority are always the same:

• "How many vacation days do I have left?"
• "Where do I download my payslip?"
• "How do I request leave for an exam / moving / a family member's illness?"
• "What does the health plan / medical insurance cover?"
• "What's the policy on remote work / per diems / overtime?"

At a conservative average of 5 minutes per query (finding the answer, writing it up, replying), that's 12 to 25 hours monthly of a qualified person acting as a human search engine. In companies with 300+ employees, that number triples and usually justifies an entire administrative role.

How do you eliminate it? With an internal assistant built on RAG (Retrieval-Augmented Generation): the AI answers by consulting your policies, your collective agreement, and your procedures — it doesn't make things up, it cites the source. If a question exceeds its knowledge base or involves a sensitive case, it hands off to a human with the context already assembled. In real deployments, this assistant resolves 70% to 85% of queries on its own. Those 20 monthly hours don't vanish from payroll: they get reinvested in interviews, culture, and development — which is what you hired HR to do.

If you want to understand the technical mechanics behind this, we explain it in what RAG is and how a knowledge base works.

AI CV screening: speed without the black box

A job posting on LATAM job boards receives between 100 and 400 applications. Reviewing each CV takes 3 to 5 minutes if done seriously: that's 5 to 30 hours per search, and the reality is that after CV number 50, nobody evaluates the same way.

AI screening works like this:

1. You define the profile with explicit criteria: must-have skills, minimum experience, languages, location.
2. The AI reads each CV (PDF, Word, job-board text) and scores it against those criteria, with a written justification for each score.
3. HR receives a ranked list with the reasoning in plain view and decides who to interview.

The critical point — and where many off-the-shelf tools fail — is bias. A model that learns from "who we hired before" replicates the company's historical biases. That's why the right approach is to evaluate against job requirements, never against past hiring patterns, and to maintain traceability: every automatic rejection must be explainable. Done well, AI screening is less biased than human screening, because applicant number 300 gets the same attention as the first, at 9 in the morning or 7 in the evening.

Want to know how many hours per month your HR team would recover? Book a 30-minute call and we'll build the diagnosis around your real numbers, at no cost.

Guided onboarding: so day 1 doesn't depend on anyone's memory

Onboarding is the most standardizable HR process and the one most often done half-heartedly. A typical new hire involves 15 to 30 steps spread across HR, IT, and the team lead: system accounts, equipment handover, signing paperwork, mandatory training, introductions.

Automated, the flow looks like this: the new employee receives a guided sequence (by email, WhatsApp, or Slack) that delivers what they need each day, requests any pending documentation, and answers questions with the same RAG assistant. On the internal side, each step triggers the corresponding tasks — user account creation, equipment assignment, a reminder to the team lead — without anyone having to chase anyone. Companies that implement it reduce administrative time per hire from 6-10 hours to under 2, and the new employee stops spending their first week waiting for access.

What to automate first: the decision table

Area	Typical hours/month (100-employee company)	% the AI absorbs	Implementation cost (USD)
Internal queries (RAG assistant)	12-25 hrs	70-85%	3,000-8,000
CV screening	5-30 hrs per search	60-80% of initial filtering	2,500-6,000
Guided onboarding	6-10 hrs per hire	70-90%	2,000-5,000
Engagement surveys + analysis	4-8 hrs	50-70%	1,500-4,000

The advice we always give: start with the internal query assistant. It has the fastest return, the lowest risk (it doesn't touch decisions about people), and it builds the internal trust you need to tackle the rest. We cover all three areas in our AI for HR service.

The ethical and legal limit: AI doesn't decide about people

This isn't fine print, it's system design. Our rule on every AI in human resources project:

• AI pre-screens, ranks, and assists. A human decides. Hiring, terminating, promoting, or disciplining never comes out of a model.
• Every automatic score is explainable and auditable. If a candidate asks why they were rejected, there's a concrete answer.
• Employee and candidate data does not train third-party models. We use APIs with no-training agreements or models on our own infrastructure.
• The employee knows when they're talking to an AI and always has a direct path to a person.

Regulation is moving in that direction: the European AI Act classifies employment as a high-risk use, and several LATAM countries are replicating the mandatory human-oversight criterion. Designing this way from day one isn't just ethical: it saves you from rewriting the system in two years.

When AI in HR is NOT worth it

Honesty first:

• If you have fewer than 30 employees, the query volume rarely justifies a dedicated assistant. A well-built FAQ document and a simple chatbot shared with customer support may be enough.
• If your policies aren't written down anywhere, there's no RAG possible: the AI can't consult what doesn't exist. Document first (it takes 2-3 weeks), automate later.
• If your selection process is broken — poorly defined profiles, searches that change halfway through — the AI will filter fast against the wrong criteria. Automating a broken process just gives you faster errors.
• If you're looking to replace the HR team, you're reading the wrong article: this multiplies people, it doesn't substitute them.

How to get started

The path we recommend: pick one area (internal queries is the best first step), gather the documentation you already have, and run a 30-day pilot with clear metrics — queries resolved without human intervention, response time, internal satisfaction. It's the same logic we apply to any AI automation project, and we walk through it step by step in how to implement AI in your SMB.

If you'd rather do it with support: at Deepyze we design and implement HR assistants, screening, and automated onboarding for companies in Argentina and across LATAM, with fixed pricing, a team in your time zone, and post-implementation support. Tell us about your case and within 24 hours you'll have a concrete proposal.