AI Agent in HR: Real Case Study of RAG Implementation and Onboarding Automation in Ukraine

Why Onboarding Still 'Eats Up' Resources in Ukrainian Companies

According to SHRM, companies spend an average of 6 to 9 months of an employee's salary just to replace them after they leave. A significant portion of these costs results from poor onboarding. In Ukrainian small and medium-sized companies, the situation is further complicated by staff shortages: an HR manager simultaneously handles recruitment, document processing, training, and twenty other parallel tasks. This is exactly why an AI agent in HR has transformed from a trendy buzzword into a real survival tool.

In this article, we'll break down a concrete case: how a small Ukrainian IT services company implemented a RAG system (Retrieval-Augmented Generation) and automated the entire onboarding cycle — from a new employee's first day through the end of their probation period. You'll get the numbers, architectural decisions, and practical conclusions that you can adapt for your own business.

---

What is RAG and Why It's Critical for HR Automation

The Problem of 'Hallucinations' in Corporate Context

When you ask a regular language model to answer a new employee's question — for example, "what's the vacation approval procedure?" — it either makes up a plausible but false answer or says it doesn't have access to internal documents. Both scenarios are catastrophic for HR.

RAG (Retrieval-Augmented Generation) solves this problem by combining two components:

• Vector knowledge base — all internal company documents (regulations, instructions, templates, policies) are converted into numerical vectors and stored in a specialized database
• Language model — receives not a general task, but a specific excerpt from a real document and formulates an answer based on it

We discussed the architectural principles of this approach in more detail in the article Hybrid RAG Architecture: Precise Document Search and Protecting AI Agents from Hallucinations. In short: the system doesn't "make things up" — it cites. This is fundamentally important for legal and HR documents.

Why This Is Particularly Relevant for the Ukrainian Market

Ukrainian companies often have documentation in chaotic condition: some regulations in Google Drive, some in Notion, some instructions exist only "in the head" of an experienced employee. The RAG system forces companies to systematize knowledge — and this itself is valuable, regardless of AI.

---

Real Case Study: IT Services Company, 85 Employees, Kyiv

Initial Conditions and Problems

The company develops and maintains software for EU clients. They hire 3-5 new people every month. Before implementing the AI agent, the picture was:

• One HR manager handled the entire cycle — from job posting to onboarding completion
• A new employee in the first week asked an average of 47 questions (counted manually over three months)
• HR spent 12-15 hours per month solely answering typical new employee questions
• 23% of new employees in the first month felt "lost" and decreased productivity due to lack of information
• Onboarding documentation hadn't been updated for more than 8 months

Architecture of the Implemented Solution

The project was implemented over 6 weeks. Here's what was done:

First stage — audit and knowledge structuring (2 weeks) Collected all internal documents: labor regulations, tool instructions, corporate policies, FAQ on benefits, communication schemes between departments. Total — 143 documents with approximately 380 pages.

Second stage — building the RAG base (1.5 weeks) Documents were split into semantic chunks (fragments), vectorized, and uploaded to the database. For vector search, an open-source solution based on pgvector (PostgreSQL extension) was used — an economical solution for companies that don't want to pay for cloud vector databases.

Third stage — configuring the AI agent (2 weeks) The agent is integrated into corporate Slack. A new employee writes a question in a special channel or directly to the bot — and receives an answer with a link to the specific document from which the information was taken. If a question falls outside the knowledge base, the agent automatically redirects to the HR manager.

To understand how such systems can scale within more complex business processes, it's useful to familiarize yourself with the material Multi-Agent Systems: How Multi-Agent Orchestration Changes Business Automation — it describes the logic of coordinating multiple agents within a single enterprise.

Automated Onboarding Route

Besides answering questions, the agent guides a new employee through a structured route:

• Day 1: Welcome message, first-day checklist, links to all necessary accounts
• Day 3: Reminder to verify access, request meeting with team lead
• Day 7: Micro-survey "how are you feeling?", automatic response analysis
• Day 14: Check on completion of mandatory training modules
• Day 30 and 90: Automatic reminder to HR to review probation results

The entire route is configured through n8n (open-source automation platform) and requires no HR intervention in 80% of cases.

---

Results After 3 Months of Launch

Quantitative Indicators

| Indicator | Before Implementation | After Implementation | Change | |---|---|---|---| | HR time on answering new hires | 12-15 hrs/month | 2-3 hrs/month | -80% | | Number of questions unanswered for over 2 hours | 34% | 6% | -28 p.p. | | Onboarding satisfaction (survey) | 62% | 89% | +27 p.p. | | Time to first 'real' task | 8.5 days | 5.2 days | -39% | | Document processing errors | 11 cases/month | 2 cases/month | -82% |

Financial effect: The HR manager freed up about 130 hours per year — and directed them to strategic recruitment. The company filled 2 open positions that had been vacant for 3 months.

Qualitative Observations

The team noted several unexpected bonuses. First, new employees stopped "tugging" their colleagues with minor questions — this positively affected the atmosphere in departments. Second, the process of creating the knowledge base itself revealed 19 regulations that contradicted each other — a problem that existed for years but nobody noticed.

A third unexpected effect: the HR manager gained analytics on the most popular questions from new hires. It turned out that 40% of questions concerned the expense approval procedure — a signal that this process is too complex and should be simplified.

---

How Much Does Implementing an AI Agent in HR Cost in Ukraine

Cost Structure

One-time implementation costs:

• Audit and documentation systematization: from 15,000 to 40,000 UAH (depends on the level of chaos)
• Development and configuration of RAG system and agent: from 50,000 to 120,000 UAH
• Integration with Slack/Teams/Telegram and corporate tools: from 10,000 to 25,000 UAH

Monthly operational costs:

• API costs for language models (GPT-4o or Claude): from 1,500 to 5,000 UAH depending on load
• Hosting of vector database and automation: from 800 to 2,000 UAH
• Support and knowledge base updates: 2-4 hours of HR per month

For a company with 50-150 employees with active hiring, return on investment (ROI) occurs within 4-8 months. If you want to dive deeper into the economics of such solutions, read AI Automation for Mid-Market Business: How to Grow Without Increasing Headcount.

Where You Can Save

• Use open-source models (Mistral, LLaMA) for internal deployment — reduces API costs to zero but requires a server
• Build the knowledge base gradually, starting with 20-30 most important documents
• Choose Telegram instead of Slack as the interaction channel — cheaper and familiar to Ukrainian teams

Interestingly, some tools your team already uses have built-in AI features — read more about this in the material Notion, Wrike, SAP Joule: Which AI Agents Are Already Built Into the Tools You Use.

---

How to Prepare for Implementation: Step-by-Step Plan for an HR Leader

Step 1: Audit Current State

Before building any AI system, document "as-is": how much time HR spends on routine questions, which questions are asked most frequently, where documents are stored and whether they're current. Without this audit, you risk automating chaos.

Step 2: Define Scope of First Iteration

Don't try to automate everything at once. For the first MVP (minimum viable product), enough:

• 30-50 key documents
• Answers to typical new hire questions
• Basic onboarding route for the first 30 days

Step 3: Choose Knowledge Management Model

Who will update the document base? This is a critical question. An AI agent in HR is useful only as much as its knowledge base is current. Assign a responsible person and establish a regular review cycle (at least once per quarter).

Step 4: Testing with Real New Employees

Before full launch, conduct a 2-week pilot with 2-3 new employees. Gather feedback: which questions couldn't the agent answer, where were answers incorrect, what was missing.

Step 5: Scaling and Measuring Results

Define 3-5 key metrics in advance — and measure them monthly. Without measurements, you won't know if the system works and won't be able to justify investments to business owners.

---

FAQ: Answers to Common Questions About AI Agents in HR

Can an AI agent completely replace an HR manager? No — and that's not the goal. An AI agent in HR automates routine, repetitive tasks: answering typical questions, reminders, checklist verification. Strategic tasks — building culture, complex negotiations, talent development — remain with people. The correct goal: free HR to do work that truly requires human touch.

How to protect confidential HR data when using AI? This is a key question for any business. There are several approaches: deploying the model on your own server (on-premise), using data processing agreements (DPA) with cloud providers, restricting access so the agent only sees depersonalized regulations but not personal employee data. Never upload passport data, medical information, or salary reports to the RAG base.

How long does it take to implement RAG for HR in a small business? For a company up to 50 people with relatively organized documentation — from 4 to 8 weeks from start to first working prototype. Most time goes not to technical development but to auditing and structuring documents. If documentation is chaotic, add 2-3 weeks.

Which language model is better to use for a Ukrainian-language HR agent? GPT-4o and Claude 3.5 Sonnet show the best quality for understanding and generating Ukrainian text among commercial models. For budget solutions, Mistral Large or Gemini 1.5 Flash are suitable. Important: the quality of answers depends primarily on the quality of documents in the base, not just on model choice.

Would such an AI agent work for a company with 10-15 employees? Yes, but with different ROI calculations. At this business size, the main value is not saving HR hours but onboarding quality and knowledge systematization. If you hire even 1-2 people per month, or if the owner themselves performs HR functions — onboarding automation frees up critical attention resource.

---

Conclusion

An AI agent in HR is not the future but already a real tool that Ukrainian companies are implementing today and getting measurable results from: -80% HR time on routine, +27 p.p. in onboarding satisfaction, reducing time to productivity by almost half. RAG architecture makes the agent reliable and accurate even when working with sensitive corporate documentation. If you want to understand how exactly such a system can work in your company — get a free consultation: we'll analyze your processes and suggest a specific implementation plan within your budget and scale.