AI Automation: how I cut 40 hours a week down to 4
Eighteen months ago, a recruiting consultancy in Buenos Aires hired me to “write a few scripts.” We ended up replacing two full-time positions with a system that runs 24/7.
This is the real case, with numbers, tools, and the mistakes we made.
The problem before automation
The consultancy received:
- ~500 emails daily from candidates with attached CVs
- 120 WhatsApp messages from clients asking for updates
- 40 interview requests via LinkedIn
- 3 weekly reports done manually in Excel for each client
Three people spent 40 hours a week on tasks that were 90% classification, extraction, and formatting.
The final architecture
Inputs (Email, WhatsApp, LinkedIn)
↓
[n8n] Trigger + Normalization
↓
[Python] Data extraction with LLM
↓
[Supabase] Storage + classification
↓
[n8n] Decision tree + Actions
↓
Outputs (Auto email, Notification, Report)Step 1: Ingestion and normalization with n8n
n8n (self-hosted on Railway) connects:
- IMAP for Gmail and Outlook emails
- WhatsApp Business API via 360dialog
- LinkedIn via API + PhantomBuster webhooks
Each input is normalized to a standard JSON:
{
"source": "email",
"sender": "juan@example.com",
"raw_content": "...",
"attachments": ["cv_juan.pdf"],
"timestamp": "2025-04-15T10:30:00Z"
}Mistake we made: we tried to process PDFs directly in n8n. It’s slow and expensive. The solution was to save the file to S3, send the URL to a Python function, and have the workflow continue via webhook.
Step 2: Structured extraction with LLM
We use OpenAI GPT-4o-mini via API. It’s not the largest model; for structured extraction you don’t need more.
Prompt template (simplified):
system_prompt = """
Extract CV information into this JSON schema:
- name: string
- email: string
- phone: string
- years_experience: number
- technologies: string[]
- seniority: "junior" | "mid" | "senior" | "lead"
- availability: "immediate" | "1_month" | "3_months"
If information is missing, use null. Don't make up data.
"""Real cost: ~USD 0.003 per CV processed. With 500 CVs daily = USD 4.50/day.
Compared to a junior analyst (USD 1,200/month), ROI is reached by day 5.
Step 3: Classification and routing
Once structured, candidates are classified:
| Score | Action |
|---|---|
| > 85% match | Automatic email to candidate + recruiter notification |
| 60-85% | Save to manual shortlist for review |
| < 60% | Automatic thank-you email + generic feedback |
The match criterion comes from a job description parsed with the same LLM.
Step 4: Automatic reports
Every Friday at 6:00 PM, n8n triggers:
- Query Supabase for the week’s candidates
- Aggregation with Python (pandas + jinja2)
- PDF generation with Playwright + HTML template
- Email to each client with recommended candidates
Previous time: 3 hours per client × 12 clients = 36 hours/week. Current time: 0 hours. The system hasn’t failed in 8 months.
The 3 things we didn’t automate
Not everything should be automatic. We left these out of the pipeline:
- Technical interviews: AI evaluates code, but not cultural fit
- Salary negotiation: requires human judgment and emotional context
- Formal complaints: an automated angry email makes everything worse
Stack and real monthly costs
| Service | Cost/month | Role |
|---|---|---|
| n8n (Railway) | USD 19 | Orchestration |
| OpenAI API | USD 135 | Extraction + classification |
| Supabase | USD 25 | Database + Auth |
| S3 (R2) | USD 8 | CVs + reports |
| PhantomBuster | USD 48 | LinkedIn scraping |
| 360dialog | USD 55 | WhatsApp API |
| Total | USD 290 |
Replaces USD 2,400/month in salaries + 36 hours/week of a founder.
Start today: automation MVP in 48 hours
If you want to prototype fast:
- n8n cloud (free up to 5 workflows)
- Zapier or Make if you don’t know code
- OpenAI API with simple prompts
- Google Sheets as a temporary database
Start with ONE single flow that bothers you. Don’t try to automate everything at once.
Conclusion
AI doesn’t replace recruiters. It replaces manual classification, copy-paste from Excel, and waiting for someone to read an email.
The 36 recovered hours are used in quality interviews, client negotiation, and growth strategy.
Want to see if your workflow can be automated? Get in touch and in 15 minutes we’ll tell you which parts have positive ROI.