When founders ask us "is an AI agent cheaper than hiring people," they almost always start the math wrong. They count salary × headcount and stop there. The actual loaded cost of a Polish call-centre operator in 2026 is roughly 2.4× the gross salary on the contract.
Here's the real math from 4 of our Polish clients (retail chain in Tricity, fintech in Warsaw, 3PL in Łódź, dental network in Kraków). All have call-centres of 6–12 people, all are now running AI agents on the first line.
What CFOs see vs. what it actually costs
| Line item | Naive view (PLN/mo) | Reality (PLN/mo) |
|---|---|---|
| 8 operators × 8 000 gross salary | 64 000 | 64 000 |
| Employer ZUS + payroll tax (~20%) | 0 | 12 800 |
| Holiday & sick leave reserve (~9%) | 0 | 5 760 |
| Manager (1 per 8 operators, ~14K) | 0 | 14 000 |
| Office space (10 m² × 8 × 80 PLN/m²) | 0 | 6 400 |
| Equipment & licenses | 0 | 2 800 |
| Training new hires (40% turnover/yr) | 0 | 4 200 |
| Recruiting (1.5 × monthly salary on hire) | 0 | 4 000 |
| Software (CRM, telephony, headsets) | 0 | 5 000 |
| Lost calls outside business hours | 0 | 8 000+ |
| TOTAL loaded cost | 64 000 | 127 000+ |
The hidden killers nobody puts on the spreadsheet
Operator turnover and replacement cost
Polish call-centre turnover in 2025 averaged 38% per year (PwC HR survey). For an 8-person team that's 3 leavers/year. Replacing an operator: 4 weeks recruiting (€800), 6 weeks ramp-up at 50% productivity (cost: 0.5 × 8000 × 1.5 = €6 000 lost output), plus the new-hire onboarding eats 40 hours of senior team time.
Total cost per replacement: €7 000–€9 000. For 3 leavers/year = €21–27K of pure overhead.
Manager overhead
Standard ratio: 1 supervisor per 8 operators. The supervisor doesn't take calls — they handle escalations, write rotas, run 1-on-1s, ship monthly reports. Their loaded cost is ~€18K/mo. That's €2 250 per operator just in management overhead.
Lost calls outside hours
For a typical Polish SMB with 800 calls/month, ~14% come in outside business hours (eve, weekends, holidays). That's 112 calls. With 22% conversion rate to a sale and €280 average deal — that's €6 900/mo of revenue NOT captured. Most CFOs don't see this because there's no line item for "calls we never picked up."
Three deployment models — actual numbers
Here's what changes when you put AI on the first line. Numbers are real averages across our 4 Polish clients in 2024–2025.
| Model | Monthly cost | Calls handled | Cost/call | Quality (CSAT) |
|---|---|---|---|---|
| 8 operators, no AI | 127 000 PLN | 6 400 calls | 19.8 PLN | 4.2 / 5 |
| AI-only on first line | 14 000 PLN | 6 400 calls | 2.2 PLN | 4.0 / 5 |
| Hybrid (AI + 3 operators) | 52 000 PLN | 6 400 calls | 8.1 PLN | 4.5 / 5 |
When AI is NOT the right answer
Three scenarios where we tell clients to keep humans:
- —Call volume below 200/mo — setup cost > savings. Unless you need 24/7 coverage for one specific scenario.
- —Highly emotional contexts (medical bad news, debt collection escalation, bereavement) — humans are required by law in some EU countries and ethically expected.
- —Complex multi-stakeholder negotiations — voice AI is great at intent, terrible at multi-turn business reasoning under pressure.
A CFO-defendable ROI calculation
For an 8-operator call-centre handling 6 400 calls/mo:
ts// Hybrid model (AI + 3 operators) current_cost = 127_000 // PLN/mo, fully loaded new_cost = 52_000 // PLN/mo (AI infra + 3 ops loaded) monthly_saving = current_cost - new_cost // 75_000 PLN annual_saving = 12 * monthly_saving // 900_000 PLN setup_cost = 60_000 // one-off (build + integrate + train) payback_months = setup_cost / monthly_saving // 0.8 months // Plus the calls you start picking up after-hours new_revenue = 800 * 0.14 * 0.22 * 280 // ~6 900 PLN/mo // → annual ROI: (900_000 + 82_800 - 60_000) / 60_000 = 15.4×
How to start without big upfront commit
We don't recommend ripping out a whole call-centre. The smart path:
- —Week 1–2: AI on after-hours and weekends only. Zero firing risk; you only catch calls humans miss.
- —Week 3–6: AI as first responder during business hours; routes to humans on intent unclear or after 2 turns.
- —Week 7–12: Reduce operator headcount by attrition, not layoffs. Most call-centres lose 1 person/quarter anyway.
- —Month 4+: Re-allocate 2–3 operators to revenue functions (outbound sales, account expansion). Now they're a profit center, not a cost center.
This path lets the CFO see savings month-over-month without Day-1 disruption. By month 4 you're saving 50K PLN/mo and the call-centre manager is now driving outbound revenue.