Skip to content

Customer service20:05

AI will replace up to 45% of customer service tickets, not the entire department

How many tickets will AI resolve without human involvement in Polish e-commerce, telco, and B2B support? We provide thresholds from deployments, costs, and the legal limits of human oversight.

AI-generated image.

Listen to the article summary

Synthetic voice.
0:00
0:00

Treat this text as practical notes shared for free: we show how things usually unfold, not how they must look for your specific case. The claim that AI will take over customer service jobs has circulated in boardrooms since Klarna announced an assistant that handled 2.3 million conversations in a month. It sounds concrete: we have the technology, we have the case study, we have the savings. And this claim is not entirely false. It is, however, poorly calibrated. That calibration makes the difference between reducing 20 roles and losing customers who can never be won back.

The strongest counterargument

To state the argument fairly: Klarna announced that its OpenAI-based AI assistant handled two-thirds of customer service chats, matching the work of 700 full-time agents. Average resolution time dropped from 11 minutes to under 2 minutes, and repeat inquiries fell by 25%. On the Polish market, a study by EY-Parthenon and PSMB showed that 83% of contact centers report cost reductions and 60% report shorter calls. You cannot simply dismiss the claim.

AI-generated image.

What this argument overlooks

The argument relies on two silent assumptions: that deflection means a resolved issue and that the savings generate no secondary costs. In practice, we measure the reopen rate: how many tickets return after being closed by AI. For our clients, it ranges from 3 to 12%, with a pain threshold at 5%. Above this threshold, savings are consumed by the secondary human touchpoint. Then comes the law. GDPR Art. 22 requires human oversight for automated decisions, and AI Act Art. 50 mandates informing customers that they are speaking with a machine. An incorrect chatbot answer about price or cancellation rights is the company's liability, not the model's. And Klarna itself: in May 2025, it started hiring customer service staff again. They did not drop AI, but the hybrid model stayed.

Our position

AI does not replace a customer service department. It replaces repetitive tickets and first-line support with a clear escalation threshold. For us, this starts with a 6-week audit of tickets by category. In B2C e-commerce, the automation potential is 30–45% of tickets, in telecom and finance 20–35%, and in B2B technical support only 10–20%. These thresholds stem from our implementations in companies with 40 to 400 employees, not from sales decks. The cost per AI interaction is 0.5–2 PLN net, while an agent conversation costs 8–20 PLN. However, TCO must include model maintenance, integrations, audits, and escalations. Without this, the calculation is fictional.

AI-generated image.

What this means this quarter

For this quarter's decisions, a 6-week audit and ticket categorization usually prove effective: order status, FAQ, address changes, and return labels are typical areas for AI. Quality complaints, fee disputes, credit decisions, and vulnerable customers stay with humans. An 8–12-week pilot on a single channel reveals whether the reopen rate stays below 5% and escalations do not exceed 20–25%. We would revise our stance if repetitive categories made up over 60% of volume in the audit and the reopen rate stayed below 2% after three quarters. So far, we have not seen this in our implementations.

Source materials

Sources consulted during research. The text above is our own; these third parties are not responsible for its content and have not authorised it.

No-obligation call

Schedule a workshop with Paul

Thirty minutes or a full workshop, choose what suits you. Book a slot directly in the calendar and get instant confirmation.

Paul Lazniak

HEXART Founder