AI customer service needs a clear route to a person

AI customer service works better when customers have a clear path to resolving their requests. That includes reaching a person when judgment is needed, information is missing or human assistance is requested. The handoff should carry the conversation's context and explain the next step without promising a resolution that has not been approved.

An immediate response can reduce the initial wait. But for someone who received the wrong product or needs a quotation corrected, the problem ends when an appropriate action is completed. Service design should therefore begin by defining what resolution means for each type of request.

The Klarna case through two news sources

On September 10, 2025, Reuters reported that Klarna was reconsidering its emphasis on AI cost reductions to give service more attention. Bloomberg reporting published by Folha on October 10 highlighted renewed customer service hiring to preserve the option of speaking to a person.

The first story emphasizes changing priorities; the second places support within a wider discussion of employment. Both concern the same company and describe its continued belief in AI. Our interpretation is that the reporting supports examining service design, without concluding that every form of service automation has failed.

Define when a conversation should be handed over

A small team can begin with a list of recognizable reasons for escalation. Test the list against conversations authorized for this purpose, removing information that is unnecessary. The people who handle requests every day should help define the criteria because they understand exceptions that forms do not capture.

• The customer explicitly asks to speak with a person.
• The answer depends on information that is unavailable or contradictory.
• The same guidance has already been given and the problem remains unresolved.
• The request involves a commercial exception that automation may not approve.
• A complaint requires investigation or coordination between departments.

These criteria are an operating proposal, rather than a universal rule. A repair business, a school and a professional services company will have different needs. The important point is that the handoff has an owner and destination, instead of merely ending the automated conversation.

Send the context along with the customer

The team member needs a short record of why the customer contacted the business, the action requested, what has already been attempted and what remains unconfirmed. A summary can help, but the original conversation should remain accessible to authorized staff so they can check details before deciding.

Avoid asking again for information that has already been provided and remains valid. If identity needs confirmation or a detail needs correction, explain why in plain language. State support hours and a realistic response expectation, especially when the team operates in another time zone.

A possible workflow for a service company

In a hypothetical example, an installation business receives a request through its website to change an appointment. Automated support can identify the request and explain the available channels. If rescheduling requires checking a technician's calendar, the request goes to the responsible person with the preferred time and conversation history attached.

The conversation should describe a new appointment as confirmed only after the company's process has confirmed it. Until then, the customer receives an acknowledgment that the request is under review. This wording distinction prevents an intention from becoming an unsupported commercial promise.

Measure resolution over several days

Track the time to first response, time to completed action and repeat contacts about the same issue. Also record how many handoffs reached the correct queue and how many were left without an owner. These measures help reveal whether a quick initial response is simply moving work elsewhere.

Read a sample of closed cases and check whether the outcome was clear to the customer. A short conversation may indicate efficiency or abandonment. Without looking at the result, a business could celebrate fewer messages while overlooking a loss of trust.

Review the figures with the support team, not only with the people maintaining the tool. When an unusual case becomes common, change the process or the published information. When a handoff fails, identify where ownership was lost before adding another automated response.

Improve the offer when the same question returns

Repeated questions about timing, pricing and scope also identify weaknesses in the website and commercial proposal. Vero 4 recommends using those records to improve information available before purchase. Customer service becomes more useful when it helps the business explain its offer clearly and deliver what it has agreed.

Sources and further reading

• Reuters, via WHBL, 2025-09-10.
• Bloomberg, via Folha de S.Paulo, 2025-10-10.