An Automated Service Should Hand Over the Uncertainty

A design essay on escalation: a human needs the unresolved question, relevant context and authority to act, rather than a forwarded conversation alone.

A handover is more than a change of speaker

An automated service can announce that a person will take over while transferring very little that makes the next interaction useful. The customer repeats the account, the staff member searches for the relevant detail and the underlying uncertainty remains unexplained. A human has entered the exchange, but the decision has not become easier to make.

The design question is what should cross the boundary between automation and human responsibility. This essay proposes that the handover should transfer the unresolved question, the evidence already obtained and the authority required to resolve it. It does not claim that every service can eliminate waiting or every difficult case can be settled immediately. A good escalation makes the difficulty intelligible instead of concealing it behind a change of channel.

Describe why the case cannot continue automatically

A system may lack information, encounter conflicting records or reach a decision it is not authorised to make. These are different reasons for escalation. A staff member needs to know which one applies before deciding whether to request another document, correct a record or seek approval from someone else.

Consider a hypothetical booking service where the customer has a confirmation but the supplier’s record shows no reservation. Forwarding the entire conversation leaves the human to reconstruct the issue. A precise handover identifies the two conflicting records, the last verified step and what has not been established. The difference is operational: the human can investigate the inconsistency rather than spend the opening of the interaction discovering that one exists.

Preserve evidence without carrying irrelevant material

Useful context should be selected for the decision being transferred. A long transcript may contain the necessary evidence while making it difficult to locate. A summary may be easier to read while omitting the sentence that changes the case. The design should preserve access to the relevant source and distinguish a summary from the underlying record.

Selection also requires restraint. The handover should not carry unrelated personal details merely because the system encountered them. Its purpose is to support the next authorised action. A practical structure identifies verified facts, customer statements, system interpretations and open questions separately. This makes it possible to correct an interpretation without pretending that every part of the record has the same evidential status.

Match the destination to the authority required

Escalation is incomplete if the receiving person can only restate the same limitation. A case concerning a discretionary refund needs a route to the relevant approval. A disputed identity record may need a different team from a routine service enquiry. Assigning the case by subject alone can miss the authority that actually determines whether progress is possible.

The organisation should define what each receiving role can decide and how a further escalation is initiated. This is not a demand to grant broad discretion to every staff member. It is an argument for making the route through existing authority clear. The customer should not have to discover the organisation chart by repeatedly explaining the problem to people unable to act on it.

Do not convert an estimate into a promise

A handover often includes an expectation about when someone will respond. That expectation should reflect the actual service arrangement. An automated message can create a new obligation if it confidently promises a resolution that the receiving team has neither accepted nor the capacity to provide.

A more responsible message identifies the next step and any uncertainty about its timing. If the organisation has a defined response commitment, communicate it accurately. If more information is required before a timetable can be given, say what is missing. The goal is not to sound less helpful. It is to avoid manufacturing reassurance at the precise moment when a customer already needs a reliable account of what will happen next.

Allow the human to correct the route

The receiving person may discover that the system classified the issue incorrectly. A useful process lets that correction change the case rather than force the human to work around a fixed category. It should also preserve why the route changed, so the next person does not repeat the original assumption.

For example, a hypothetical billing complaint may turn out to concern an order that was never fulfilled. The relevant decision then changes from explaining a charge to establishing delivery and the available remedy. A rigid escalation label can delay that recognition. Human involvement deserves more than a ceremonial opportunity to speak: it needs room to revise the description of the problem when the evidence supports a different one.

Review whether the uncertainty was resolved

Counting a case as handed over answers whether it left one queue. It does not establish whether anyone acquired the information and authority needed to complete it. A review should therefore follow the decision: what remained unclear, who accepted responsibility and what action eventually addressed the issue?

Repeated failures can identify a missing document, an ambiguous policy or a destination without sufficient authority. Those are design questions rather than automatic evidence that the human or the automated system performed badly. A service becomes more coherent when escalation transfers responsibility in a form another person can use. The next voice should inherit a defined problem and a workable route, rather than merely the customer’s accumulated frustration.

Gabriel Nicolaev — CyGuru

Editorial photograph for illustration.

FOLLOW THE CONNECTION

Another angle on this subject

Continue through related ideas, decisions and consequences.

World & Power

Closing Public Infrastructure Is a Public Decision, Too

A policy essay on retiring public infrastructure: distinguish an asset that should close from a service obligation that still needs an accountable future.

Human Behaviour

A Disagreement Needs a Question That Could Change Your Mind

An opinion essay on distinguishing a conversation that can produce understanding from an argument arranged to protect a verdict.

Technology & AI

An AI Agent Should Not Inherit Every Permission Its User Holds

The next operating question for AI agents is how much authority each task actually requires, and how a mistaken action can be contained.