The Chat Window That Opens Itself And Is Not A Person

A bubble appears in the corner asking how it can help. It cannot help, and after several exchanges it offers a link to the page you were already reading.

Support is measured by contacts avoided

The central metric in customer service operations is the number of enquiries resolved without reaching a person, variously called deflection or containment.

Every contact that ends inside the automated layer is recorded as a success, whether the customer's problem was solved or the customer simply gave up.

Those two outcomes are difficult to distinguish in the data, and the metric does not try particularly hard to separate them.

Cost per contact varies enormously by channel

A telephone call occupies one trained person for its whole duration. A chat conversation lets one person handle several at once. An automated exchange occupies nobody.

The gaps between those costs are large, which is why every channel that touches a human gets progressively harder to reach as an organisation grows.

The savings are real and immediate, while the cost of frustrated customers is diffuse, delayed and lands on a different department's numbers.

Triage looks identical to obstruction

The bot's genuine function is sorting. Password resets, delivery tracking and opening hours really are better handled automatically, and most enquiries are one of those.

The system is tuned for that majority, which means the unusual problem is repeatedly matched to the nearest common one and answered incorrectly.

From inside the conversation there is no way to signal that your case is unusual, because the mechanism for expressing that is the human you are trying to reach.

Hiding the phone number is a design decision

Contact routes are arranged as a funnel, with self-service articles first, automated chat next, email after that and the telephone number several layers deep or absent entirely.

Each layer removes a share of the traffic, and the ordering is chosen to maximise that removal rather than to match how urgent a problem might be.

Users respond by learning the shortcuts, typing the word for a human repeatedly, which the system then adds to its list of things to handle automatically.

The cost moved rather than disappeared

Time spent explaining a problem three times to a system that cannot act on it is real work, performed by the customer and recorded nowhere.

Because it appears on no balance sheet, there is no internal pressure to reduce it, and the deflection figure keeps improving as the experience deteriorates.

Organisations that treat resolution rather than containment as the target end up with fewer contacts anyway, because the problems get fixed at the source.