Why menu-based bots frustrate shoppers, and what replaced them

Chatbots earned their bad reputation honestly. For years, what got installed on websites was a menu tree dressed up as a conversation: press 1 for orders, press 2 for returns, press 3 to go back to the previous menu.

If your question wasn’t on the list, there was no way out. And most real questions weren’t on the list.

The structural problem

A menu bot forces the customer to do the translation work. They arrive with a specific question — “will this fit a 60cm gap?” — and have to guess which of six generic categories the business filed that under.

That translation fails constantly, for a simple reason: the categories were designed around how the business is organised internally, not around how customers think.

Worse, the failure is silent. The customer doesn’t complain that the menu didn’t have their option. They close the tab, and the business sees a resolved session.

What actually changed

Two things, and only one of them is the obvious one.

The obvious one is natural language understanding: the customer can now type the question as they’d say it out loud, and doesn’t need to find the right button.

The less obvious one matters more commercially: the system can now be connected to the catalogue and can show products. A menu bot could only return pre-written text. A modern chatbot for online store can check stock in real time, compare two items and display them inside the conversation, with a route to the cart.

That’s the difference between a support tool and a sales tool.

What still goes wrong

Upgrading the technology doesn’t automatically fix the experience. Three failure modes survived the transition:

Trained on the FAQ, not the catalogue. The system answers fluently with information that was accurate six months ago. Fluent and wrong is worse than clunky and right.

No exit. Natural language understanding with no handover path is a nicer-sounding dead end. Same trap, better vocabulary.

Hidden behind a greeting. A bot that opens with three lines of introduction before letting the customer type is still wasting the window in which the sale exists.

How to evaluate one in ten minutes

Run three tests on any tool you’re considering:

The variant test. Ask about a specific colour or size of a real product. If the answer is generic, it isn’t connected to the catalogue.

The comparison test. Ask which of two products suits a described use case. This is where a sales tool separates from a support tool.

The escape test. Ask for a human. Time it, and check whether you have to repeat anything.

A system that passes all three will pay for itself in a category where customers hesitate. One that fails the third will quietly cost you the conversations that mattered most.