Nedbank Chat Bot

Nedbank wanted to expand how it engages customers, and chat was the most efficient channel available. The project brought an existing chat service together with the bank's own app into a single conversational experience.

The intent was concierge rather than chatbot: an assistant that can move a customer between services without exposing the seam between the systems providing them.

The design work was conversational architecture first, and a component library second.

Incorporate an existing chat app into the Nedbank Money app and build the new UI components into the current library — for web, mobile and tablet.

A conversation has no navigation, so capability has to be surfaced in the dialogue itself. We used suggested prompts and structured replies at the points where a customer would otherwise have to guess, making what the assistant can do a matter of recognition rather than recall.

Continuity was the governing principle: history persists, context carries across a handover between automated and human assistance, and the customer is never asked to repeat information the system already holds. Handovers are stated plainly rather than hidden, because a customer who knows who they are talking to asks better questions.

Error recovery was designed as a normal path, not an exception. When intent isn't understood, the assistant narrows rather than apologising — offering the closest options it can act on — and always leaves a route to a person.

We worked closely with the content team and briefed the agents chatting with customers, using those real conversations to shape the conversational patterns that were later encoded into the bot.

New components — message types, structured cards, prompt chips, status and typing indicators — were built into Nedbank's existing library rather than alongside it, so the chat inherits the bank's type scale, spacing and colour semantics.

Because responses originate in different underlying services, the presentation layer had to be strictly consistent: the same card structure and the same visual weight regardless of source. Responsive behaviour was specified across web, mobile and tablet from the same component set.