The challenge
The brand was launching a new digital experience for a young audience. Its team could not staff a conversation with every visitor, around the clock. But a generic chatbot was not an option either: the brand's voice had to be right, and with a young audience, the safety bar was as high as it gets. One bad reply could undo the whole launch.
What we built
We built five AI assistants, each with its own personality, each speaking in the brand's voice. They greeted visitors, answered questions, made suggestions and kept the conversation going, day and night.
- Each assistant followed a written guide to the brand: what to say, what never to say, and how to sound.
- Replies came fast. Most arrived in under four seconds, even at the busiest moments.
- It ran without a support team on shift. The brand's people only stepped in when they chose to.
How we kept it safe
Safety was designed in from the start, not added later. Every assistant was built on the assumption that it might be talking to someone under 16.
- Every message was checked before it reached a customer.
- Customers always knew they were talking to AI.
- Assistants forgot each conversation when it ended. No personal history was kept.
- One switch could turn everything off, instantly, if the brand ever needed it.
For the next phase, we wrote up a privacy design for the brand's legal team: no detail that identifies a person is ever sent to an outside AI service, and everything the AI makes is clearly labelled.
The results
The assistants handled 678,724 customer interactions with 247,091 people. The brand renewed and expanded the work into a second phase.