Daisy vs Scammers

Meet Daisy, a lifelike state-of-the-art Conversational AI designed to speak with scammers and keep them on the phone for as long as possible to waste their time and keep them away from the public. To make Daisy truly convincing, we knew we couldn’t just give her a voice. We had to build an entire digital persona from the ground up. This meant creating a rich backstory complete with friends, family, hobbies, and even pets. Every detail, no matter how small, contributes to Daisy’s authenticity, allowing her meandering stories to feel genuine, like those of your own beloved nan. Her “late husband, Norman,” for example, isn’t just a throwaway line; he’s a touchstone in her rich, fabricated memories.

Real human conversations are fluid, with overlaps, interruptions, and utterances. Daisy’s conversational model is trained to handle these gracefully. If a scammer interjects, Daisy doesn’t just cut them off or ignore them; she’s designed to acknowledge the interruption, perhaps with a “Oh, bless your heart, let me just finish this thought, dearie,” before seamlessly continuing her story or addressing the new input. This natural give-and-take is crucial for maintaining the illusion of a genuine person.

Minimising latency between Daisy and the scammer is essential to avoiding suspicion and we committed to ensuring that our average latency remained well below the 1000ms threshold where conversations start to feel robotic. Prompts were optimised for LLM processing, routing configured to minimise cross data-centre communication, and copious layers of caching and prefetching employed to allow the conversation to flow smoothly.

To ensure total believability, her voice was modelled on a real-life grandmother of a staff member from VCCP, giving her that warm, familiar, and utterly convincing tone. Beyond her direct impact in tying up scammers, Daisy has also been a powerful awareness-raising tool for the crucial 7726 scam reporting number. Her highly publicised existence, often featuring clips of her frustrating conversations with fraudsters, serves as a compelling and memorable example of exactly the kind of deceptive calls and messages people should be reporting. By showcasing how easily even a sophisticated AI can be targeted, Daisy underscores the pervasive nature of these scams and the importance of taking action.

Her charm, emotional nuance, and realism made her a surprisingly effective line of defence — one that left scammers frustrated, exposed, and out of time.

Meet Daisy, a lifelike state-of-the-art Conversational AI designed to speak with scammers and keep them on the phone for as long as possible to waste their time and keep them away from the public. To make Daisy truly convincing, we knew we couldn’t just give her a voice. We had to build an entire digital persona from the ground up. This meant creating a rich backstory complete with friends, family, hobbies, and even pets. Every detail, no matter how small, contributes to Daisy’s authenticity, allowing her meandering stories to feel genuine, like those of your own beloved nan. Her “late husband, Norman,” for example, isn’t just a throwaway line; he’s a touchstone in her rich, fabricated memories.

Real human conversations are fluid, with overlaps, interruptions, and utterances. Daisy’s conversational model is trained to handle these gracefully. If a scammer interjects, Daisy doesn’t just cut them off or ignore them; she’s designed to acknowledge the interruption, perhaps with a “Oh, bless your heart, let me just finish this thought, dearie,” before seamlessly continuing her story or addressing the new input. This natural give-and-take is crucial for maintaining the illusion of a genuine person.

Minimising latency between Daisy and the scammer is essential to avoiding suspicion and we committed to ensuring that our average latency remained well below the 1000ms threshold where conversations start to feel robotic. Prompts were optimised for LLM processing, routing configured to minimise cross data-centre communication, and copious layers of caching and prefetching employed to allow the conversation to flow smoothly.

To ensure total believability, her voice was modelled on a real-life grandmother of a staff member from VCCP, giving her that warm, familiar, and utterly convincing tone. Beyond her direct impact in tying up scammers, Daisy has also been a powerful awareness-raising tool for the crucial 7726 scam reporting number. Her highly publicised existence, often featuring clips of her frustrating conversations with fraudsters, serves as a compelling and memorable example of exactly the kind of deceptive calls and messages people should be reporting. By showcasing how easily even a sophisticated AI can be targeted, Daisy underscores the pervasive nature of these scams and the importance of taking action.