Daisy: the AI granny scambaiter

Read the campaign story

O2 introduced Daisy in November 2024 as an AI character that could converse with scam callers. The brand announcement credits advice from Jim Browning and places the activity within its Swerve the Scammers campaign. It also describes public-facing content featuring Amy Hart. The One Club records VCCP’s campaign work. Daisy was a demonstration and awareness device, not a service every customer could assume was screening their calls.

O2 explicitly told people to leave scambaiting to experts and report suspected fraudulent calls and texts through the appropriate reporting route. That instruction matters to the case: the entertaining content and the recommended audience behavior were different. Published examples of time spent on calls do not establish how many crimes were prevented or how much customer money was saved. No such inference is made here.

Editorial interpretation: a familiar character makes an invisible technical contest easy to follow. The usual power relationship is reversed, which gives the story a satisfying shape. But the educational purpose depends on what viewers do afterward. If an audience imitates the risky demonstration rather than following the safe advice, the communication has failed a more important test than entertainment.

The character also introduces a responsibility to avoid turning a group stereotype into the whole explanation. Age is not a complete account of susceptibility to fraud, and a comic persona should not replace practical information for varied audiences. An adaptation needs factual review, clear boundaries around the demonstration and a direct route to help. The spectacle earns attention; the next step determines whether the attention is useful.

Separate the hero’s action from the audience’s action

A campaign protagonist can do something the public should not copy. State the distinction at the moment it is most likely to matter, rather than hiding it in a footer. For a fraud-awareness story, the desired response might be pausing, verifying through an independent channel or reporting a suspicious message. Test recall of that safe step alongside recall of the character. If viewers remember only the stunt, redesign the ending and supporting content.

Make technical capability visible without overstating coverage

A demonstration can explain a system’s purpose, but it cannot prove that the system works in every context or is available to every customer. Identify the exact setting being shown and distinguish a pilot from an operational service. Have the technical and customer-support teams review the language together. This prevents a publicity claim from becoming an accidental service promise that the business cannot fulfill when customers ask for it.

Use protection outcomes that the evidence can support

Minutes spent on a call are an activity measure. They are not automatically crimes prevented, and publicity reach is not evidence that people changed their behavior. A useful evaluation can test recognition of suspicious patterns, recall of reporting steps and successful use of the help journey. Avoid collecting unnecessary details of personal fraud experiences for a marketing study. Let qualified fraud teams determine what prevention claims can responsibly be made.

A small test for another brief

Create a clearly labelled fictional scam-awareness scene with qualified reviewers, then test whether viewers choose the safe next step.

Measure recognition, safe-action recall and completion of the help journey. Keep these separate from video views and character popularity.

Stop if viewers interpret the exercise as an invitation to bait scammers or if the guidance no longer matches official advice.

• Use actors or an expressly simulated interaction rather than live scammers.
• Have fraud specialists review the scenario, reporting instructions and limits.
• Offer a short decision exercise and direct people to current official help, without collecting sensitive incident details.