O2's AI Granny: A genius AI scambaiter who outwits scammers
Telecom giant O2 created 'Daisy', an AI grandma that keeps phone scammers on the line with endless conversations, fighting phone fraud and educating consumers.
Imagine: an elderly lady chatting with phone scammers for hours, smartly stringing them along, while wasting valuable time that they would otherwise use to scam vulnerable people. Sounds like a lovely form of justice, doesn't it? This is exactly what British mobile operator O2 achieved with “Daisy” — a fully AI-powered grandma available 24/7 to frustrate and con fraudsters.
An ingenious weapon against phone fraud
The British telecom operator O2, in collaboration with creative agency VCCP, developed “Daisy” — an AI-powered scam baiter in the shape of an elderly woman. Daisy is an AI persona of a quick-witted “grandma” who answers fraudulent calls and deliberately lets scammers talk in circles so they waste their time and can't reach real customers.
Essentially, O2 built a real-life conversational AI that can talk on the phone like a chatty grandmother, deliberately keeping fraudsters busy with twisty conversations. The objectives were twofold: (1) directly occupy scammers' phone lines (each minute they talk to Daisy is time not spent defrauding an actual victim), and (2) educating the public about scammers' tactics through the collected material.
Watch Daisy's introduction here:
Amazing results
The O2 “AI Granny” proved to be extremely effective. During tests, Daisy kept some fraudsters on the line for more than 40 minutes — an eternity in the world of phone fraud, which probably drove the criminals crazy. Within months, Daisy processed over 1,000 fraudulent calls as “O2's Head of Scammer Relations.”
O2 publicly released a video of Daisy in action, along with insights she gathered: for example, confirmation that many scammers read from identical scripts and become aggressive when challenged. The campaign generated broad PR: media and social networks praised O2's creative use of AI for a good cause, and it stimulated conversations about fraud prevention among O2's customers (an important goal).
The main video featuring Daisy's “calls from hell” — starring a real influencer victim — garnered a lot of shares and helped viewers spot red flags (as Daisy deliberately provoked scammers to reveal their printing tactics).
Strategically, the effort positioned O2 as a champion for customers using cutting-edge technology. It also won industry awards for creative innovation in AI. Perhaps more importantly, Daisy got people talking: she “became a viral sensation” with more than 2 billion impressions of campaign content, according to WPP. The campaign drove traffic to O2's fraud advice site and likely increased brand trust. O2's marketing director said it shows “how far we want to go to keep customers safe... Our secret weapon is Daisy.”
Watch Daisy live in action here:
Advanced technical execution
How does Daisy work technically? The complexity behind this seemingly simple solution is fascinating:
AI speech generation technology
Daisy was developed with a mix of generative AI technologies. The team recorded the voice of a real grandmother (a family member of a VCCP employee) to create a natural text-to-speech voice that sounds authentically older. This ensured that the fraudsters heard a human voice, which was critical to credibility.
Custom language model
They trained a modified Large Language Model (LLM) on scambaiter scripts and common fraud phone calls, giving Daisy a knowledge base of questionable schemes and some humorous side tracks (like talking about her cat or knitting) to deploy.
The team refined the model to find the right balance between:
- Credibility (sounding like a real, slightly confused older woman)
- A waste of time (strategies to prolong the conversation)
- Natural dialogue patterns (including relevant but twisty answers)
Advanced call simulation
Daisy runs autonomously 24/7 and answers incoming fraud calls to a number that O2 deliberately distributed on scammers' lists. During conversations, Daisy responds in real time with relevant but cumbersome answers (thanks to the LLM's natural language-generating abilities), imitating how a sweet, confused grandma would react.
She was even coached by renowned human scambaiter Jim Browning to maximize time-wasting tactics, such as:
- Asking to repeat information
- Get a “supervisor” who will never arrive
- Pretend she's going to fetch non-existent bank cards
- Telling family stories
- Misunderstanding instructions
For the campaign videos, Daisy was visualized as an avatar — they used a diffusion model to generate a photorealistic digital grandma and an image-to-video tool to animate her for marketing content.
Autonomous operation
One of the most impressive aspects is that Daisy works completely autonomously. The system:
- Answers incoming calls
- Identifies fraudulent calls by signals in the call
- Applies different diversion tactics depending on the type of scam
- Keeps statistics on call length and type of fraud
- Extracts insights from conversations for analysis
This autonomy enables O2 to stop phone scammers 24/7 without employing staff.
What makes this case so special?
The O2 Daisy campaign is characterised by several innovative aspects:
- Technology for good: Instead of simply demonstrating AI as a technological trick, O2 used it for a real social problem — protecting vulnerable people from fraud.
- Character and personality: By giving AI a specific persona (a grandmother with name, voice, and personality), O2 made the technology relational and human, making the campaign more memorable.
- Immediate measurable impact: Every minute Daisy spent with a scammer was a concrete gain in the fight against fraud — a clear, tangible metric for success.
- Educational element: The campaign went beyond just frustrating fraudsters; it also educated consumers about fraud tactics by sharing real conversations.
- Strategic brand symbiotics: The campaign perfectly strengthened O2's brand position as a telecom provider committed to customer safety, bringing brand values and technological innovation together.
Together, these elements created a campaign that attracted attention, amused and had a real impact.
AI supercharging elements
The Daisy case embodies several core elements of AI supercharging:
- Scalable personification: A single AI system was able to have thousands of unique conversations, each adapted to the specific fraud scenario, without additional human effort.
- 24/7 capabilities: Unlike human anti-fraud teams, Daisy was able to work tirelessly, making continuous protection possible.
- Automated creativity: The language model was able to generate new, relevant response tactics that engaged fraudsters in real time, with minimal human supervision.
- Data-driven refinement: Each interaction provided insights that O2 could use to improve both Daisy and its wider fraud prevention strategies.
- Cross-functional value: What started as a marketing campaign also provided direct value for customer service and security teams — an example of how AI supercharging can break traditional departmental silos.
These supercharging elements made possible an initiative that would have been impossible with traditional means, regardless of the number of people or resources.
Challenges and Solutions
The implementation of the AI Granny came with several challenges:
- Naturalness of conversations: The AI had to sound convincingly like an older woman, including speech patterns and word choice. Solution: The voice was developed by recording a real older woman and the language model was specifically trained for age-appropriate language patterns.
- Balancing credibility with a waste of time: If Daisy was too smart, fraudsters would hang up quickly; too naive, and they could dominate the conversation. Solution: Fine-tune the model with input from professional scambaiters such as Jim Browning, who shared strategies for optimal waste of time.
- Ethical considerations: While focused on fraudsters, O2 had to consider the ethics of AI deception. Solution: The project was strictly limited to known fraud numbers and presented transparently in all public communications.
- Technical complexity: Combining speech recognition, LLM response, and realistic text-to-speech in a smooth phone conversation was technically challenging. Solution: Collaborate with specialized AI partners and extensive testing to ensure a seamless conversation experience.
- Measurability of impact: How do you prove that the effort actually reduces fraud? Solution: Metrics such as length of calls, frequency of calls back to numbers used, and extrapolation of the number of potential victims who were protected.
These challenges underscore that even with advanced AI, careful human control, and strategic thinking remain essential for successful implementation.
Impact and potential for your marketing team
The O2 Daisy campaign offers valuable insights and opportunities for various marketing teams:
- AI as a value keeper: The campaign shows how AI can be used to actively promote brand values (in this case customer protection) instead of just communicating.
- Content generation via AI interactions: The conversations themselves became valuable content — consider how AI interactions in your domain can provide interesting content.
- PR value of targeted AI innovation: The newsworthiness of creative AI applications can generate considerable media attention, especially if they address a societal problem.
- AI personification: Giving AI a name, personality, and story makes it more accessible and memorable for consumers.
- Data insights from AI conversations: The analyses of thousands of fraud conversations provided valuable insights that informed O2's broader anti-fraud strategies.
You don't have to have the size or budget of O2 to apply similar principles. Smaller brands can develop more targeted AI conversation tools for their specific market niche or customer problem.
Practical takeaways
- Identify sector-specific issues: What problem in your industry could be addressed with conversational AI? Think of common customer challenges or time-consuming interactions.
- Consider personification: An AI agent with personality can be more powerful and memorable than a generic chatbot. Consider which persona would resonate with your target audience.
- Find win-win applications: The best AI applications, such as Daisy, offer both immediate value (fraud prevention) and marketing value (PR, brand reinforcement).
- Start with existing tools: Start with available conversational AI platforms and personalize them for your specific use case before you start from scratch.
- Make AI conversations part of your content: The interactions themselves can become valuable content for marketing, training, or product development.
- Measure and share impact: Develop clear metrics to demonstrate the effectiveness of your AI solution and share these results as part of your marketing strategy.
- Maintain human control: Even autonomous AI like Daisy requires human supervision and guidance to stay effective and ethical.
Ready to outsmart fraudsters (or other challenges)?
O2's Daisy campaign shows how AI can go beyond just increasing efficiency or enhancing creativity — it can play an active role in protecting customers and promoting brand values. By developing an AI persona that not only talks but actually acts on behalf of your brand, you're creating a powerful new dimension in your brand experience.
The combination of practical utility (fraud prevention), narrative elements (Daisy's grandmother persona), and mediagenic results (40+ minutes of conversations with frustrated fraudsters) made this campaign a perfect storm of brand impact.
This article is part of our “AI Supercharged Campaigns” series, highlighting the most innovative and inspiring AI marketing cases that can inspire your team.