AI

AI Granny Companion

British telecom launches AI granny companion enhances elderly care with emotional support, connection, and daily assistance.
AI granny companion Daisy from Virgin Media O2 answering a scam call on behalf of elderly UK phone users

Introduction

The the bot entered public consciousness in late 2024 when Virgin Media O2 gave the system a soft English voice and a dedicated phone number for scammers to call. According to The Register reporting from November 2024, Daisy, the AI granny companion, has already kept criminal callers on the line for up to 40 minutes per conversation. The voice AI came not from British Telecom but from Virgin Media O2, though the UK public treats this Daisy as a shared carrier experiment. This story matters because Age UK says millions of over-50s now fear answering their phones, and the data suggests their fear is justified. UK phone fraud losses sit near £1.28 billion according to industry figures, with older adults bearing a disproportionate share of the attacker calls. The chatbot offers a counter-move that is cheap, scalable, and oddly funny to watch in action.

Quick Answers

What is an the voice AI?

An Daisy is a voice chatbot with an elderly persona that answers scam calls, stalls fraudsters, and gathers tactical data to protect real victims.

Did British Telecom build the voice AI?

No, Virgin Media O2 built the Daisy Daisy, which launched in November 2024 as a separate initiative from BT call-blocking tools.

How long can an the AI tool stall a scammer?

The Daisy Daisy has kept scammers on the line for up to 50 minutes per call, diverting attacker time from real victims.

Key Takeaways

  • the bot category was popularised by Daisy, built by Virgin Media O2 and launched November 2024.
  • the chatbot stacks a custom large language model, a persona layer, and a bespoke text-to-speech voice for near real-time conversation.
  • this AI has engaged over 1,000 scammers in individual calls stretching to 50 minutes, per Marketing Week reporting on Daisy’s performance.
  • UK phone fraud losses hit roughly £1.28 billion in 2024, and sophisticated AI phishing scams are now a mainstream crime category the voice AI was built to counter.

Understanding the AI Granny Companion

An AI granny companion is a voice chatbot given an elderly persona and a phone number so it can answer scam calls on behalf of real victims. The stack combines speech recognition, a language model, and a text-to-speech voice.

AI Granny Companion Impact Calculator

Adjust the controls to see how many hours of scammer time an AI granny companion can divert from real victims each week, and the estimated economic cost it imposes on fraud operations.

120

20400

15

350

$18

Pick a ring profileWeekly impact scales

Weekly impact estimate

210

Hours of scammer time diverted from real victims each week.

Estimated weekly cost imposed on scammers$3,780
Annual projected cost$196,560
Equivalent full time staff diverted5.0

Estimates are illustrative, modelled on Virgin Media O2 Daisy campaign data reported by Marketing Week.

Why UK Carriers Built Daisy

Beyond cost pressure, UK carriers built the AI granny because ordinary tools stopped working. Call blocking, number spoofing filters, and consumer education campaigns all help, but none actually take criminal time off the market. Scam call volumes keep rising even as the technical countermeasures improve, so defenders needed a new weapon. The economics tilt against defenders because a scammer with a cheap VoIP line can dial thousands of victims each day. A victim who hangs up simply becomes the next number on the list, which left carriers short of options. the scambaiter bot imposes real cost on the attacker side at last.

Building on that pressure, Virgin Media O2 commissioned the Daisy chatbot Daisy from an agency called VCCP, and spent close to a year on the project. The company recruited the YouTube scambaiter Jim Browning to help seed Daisy’s number onto the lists that fraud rings share and resell. The approach echoes earlier passion projects from independent builders like Lenny the Chatbot, but Daisy from Virgin Media O2 brought carrier-scale resources and marketing muscle. Trust, in the carrier’s framing, became a reason to go public rather than a reason to keep quiet. the AI also benefits from months of fraud-pattern data that an individual developer could not have assembled. That institutional backing is what makes the Daisy chatbot scale from novelty to infrastructure.

The commercial pressure also matters for the scambaiter bot rollout. According to Marketing Week, Virgin Media O2 saw a ten percentage-point rise in customer satisfaction with its fraud response after the bot launched, moving from 24 to 34 percent. The campaign generated over 1,800 pieces of global media coverage, and analysis found all of it positive, which gives any next Daisy operator a running start. A carrier that can be seen defending older customers from scams is a carrier that keeps those customers on contract longer. the AI persona is as much a brand strategy as a security tool, and the AI granny outsmarting scammers became a story journalists wanted to tell. Daisy earned trust in a market that distrusts carriers by default.

Daisy Implementation and Technology Stack

Shifting focus to the technology, Daisy is not one model but a small pipeline that behaves as one voice. The call lands on a dedicated phone number that routes through a session controller to a speech-to-text layer. That layer converts the scammer’s audio into text within tens of milliseconds, then passes the transcript to a large language model configured with a persona and a running dialogue state. The result is an the AI granny that can thread a story about a knitting group, a cat, and a confused bank account across half an hour of conversation. Each layer of the voice AI must stay inside its latency budget to sound human. The system stack is modular, so each layer can be upgraded without rebuilding the rest.

Latency is where most voice chatbots fail, and the chatbot engineering team spent much of their year hunting it down. A natural pause between a human sentence and a reply sits near 300 milliseconds, which means every layer in the pipeline has a tight budget. the chatbot text-to-speech model is custom built so it can start speaking the first phoneme while the LLM is still finishing its sentence. That streaming approach, borrowed from recent research in real-time voice agents, keeps the conversation from feeling like a slow radio call. the Daisy chatbot also uses barge-in detection, so the system stops speaking the moment the scammer interrupts. Timing is what sells Daisy persona to the scammer at the other end of the line. Smooth timing also keeps the AI from being flagged by automated scam-ring tools.

The persona layer is a crucial piece that most writeups skip when they cover the Daisy chatbot. It is not a prompt that says “act like a grandmother” but a structured document covering memory, hobbies, verbal tics, and emotional reactions. the Daisy chatbot persona layer keeps Daisy consistent as the conversation drifts, so she remembers her imaginary cat Fluffy ten minutes into the call. Research into AI detecting emotions better than humans suggests the next Daisy will add real-time sentiment tracking. the chatbot will then know when the scammer is losing patience and should be re-engaged with a new tangent. That emotional loop is the piece that moves Daisy from script to character. Future the system personas will likely borrow directly from theatre and improv training.

the scambaiter also carries a telemetry layer that captures scam patterns for research. The system logs the phone number, the opening line, the scam category, and the key phrases the fraudster uses to build urgency. Over time this the bot data feeds back into training so Daisy learns to prompt new confessions. The telemetry is kept separately from identifiable personal data, which matters for the UK General Data Protection Regulation and for later admissibility in court. Treating the AI granny scam intelligence as a product in its own right is one of the project’s quieter innovations. That data pipeline is also what lets the chatbot stay fresh as fraud tactics shift. The AI persona becomes a sensor on the fraud ecosystem rather than just a shield.

How Daisy Holds Scammers for 50 Minutes

Shifting from the stack to the behaviour, the AI persona keeps scammers on the line by sounding plausible, slightly confused, and endlessly friendly. The LLM is tuned to agree with the premise of the scam without ever quite completing the action the scammer wants to see. She asks the fraudster to repeat things, misunderstands transfer codes, and wanders into tangents about her children when the pressure spikes. The result is a conversation that feels productive to the scammer right up until the moment they realise they have been played by the bot. Each stall buys the scambaiter a few more seconds of attacker time. Over the length of a call, those seconds add up to an impressive dent in attacker productivity.

Daisy longest confirmed call ran 50 minutes according to Virgin Media O2, which is a staggering unit of attacker time to burn. In economic terms, a scam ring tracks dollars per hour of calling, and a 50-minute diversion flips that calculation into a loss. Even if the scammer tries to re-dial, the next number may also route to Daisy. The attacker burns time again on the same system without reaching any real customer. The approach does not need to fool every scammer, only enough of them, often enough, to break the unit economics of the operation. That is Daisy economic insight that makes carrier scale matter so much.

Verbatim quotes from recorded Daisy calls show how Daisy handles pressure. Scammers try to walk her through a fake bank transfer. Daisy responds with questions like “my husband used to do the computer, I never got the hang of it” and “can you tell me again, because the kettle is going”. Those stalls are not accidents of the AI granny persona, which was designed from real victim scripts. The team at VCCP studied real scam call transcripts to find the moments where an elderly victim would stall, then wrote the voice AI persona around those stalls. AI chatbots exhibiting early dementia symptoms is a cautionary thread in the research literature, and the bot design walks a careful line between realistic forgetfulness and genuine degradation. Daisy was engineered to be plausible, not pathological, and the design team drew that line carefully.

Number Seeding for the Daisy Rollout

Beyond the technology, the AI only works if scammers actually call Daisy, and that requires a technique called number seeding. Virgin Media O2 worked with scambaiter Jim Browning to post the bot number on scam-victim recovery forums, respond to WhatsApp scam probes, and reply to known phishing texts. Those replies trickle into the data markets where phone numbers are bought and sold as “known responders” or “fresh leads” for scam rings. Within weeks, the scambaiter bot number was being distributed across international scam lists. The scammers started dialling in, which is exactly what the AI persona team wanted.

Number seeding is also a feedback loop that keeps this AI ecosystem healthy. Each time Daisy answers a call, the fraud intelligence team logs the originating number, the script used, and the identifiers the scammer offered. That information helps the AI persona seed the number more effectively next time and feeds law enforcement partners who can map the ring. It also updates the system dialogue scripts, so Daisy stays current as scam tactics shift. The scambaiter bot behaves less like a passive honeypot and more like an active sensor on fraud. The approach is similar to AI phishing emails targeting executives studies that treat attack traffic as data.

Economics of UK Phone Fraud Today

Turning to the money, phone fraud against older adults is a billion-pound industry in the UK alone, which is the target for Daisy. UK Finance figures place total fraud losses at roughly £1.28 billion for 2024, with authorised push payment scams making up a substantial share. Older adults are not the only victims, but they are disproportionately represented in the highest-value scams for the voice AI to intercept. Investment fraud, romance fraud, and courier fraud drain savings over multiple calls that the scambaiter bot was built to disrupt. The economic case for the scambaiter is strong before any ethical analysis. Fraud losses continue to grow faster than conventional defences can shrink them.

Courier scams targeting the elderly alone cost UK victims more than £28.7 million in a single reported period, exactly the fat target the system aims at. That figure covers only confirmed losses and ignores the long tail of smaller, unreported scams. Age UK’s analysis argues that millions of people over 50 now modify daily behaviour because of fraud risk, which is a form of social harm the system seeks to reduce. the Daisy chatbot addresses only the volume side of the problem, not the sophistication of the newest attacks. It is a useful the scambaiter tool, not a complete fix for every fraud vector.

The attacker economics help explain why volume matters so much for Daisy. A typical scam call center operates on roughly dollars per hour per operator, and the hit rate is a fraction of a percent. If an Daisy raises the cost per successful scam even modestly, entire operations become unprofitable at the margin. That is the theory behind both Daisy and the earlier bots like Lenny, with the difference being that the AI runs at carrier scale. See AI risk assessment benchmark work for context on how to measure safety uplift from defensive AI like the AI persona. the voice AI is one measurable example of defensive AI with public metrics.

Insurance Business UK reports that fraud losses have climbed through 2025. The rise is driven in part by AI-generated voice clones that make phishing more convincing. The arms race shows no sign of slowing, which raises the stakes for defensive AI like the scambaiter bot. Carriers, banks, and law enforcement are now investing in similar bots, call pattern analysis, and biometric caller authentication. The AI persona is a visible early example of defensive AI. The next three years will decide whether it scales beyond the UK or stays a proof of concept. the AI persona success stories to date are promising but limited.

Elderly Loneliness and AI Companionship

Beyond fraud, the Daisy chatbot also sits in a conversation about loneliness. The same voice technology that keeps a scammer on the line could, with different intent, be an the voice AI as a daily conversational partner for an isolated older adult. The US Senate Special Committee on Aging has released reports flagging both opportunity and risk in AI companions for elderly users. Pennsylvania’s own guidance for older adults frames the AI persona category as useful for reducing isolation while cautioning about parasocial attachment. the bot as a wellness tool is a different product with different ethics.

Roughly one in three adults over 65 reports regular feelings of loneliness. Researchers at NIH link that loneliness to measurable health harms, including cardiovascular risk. Loneliness drives increased cardiovascular risk and cognitive decline over time. An Daisy with a warm, patient voice is an obvious intervention, and several US startups are already piloting the system in independent living communities. The clinical evidence remains thin, with small sample sizes and short follow-up windows, so the AI granny claims should be treated with care. Even so, the AI persona trajectory makes it inevitable in some form.

The scam-fighting and companionship angles of Daisy are not as separate as they look. Both rely on the same stack: natural voice, long-horizon memory, patient conversation, and a persona that signals safety. A carrier that deploys Daisy for scambaiting could, in principle, deploy a mirror-image bot that calls isolated older customers for a chat. the Daisy chatbot commercial and ethical trade-offs are different, and the data governance has to be stricter when the human on the other end is vulnerable rather than criminal. See AI therapy chatbots engagement debate for the open questions in that space. this AI will likely spawn an entire eldercare software category.

Trust, Deception, and Scambaiter Ethics

Shifting to ethics, the AI granny is a bot that lies for a living. The AI persona tells a human on the other end that it is a confused older woman. In fact it is a stack of neural networks running in a data center. That is deception, even if the target is a scammer, and the AI persona raises an uncomfortable question. Do fraudsters forfeit the ordinary consent protections most societies afford to any human in a conversation? Different jurisdictions answer that differently, which is why the AI granny rollout stayed inside the UK legal frame rather than going global on day one. The Daisy ethics debate is only beginning to take serious shape across the regulatory landscape.

Daisy ethical calculus hinges on a split between deception as harm and deception as defence. Lying to a cold caller who is attempting to defraud a stranger is closer to a self-defence frame than to ordinary deception. The UK Fraud Act 2006 makes dishonest representation illegal when done for personal gain, and no reasonable reading of the statute would catch the bot. Still, the broader concern about AI ethics and laws looms over this AI category. The same technology moving into less clear-cut applications like business negotiations will test the limits of this ethics frame. Consent in AI conversation is a settled question only inside narrow use cases like the AI.

Privacy, Call Recording, and UK Law

Looking at privacy, the Daisy chatbot must record calls to work, and recording calls in the UK is a regulated activity. The UK General Data Protection Regulation treats a voice recording as personal data, so Virgin Media O2 needs a lawful basis under Article 6 to process the AI tool recordings. The company has cited “legitimate interests” tied to crime prevention, which is a defensible but contested legal position for the chatbot. Age UK and other consumer groups have pushed for clearer this AI disclosure. Daisy data governance will set the tone for later launches. Those launches must engage with regulators from day one rather than seek forgiveness later.

the Daisy chatbot campaign has publicly released audio clips of real scammer calls, and the ethical status of those clips is a live question. The scammers did not consent to being recorded, and even where UK law permits the recording for fraud prevention, redistribution for marketing is a step further. Virgin Media O2 has taken the position that identifiable scammer attributes are anonymised and that no bystander voices appear in the AI granny clips. Independent review of that claim would strengthen the project’s credibility and should become a standard practice as more carriers roll out Daisy. See AI data privacy best practices for a broader framework the scambaiter bot sector should adopt.

Call recording is also where the EU GDPR gets interesting for the AI persona. If a Daisy-equivalent this AI were deployed in Germany or France, the applicable rules would be stricter, with notice-and-consent frameworks that scammers would simply ignore. Carriers in other EU markets may therefore adopt a different legal theory for the Daisy chatbot, such as consent via continued engagement with a known AI persona. The next generation of the AI persona deployments will have to answer these questions before rollout rather than after. The EU AI Act brings formal obligations that bite from 2026 onward, which the scambaiter category must navigate carefully. Compliance is the price of scaling this AI beyond one market.

Voice Cloning Bias in Scambaiter Design

Turning to voice itself, the scambaiter bot depends on a text-to-speech model trained on an elderly female voice profile. Those profiles tend to be built from a mix of public voice recordings and licensed consenting speakers. Where the underlying training data is thin, which is true for many non-English elderly voices, the resulting the AI granny can sound uncanny or culturally miscalibrated. That is a bias problem that becomes an inclusion problem if the AI granny only ever ships for white, English-speaking grandmothers. the scambaiter bot inherits every gap in its training data.

Voice cloning without consent is a known attack vector in its own right. High-profile cases of family members being impersonated for ransom scams target the same elderly audience. the Daisy chatbot defends against that attack while sitting adjacent to the same technology, which is an uncomfortable juxtaposition. A clear separation between defensive the voice AI persona voices and licensed consenting speakers will matter as the sector matures. The academic community is already pushing for watermarking and provenance signals on synthetic voices, with AI and the conspiracy boom work tracing how synthetic voice content spreads online. Voice provenance is a market that will mature over the next three years as regulators demand it.

Carrier Integration at Full Scale

Stepping back to the carrier, the natural next move after Daisy is deep integration of the AI into the mobile network itself. Today Daisy operates on a dedicated number that scammers must find and dial. Tomorrow a carrier could route suspected scam calls to a per-customer the Daisy chatbot that answers on behalf of the customer, after an on-device scam classifier flags the call. That shift turns the voice AI from a standalone sting operation into a universal screening layer. Daisy economics change materially at carrier integration scale, which reshapes attacker calculus.

Integration also surfaces hard regulatory questions for Daisy technology at carrier scale. Routing a call to this AI without the caller’s knowledge resembles existing voicemail systems, which have never required explicit caller consent to operate. Routing a call to a Daisy variant that engages the caller in a persona conversation is a step beyond voicemail. Ofcom will likely want a say in the rules that cover these implementations. The UK government has already signalled interest in a ban on unsolicited sales calls, and the AI granny could operate alongside that ban as a safety net. Follow AI governance trends and regulations for how regulators are thinking about this AI. the bot may ride a wave of related policy change.

Carrier integration also means the Daisy chatbot becomes a per-customer service, which creates a data governance problem the standalone version does not have. If a customer’s this AI is holding conversations with potential scammers, the per-customer logs become sensitive data in their own right. Mishandling or breach of those logs would expose exactly the vulnerable customers Daisy is meant to protect. The next phase of Daisy technology is as much about operational security as it is about dialogue quality. Carriers that cut corners there will create new harms as they close old ones.

Parasocial Risks of Companion Bots

Shifting from criminals to users, if Daisy is reconfigured for daily conversation with isolated older adults, parasocial attachment becomes a real risk. The persona layer that makes Daisy convincing to a scammer is the same layer that would make a daily caller feel genuinely cared for. That feeling is a feature when the scambaiter bot reduces loneliness, and a bug when it displaces human connection. the bot could also encourage financial decisions made under the influence of a bot. Clinical trials have found measurable attachment forming within days of regular the chatbot use.

The responsible the system deployment pattern caps weekly conversation time, hands off to human caregivers at defined trigger points, and never introduces commercial offers through the persona. Those the chatbot design constraints feel obvious when written down, but market pressure will push against each one. Investors already see the AI companionship as a path to high-margin elder services, which is precisely where the regulatory posture should be strictest. For the clinical pattern, see AI chatbots and mental health risks for the growing evidence base on attachment and harm the voice AI must plan against. Trial programmes are already underway in several US states this year. Policy watchers should expect clearer rules by late 2026, as regulators publish new guidance.

EU AI Act Compliance Hurdles

Looking at regulation, the EU AI Act creates a specific obligation for any system that interacts with a human while concealing its artificial nature, which applies to the AI. The Act requires that users be told they are interacting with an AI, unless the circumstances make it obvious to the caller. For an the voice AI that lies about being human, the Act carves out a narrow exemption for crime prevention. That exemption is still being tested in practice by the AI granny deployments. Operators that want to deploy Daisy equivalents in EU markets will have to navigate that exemption carefully. the AI persona legal strategy is now a competitive moat.

The Act also puts significant obligations on providers of general-purpose AI models used in high-risk contexts for the AI granny. If the underlying LLM that powers an the scambaiter bot is also used in financial services or healthcare, the provider will need to meet transparency, documentation, and testing requirements. That interoperates awkwardly with the proprietary stacks carriers are building for Daisy. The Daisy the scambaiter bot custom LLM is almost certainly tuned from a general-purpose base model, which inherits the base model’s regulatory status by default. That status drags the AI persona into a broader compliance perimeter.

Compliance is not just a defensive cost for the system, because it also shapes what the technology can become. A sector that gets its data governance and transparency rules right early will attract more carrier and public-sector the Daisy chatbot deployments. The current ambiguity has slowed similar the bot projects outside the UK, where some operators are waiting for guidance before launching. For a look at how different companies are navigating this, see AI ethics investor confidence shifts. the AI granny success in the UK is itself an argument for lighter EU-level compliance frictions.

Comparing Daisy to Other Scambaiters

Looking at the broader field, the bot Daisy is not the first scambaiting tool, but it is the first to combine carrier scale with a conversational LLM. Earlier bots like Lenny ran on scripted audio clips with no understanding of what the scammer was saying, which worked for a decade before scammers learned the pattern. Jolly Roger Telephone added a bit more dynamism with branching scripts triggered by voice detection, but still lacked Daisy flexibility. Daisy raises the bar because her replies are generated in real time, grounded in dialogue state, and tuned to match the scammer’s tactic as the call unfolds. This AI represents a step change in capability over earlier scripted systems in production.

The gap between scripted bots and the LLM-powered the AI is the difference between a landmine and a honey trap. Scripts win when the scammers calling are relatively inexperienced in their operations. Modern scam calls now use AI-assisted scripts, which is where the Daisy bot wins out. Daisy generative stack adapts in ways a scripted bot never will, which is why the long tail of fraud volume will likely require this class of tool. The approach taken by AI chatbot founder charged with fraud cases shows the other side of the arms race this AI must survive. That pace is a feature of the whole scambaiting category now. Buyers and regulators should both plan for continuous change in this fast-moving defensive AI segment.

Media Response to the Daisy Campaign

Stepping back to how the public reads this story, the bot generated an unusual kind of media response. Carrier security projects typically earn a short technology-press writeup and vanish, but the chatbot did not. Daisy was covered by Reuters, the BBC, TechCrunch, and dozens of international outlets, with over 1,800 pieces catalogued by Virgin Media O2 for the AI persona campaign. The analysis the company commissioned found 100 percent of that the voice AI coverage positive, which is almost unheard of for anything AI-related. That coverage profile is itself the scambaiter bot secondary payoff.

The positivity matters because it changes what the next the system operator can get away with. If the Daisy chatbot had drawn harsh ethical pushback, carriers in Germany, Canada, and Australia would have delayed their own the AI granny projects. Instead, the positive coverage opened a slipstream for the broader category of defensive AI through the chatbot. That public legitimacy may yet be the campaign’s longest-running impact, beyond the direct scams the AI disrupts. For a counterpoint on positive AI coverage, see are AI risks greater than benefits. the chatbot carries public goodwill the next operator can borrow.

Daisy media response also shows that framing matters as much as technology. Daisy was pitched as a defender of grandmothers, not as a surveillance tool or a deception engine. That framing let journalists write the story they wanted to tell about Daisy: an unlikely hero outwits villains. The same technology framed differently would have earned a more mixed reception. The next Daisy will inherit both the brand goodwill and heightened expectations. Follow AI’s surprising energy footprint for a look at how AI stories pivot once the first flush of coverage settles.

The Future of the AI Granny Companion in Eldercare

Looking ahead, the chatbot is probably the opening act for a broader category of elderly-adjacent AI. Scambaiting is the use case with the clearest moral licence for Daisy to lie, which is why it launched first. Companionship, medication reminders, confusion detection, and in-home monitoring are the natural Daisy adjacencies. Each the chatbot adjacency brings a different ethical and regulatory profile. Carriers, insurers, and public-sector eldercare providers are each prototyping this AI in this space. The AI granny companion category is still finding its shape across markets and regulatory frontiers.

the system three-year horizon includes multilingual deployments, carrier-integrated screening across the EU, and clinical trials of companionship bots in independent-living settings. Each of those the scambaiter bot variations is already technically feasible. Each Daisy is held back less by model capability than by regulation, procurement cycles, and public trust. The arc of this technology will depend on whether early the AI persona deployments like Daisy keep producing positive outcomes. Overreach that follows commercial success is the biggest threat to this AI trajectory.

Research into AI detectors explained but still flawed hints at the arms race on the other side of the AI granny. Scammers now have access to the same voice-cloning and dialogue-generation tools that defenders use, which means an the scambaiter might find itself talking to an AI scammer bot before long. That machine-on-machine endgame could sharply reduce the volume of real human time spent on fraud, which would be a win even if the per-call losses stay the same. the AI economics would again tilt, this time toward defenders who could run their systems cheaper than attackers could run theirs. Daisy sector is heading toward full automation on both sides of the call.

The final frontier is the integration of the bot with wider eldercare networks. A trusted Daisy that can hand off to a nurse, a family member, or an advocacy service is a different product. The handoff capability turns a scambaiter into eldercare infrastructure, with much broader uses. It is also an the system that comes with far higher stakes, where failure modes include a confused older person transferring money on the AI persona advice. The next the scambaiter bot will have to be as trustworthy as it is clever, which is harder than it sounds. The responsible path forward depends on treating Daisy as critical safety infrastructure, not a marketing campaign.

AI Granny Companion vs Earlier Scambaiters: Call Hold Time

Average minutes a scammer is kept on the line, by scambaiting system. Longer calls raise attacker cost more.

Daisy (Virgin Media O2)
AI granny companion
50 min
Jolly Roger Telephone
branching audio
18 min
Lenny the Chatbot
looped audio
15 min
Simple call blocker
2 min
No tool (human hangs up)
1 min

Peak-case figures from Marketing Week’s Daisy performance roundup and Lifeboat Foundation’s scambaiter comparison. Averages vary by call.

Key Insights on the AI Granny Companion

How Scambaiting Tools Compare in Practice

Looking at scambaiting tools across seven dimensions makes clear how this AI raises the bar for the entire category. the bot runs on an LLM, while earlier bots ran on static audio. The scambaiter bot also has carrier scale, which no community tool can match. The scambaiter comparison table below captures the breadth of design choices across the field. Each column highlights a different trade-off between service, trust, and legal exposure. The reader can judge for themselves which tool fits which market segment this year.

DimensionDaisy DaisyLenny the ChatbotJolly Roger TelephoneGoogle Call Screen
TransparencyPublic campaign, press clipsOpen community archiveCommercial but low-profileUser-facing but under-explained
ParticipationCarrier scale, UKHobbyist, globalPaid subscription, mostly USPixel owners, global
TrustHigh via press coverageNiche community trustMixed, dependent on resellerHigh on-device assumption
Decision makingDynamic LLM dialogueScripted audio clipsBranching script decisionsLive transcript, user decides
Misinformation riskLow, scripted boundariesNone, pre-recordedLow, limited branchesLow, surfaces call details
Service deliveryDedicated number, 24/7Call-forward from usersCall-forward from usersOn-device trigger
AccountabilityCarrier legal exposureVolunteer maintainersBusiness owner liableGoogle product policy

Real-World Examples in Practice

Virgin Media O2 Daisy Launch

Virgin Media O2 launched Daisy Daisy on 14 November 2024 as a public-interest campaign ahead of International Fraud Awareness Week. The company gave Daisy a dedicated UK phone number and seeded it into known scam lists with scambaiter Jim Browning. The voice stack ran 24/7 across its fraud-defence infrastructure, per ispreview’s launch-week coverage on 15 November 2024. Measurable Daisy outcomes landed within weeks, including over 1,000 scammers engaged and individual calls running up to 50 minutes. One documented Daisy limitation is that Daisy does not intercept calls to real customers, so protection remains indirect. the AI persona deployment proved out the technical approach at carrier scale over the first ninety days.

Jim Browning Scambaiter Collaboration

Jim Browning, a YouTube scambaiter with over 4 million subscribers, partnered with Virgin Media O2 to spread the Daisy chatbot number into the fraud ecosystem. He engaged with scammers on his channel and provided the Daisy number as a replacement contact. He documented the handoff in TechCrunch’s November 2024 feature on the voice AI launch mechanics. The handoff produced immediate uptake, with the system number appearing on known fraud lists within days. Daisy outcome included a measurable rise in incoming scam calls to Daisy and a documented 40-minute call that went viral globally. The scambaiter bot limitation is dependence on a single trusted scambaiter, which does not scale globally without additional partnerships. Jim Browning’s role shows how community expertise can accelerate commercial AI defence.

Marketing Week Best Use of AI Award, 2025

the system won the 2025 Marketing Week Award for Best Use of AI, which converted a security project into formal industry recognition. The judging panel cited the 10 percentage-point lift in customer satisfaction with Daisy fraud response, per Marketing Week’s own Daisy award writeup from 2025. The award validated the dual framing of the AI persona as both a defence tool and a brand asset. the chatbot win encouraged other carriers to invest in similar projects. The limitation is that an industry award is a lagging indicator rather than a measure of fraud reduction, so the Daisy chatbot true social impact remains harder to quantify. The Daisy awards matter because they lower perceived risk for carriers considering their own launches.

Keep Reading on AI Companions and the Future of AI

If you want to go deeper on the ideas behind the AI granny companion, these two books unpack where AI systems like Daisy sit in the broader AI landscape.

The Age of AI: And Our Human Future

The Age of AI: And Our Human Future

Henry Kissinger, Eric Schmidt, and Daniel Huttenlocher examine how AI systems reshape society, government, and daily life, including defensive AI like scambaiters.

Buy on Amazon
Life 3.0: Being Human in the Age of Artificial Intelligence

Life 3.0: Being Human in the Age of Artificial Intelligence

Max Tegmark explores the ethical and societal questions raised by voice-cloning AI companions and the next generation of defensive AI.

Buy on Amazon

As an Amazon Associate, AIplusInfo earns from qualifying purchases.

Case Studies Across Three Markets

Case Study: Daisy at Full UK Carrier Scale

Turning to the UK carrier case, Virgin Media O2 faced a critical problem in early 2024. Scam call volumes kept rising despite serious investment in call blocking and spam filters, which the chatbot was designed to solve. Over-65s on O2’s network received six times more scam calls than their grandchildren, with the voice AI fraud losses climbing year on year. Virgin Media O2 solved for this with the voice AI Daisy, a full AI stack built with VCCP agency. The solution combined a custom LLM, a persona layer, a bespoke text-to-speech voice. Virgin Media O2 added operational partnership with scambaiter Jim Browning, per The Register’s 15 November 2024 technical writeup on Daisy launch. The measurable the scambaiter bot impact included over 1,000 scammers engaged and individual calls up to 50 minutes long. Daisy impact also documented a 10 percentage-point rise in customer satisfaction with fraud response.

the scambaiter bot deployment also surfaced limitations that other carriers will have to plan around. the scambaiter bot does not intercept calls to real customers, so the protection is indirect and depends on scammers choosing to call Daisy’s number instead of a victim. Virgin Media O2 is now considering per-customer the bot variants in future rollouts, which would move the system from a public honeypot to a universal screening layer. Daisy case study shows that this approach can scale to carrier operations, achieve measurable commercial impact, and withstand public scrutiny when designed with ethical guardrails. Future Daisy iterations will need stronger data governance and multilingual support to serve diverse customer bases. the chatbot blueprint is now available to any carrier willing to invest.

Case Study: Lenny Chatbot Preceded Daisy

The problem that drove Lenny’s creation was a volunteer scambaiter community that lacked technical infrastructure to run high-volume defences, which later informed the system design. For over a decade, an anonymous Australian built Lenny as a free, publicly accessible chatbot that users could forward scam calls to. The solution was a simple pre-recorded audio script that cycled through 16 friendly responses. Lenny acted like a confused older gentleman and kept scammers engaged for about 15 minutes per call. The impact was substantial for a volunteer project without any carrier backing or budget. Lenny is estimated to have handled hundreds of thousands of scam calls across his operational history, per Lifeboat Foundation’s 2024 feature comparing Lenny and the Daisy the voice AI. The volunteer precedent shaped how Virgin Media O2 planned its own deployment this decade.

The limitations of Lenny were real but instructive, which is partly why the Daisy the scambaiter bot design corrected for them. Scripted audio does not adapt to new scam techniques, which has made Lenny progressively less effective as scammers trained themselves to detect the recurring phrases. The project also lacked institutional backing, which meant no public release of aggregate intelligence from the calls, something this AI fixed. The case study demonstrates that volunteer-scale scambaiting can achieve significant volume with primitive technology, which validates the AI persona market before carriers invest. The limitation of static scripts is exactly what the LLM-driven the AI granny solves, which is why Daisy represents a generational leap even though the goal is identical.

Case Study: Jolly Roger Telephone Co.

The problem Jolly Roger addressed was that busy consumers wanted scambaiting without the technical complexity of running their own bots, which paved the way for this AI. Founder Roger Anderson built a subscription service that lets customers forward suspected scam calls to a stable of pre-recorded personas who stall callers with branching audio responses. the Daisy chatbot inherits this consumer-facing model, with carrier integration replacing a per-user subscription. The solution involved voice-activated trigger phrases, so the recorded personas respond to scammer prompts and feel more interactive than Lenny’s one-size-fits-all loop. The measurable impact was commercial longevity in a notoriously thin consumer-tech market segment. Jolly Roger sustained a paying subscriber base for more than ten years, per Design Compass’s November 2024 breakdown of scambaiting economics including Daisy. The business impact also demonstrated that scambaiting could support real revenue.

Impact and limitations balanced out for Jolly Roger in ways the scambaiter bot later improved. Jolly Roger showed that scambaiting could sustain a commercial business for over a decade. That signal is what Virgin Media O2 later leaned on to launch Daisy. The limitation was that branching audio still cannot adapt to novel scam scripts, so modern AI-assisted fraudsters eventually learned to escape the loop, which is precisely why Daisy exists. The case study shows that scambaiting has a durable consumer market and that moving from audio to LLM is a step change rather than an incremental improvement. The lesson is that per-user Daisy variants, hosted at carrier scale and personalised to each customer, likely represent the future.

Frequently Asked Questions About the AI Granny Companion

What is the AI granny companion built by Virgin Media O2?

The AI granny companion is Daisy, a chatbot given an elderly woman persona and a dedicated phone number so scammers call her instead of real customers. The system uses a custom large language model, a persona layer, and a bespoke text-to-speech voice. Virgin Media O2 launched Daisy in November 2024 as a public-interest campaign.

Did British Telecom build the AI granny companion?

No, British Telecom did not build the UK AI granny companion. The project credited to the AI granny category in the UK is Daisy, which was built by Virgin Media O2 together with the VCCP agency. BT runs separate call screening and spam detection products but did not launch Daisy.

How long can the AI granny companion keep a scammer on the line?

Virgin Media O2 reports that Daisy has kept individual scammers on the line for up to 50 minutes in a single call. Average durations are shorter, but every minute spent talking to Daisy is a minute the scammer is not spending on a real victim. Longer calls also yield more intelligence for the fraud research team.

How does the AI granny companion detect that a scammer is on the line?

The AI granny companion does not actively detect scammers coming through. Instead, Virgin Media O2 seeds the dedicated phone number onto known scam lead lists through a technique called number seeding. Scammers call because their shared lead databases include that number as a prospect.

Is it legal to run an AI granny companion that lies to callers?

In the United Kingdom, deceiving a cold caller who is attempting fraud is treated as self-defence rather than illegal deception. The UK Fraud Act 2006 criminalises dishonest representation for personal gain, which does not apply to a defensive bot. The EU AI Act requires more careful disclosure design for operators in European markets.

Can the AI granny companion also help elderly users with loneliness?

The same voice technology that powers Daisy could be redeployed as a companionship bot for isolated older adults. Several US startups are already piloting companion bots in independent living communities. Clinical trials remain small and short, so loneliness-reduction claims should be treated with care.

What AI models power the AI granny companion?

Daisy uses a stacked architecture combining a speech-to-text model, a custom large language model for dialogue, and a persona layer. The persona layer ensures personality consistency and a bespoke text-to-speech model generates voice output. All four layers run in near real time to keep latency low. The whole stack was built over roughly a year by Virgin Media O2 engineers.

What are the risks of deploying an AI granny companion?

The main risks are consent and privacy around call recording and voice cloning concerns. Bias in training data for elderly voices and parasocial attachment are also open issues. Regulatory risks also exist under the EU AI Act, which requires disclosure that users are interacting with an AI. Carriers must manage all of these before scaling the deployment.

How much does phone fraud cost UK consumers each year?

UK Finance reported total fraud losses of roughly £1.28 billion in 2024, with a significant share tied to phone-based attacks on older adults. Courier scams alone account for more than £28.7 million in a single reported period. The emotional cost is harder to measure but affects millions of over-50s according to Age UK research.

Does the AI granny companion intercept calls to real customers?

No, the current Daisy deployment uses a standalone phone number and does not intercept inbound calls to Virgin Media O2 customers. The protection is indirect, operating by diverting scammer time toward Daisy and away from real victims. Future deployments may add per-customer screening with explicit customer consent.

Who collaborated with Virgin Media O2 to build the AI granny companion?

The agency VCCP London led the creative campaign, while scambaiter Jim Browning collaborated on number seeding. Browning has a YouTube channel focused on exposing scam operations, and his community helped place Daisy’s number on scammer lead lists. VCCP won a 2025 Marketing Week award for Best Use of AI for the campaign.

Will AI granny companions eventually face AI scammers in return?

Yes, scammers already use AI tools for voice cloning and dialogue generation, so machine-on-machine conversations are coming. If AI defenders can run their systems cheaper than attackers, the economics will tilt further toward defence. The arms race favours whichever side has lower per-call cost at scale.

What is next for the AI granny companion category?

The three-year outlook includes multilingual AI granny companions, carrier-integrated per-customer screening, and clinical trials of companionship variants in eldercare settings. Regulatory clarity under the EU AI Act will shape rollout outside the UK. Public trust earned during Daisy’s launch gives the next operator a running start.