AI use case

AI spam and scam call blocking for mobile and landline subscribers

AI in the operator's network that protects subscribers from unwanted calls: it analyses incoming calls in real time, blocks known fraudulent calls, and labels suspected scam, spam and spoofed calls on the customer's screen before they answer, so subscribers can decide whether to pick up. Fraud against the operator itself, such as SIM swap or revenue share fraud, is a separate use case.

By Len Debets · Last verified 26 September 2026 · 5 public deployments

At least 540 million
Interactions handled
Bell Canada (organization claim).
USD 500,000 to USD 7.5 million
Indicative value per year
A mobile operator with 5 million subscribers. Worked example, see how it is calculated.

What problem does it solve?

Phone scams are one of the main ways criminals reach victims. Callers pose as a bank, a tax authority, an online retailer or the operator itself, often with a spoofed number that looks local or familiar, and push people to hand over details or move money. In the UK, Virgin Media O2 and Hiya found fake Amazon, HMRC and banking calls at the top of the list of nuisance calls in early 2026, and in Australia Telstra cites the ACCC's finding that phone scams accounted for the highest overall financial losses among all contact methods in Australia in 2024. The side effect is that people stop answering unknown numbers: in Telstra's research, 42% of Australians with a mobile device say they are less likely to answer calls because of scam fears, which also hurts the legitimate organizations that need to reach them.

Scammers constantly adapt their numbers and tactics, so static block lists fall behind. Operators also have to avoid blocking genuine calls, which is why some malicious calls still slip through. The network sees signals no single phone can see: how many calls a number makes and whether a call claiming a local number actually arrives from abroad. Using those signals at scale, in real time, is where machine learning helps; BT says its vendor Hiya uses machine learning to improve scam detection the more malicious calls it encounters.

How does it work?

  1. Analyse every unknown call. When a call arrives from an unknown number, the model scores it in real time on the behaviour of the calling number (for example a high volume of calls from a single number), where the call really comes from, customer reports and other data points.
  2. Detect spoofing. Models and network checks flag calls whose presented number does not match where the call really comes from, such as an overseas call showing a local mobile number.
  3. Block or label. Known fraudulent calls are blocked or diverted to voicemail; suspected scam or spam calls are delivered with a warning label on the handset or landline display; verified businesses can show their name.
  4. Learn from reports. Customer reports (for example to the 7726 short code in the UK) and investigations are used to block the numbers behind them and to refine the blocking services, so new scam trends are identified and blocked faster.
  5. Trace and shut down. Operators work with other carriers and regulators to trace scam calls back to their origin and stop the parties bringing them into the network.
Audience
Customer facing
Autonomy
Autonomous
Adoption
Mainstream
Channels
Phone and voice, Mobile app, API and system to system

What is it worth?

Benchmarks are computed from the public deployments below: one data point per organization per KPI, with who made each claim.

Value benchmarks for AI spam and scam call blocking for mobile and landline subscribers
KPIMedianReported rangeData pointsClaimed by
Interactions handledNot pooled
2.4 million to 540 million
33 organization
Users servedNot pooled
2.5 million
11 organization

Value drivers: Risk and loss reduction, Customer experience, Inclusion and access.

Indicative value

A mobile operator with 5 million subscribers

USD 500,000 to USD 7.5 million

Customer scam losses prevented per year

How this is calculated

Formula: subscribers * victimRate * averageLoss * preventedShare. The low scenario uses every low input, the high scenario every high input.

InputLowHighBasis
Subscribers protected subscribers, subscribers5,000,0005,000,000The reference operator.
Share of subscribers who lose money to a phone scam in a year victimRate, fraction of subscribers0.0020.005Editorial assumption, deliberately far below the survey figure cited on this page (Hiya reports 16% of UK consumers fell victim to phone scams), because survey victimisation includes small and unreported losses.
Average loss per victim averageLoss, USD per victim5001,000Editorial assumption; Hiya's survey cited on this page reports an average loss of GBP 798 per UK victim.
Share of those losses prevented by blocking and warnings preventedShare, fraction of losses0.10.3Editorial assumption. O2 reports that calls labelled suspected scam are answered 42% less often; not every unanswered scam call is a prevented loss.

What it leaves out: Customer losses only, and a rough order of magnitude. It leaves out the operator's savings on scam related complaints and contacts, the value of customers trusting calls again, and the cost of genuine calls that are wrongly labelled or blocked.

Who already uses it?

5 public deployments, strongest evidence first. Grades: A regulator or audit, B the organization itself, C vendor case study, D anonymous or estimate.

Bell Canada

Canada · Telecommunications · 2025

ScaledGrade B

Bell launched AI powered Suspicious Call Detection in 2025 to block or label spam and fraudulent calls for its wireless customers. In July 2026 it added a new AI model that identifies and flags spoofed calls, which manipulate caller ID to impersonate trusted organizations or contacts, in real time. The enhancement rolls out automatically, with no opt in required, to customers on iOS and Android phones across Bell, Virgin Plus, Lucky Mobile, PC Mobile, No Name and Maxi. Bell says it was the first carrier in Canada to flag spoofed calls this way.

  • Interactions handled: at least 540 million, 2025 launch to July 2026, calls blocked or labelled
    "Since launching Suspicious Call Detection in 2025, Bell has analyzed more than 4.4 billion calls and blocked or labelled over 540 million suspicious or fraudulent calls."
    Claimed by: organization

BT Group

United Kingdom · Telecommunications · 2024

ScaledGrade B

BT's Digital Voice home phone service includes Enhanced Call Protect, an AI powered tool from Hiya that monitors incoming calls, diverts scam calls to a junk voicemail and shows a "Nuisance?" warning on the landline display for suspected spam, while showing the name of registered businesses. In its first four months it blocked more than 2.4 million scam calls and identified about 17.7 million spam calls. BT also runs an AI network level firewall against calls from abroad that use a UK number for scam purposes.

  • Interactions handled: at least 2.4 million, May to early October 2024, scam calls blocked
    "BT’s new Enhanced Call Protect on Digital Voice has successfully blocked more than 2,430,000 scam and identified 17,700,000 spam calls to landlines since the new scam protection service from Hiya was introduced in May."
    Claimed by: organization
  • Users served: 2.5 million, October 2024
    "2.5 million BT customers already receive the new call vetting service, a benefit of migrating to Digital Voice."
    Claimed by: organization

Virgin Media O2

United Kingdom · Telecommunications · 2024

ScaledGrade B

O2 launched Call Defence in November 2024 at no extra cost. Built with Hiya, it uses adaptive AI to analyse the behaviour of unknown numbers in real time and shows a warning label on the customer's screen for suspected scam or spam calls, while also blocking known fraudulent calls. It rolled out automatically on Android and on iOS 18 and later. By March 2026 it had labelled more than 1 billion calls, and O2 reports that calls labelled suspected scam are answered 42% less often and last 89% less time than unflagged calls.

  • Interactions handled: about 70 million, per month, calls labelled as suspected scam or spam (2026)
    "O2 first launched the service for its customers in November 2024, and today around 70 million calls every month are being labelled as suspected scam or spam."
    Claimed by: organization
  • Interactions handled: at least 1 billion, cumulative, November 2024 to March 2026
    "Virgin Media O2 has today reached a major milestone in its fight against fraudsters, after using AI to successfully label more than 1 billion suspected scam and spam calls to O2 customers."
    Claimed by: organization

Virgin Media O2

United Kingdom · Telecommunications · 2024

PilotGrade B

O2 created Daisy, a lifelike AI voice persona of an elderly woman, trained with help from the scambaiter Jim Browning. Daisy combines several AI models to listen and respond to scam callers in real time without human input, telling long stories and giving false details so fraudsters spend their time on her instead of real victims. O2 says Daisy has kept fraudsters on calls for 40 minutes at a time. It was launched as part of O2's Swerve the Scammers awareness campaign and is a disruption and awareness tool rather than a protection service for individual customers.

No outcome disclosed.

Telstra

Australia · Telecommunications · 2021

ScaledGrade B

As part of its Cleaner Pipes initiative, Telstra blocks suspected scam calls in its network before they reach customers. Upgrades in 2021 made blocking more aggressive, improved detection of Wangiri one ring calls from international premium numbers and of spoofed calls that pretend to come from local numbers or trusted brands, and doubled the monthly volume blocked within four months. Telstra says it keeps evolving its algorithms and detection methods and takes care not to block genuine calls. From December 2024 it added Telstra Scam Protect, an in house network feature that warns customers on screen about calls that look spoofed, arrive from overseas while showing a local number, or come from a number with a suspicious calling pattern. Its Scam Protect article (published March 2025, updated May 2026) reports blocking more than 11 million scam calls a month on average and Scam Protect warnings on an average of 12 million calls a month.

No outcome disclosed.

How do you implement it?

A model agnostic playbook: what to prepare, the order to build in, and what goes wrong.

Data you need

  • Real time call signalling and call detail records
  • Caller authentication data where available (for example STIR/SHAKEN in North America)
  • Customer scam reports and complaints
  • Registry of verified business numbers

Systems to integrate

  • Voice core and signalling platforms for blocking and diversion
  • Handset or network based caller display for labels
  • Customer reporting channels such as the 7726 short code in the UK
  • Industry traceback and intelligence sharing with other carriers and regulators

Complexity: Medium

Specialist vendors can provide the scoring and labelling (BT and O2 use Hiya), while Telstra and Bell describe their own network capabilities. Either way, the work is integration with the voice network, handset support, handling of wrongly labelled businesses, and regulatory alignment on blocking.

  1. 1

    Start with the network signals you control

    Block numbers that should never originate calls and calls that present a domestic number but arrive from abroad, before adding model based scoring.

  2. 2

    Label before you block

    For uncertain calls, a warning label lets the customer decide and gives you feedback. Block only above a high confidence threshold.

  3. 3

    Handle false positives fast

    Give businesses a way to register numbers and dispute labels, and give customers a way to report missed scams and wrongly flagged calls.

  4. 4

    Cover every customer group

    Extend protection to landlines and older handsets, where vulnerable customers are concentrated, not only to the newest smartphones.

  5. 5

    Measure harm, not only volume

    Track answer rates on labelled calls and scam reports per thousand customers, not just the number of calls blocked.

Guardrails

  • Blocking only above a validated confidence threshold; everything else is labelled, not blocked
  • Emergency and priority numbers never blocked
  • A dispute process for businesses whose calls are wrongly labelled
  • Clear customer information about what is analysed and how to switch labelling off where permitted

KPIs to instrument

  • Calls blocked and labelled per month, per category
  • Answer rate and call duration for labelled versus unlabelled calls
  • Scam reports per thousand customers
  • Disputes from businesses and share upheld
  • Share of customers covered, including landlines and older devices

Human in the loop

Fraud analysts set blocking thresholds and review new campaign patterns, a team handles disputes from businesses, and customer reports are treated as training signals. Blocking rules follow the national regulator's requirements.

Common failure modes

Genuine calls marked as scam
Hospitals, schools or delivery firms get labelled and people stop answering them. Run a fast dispute process and verified caller programmes.
Volume as the only measure
Billions of calls blocked says little about harm prevented. Track answer rates and scam reports.
Protection only for new phones
Handset based labels miss older devices and landlines. Combine network blocking with display features for all lines.
Scammers move to other channels
Scammers shift between calls, texts and messaging apps as each channel gets harder to use. Coordinate call and message protection.

What are the risks and rules?

EU AI Act

Minimal risk

Scoring, blocking and labelling calls is not listed in Annex III, is not a prohibited practice under Article 5, and the system does not interact with people or generate content, so Article 50 does not apply. A conversational scambaiting agent such as O2's Daisy talks to callers with a synthetic voice, which raises separate Article 50 transparency questions and should be assessed on its own.

Guidance

  • Scam calls and messages (Ofcom, Europe). UK regulator hub on scam calls and messages, including its statement on tackling scam calls from abroad, which covers its calling line identification guidance on how providers should process calls from abroad that present a UK mobile number.

Controls to put in place

  • Documented blocking and labelling policy aligned with national rules
  • Monitoring of false positives and business disputes
  • Privacy notice and legal basis for analysing call metadata
  • Regular review of thresholds against new scam campaigns

Frequently asked questions

How many calls do operators block or label?
Bell reports analysing more than 4.4 billion calls and blocking or labelling over 540 million since launching Suspicious Call Detection in 2025. Virgin Media O2 labels around 70 million suspected scam and spam calls a month. Telstra reports blocking more than 11 million scam calls a month on average and showing Scam Protect warnings on about 12 million calls a month.
Do warning labels actually change behaviour?
Virgin Media O2 reports that calls labelled suspected scam are answered 42% less often and last 89% less time than unflagged calls.
Can AI also fight back against scammers?
O2 built Daisy, an AI voice persona that answers scam calls and keeps fraudsters talking, in some cases for 40 minutes, to waste their time and expose their tactics. It is an awareness and disruption tool rather than a replacement for network blocking.

How to cite this page

Blits.ai AI Use Case Library, "AI spam and scam call blocking for mobile and landline subscribers", last verified 26 September 2026, https://www.blits.ai/ai-use-cases/spam-and-scam-call-blocking. Licensed under CC BY 4.0. Method: how we verify use cases.

Changelog
  • 27 September 2026: First published

Related use cases

Telecommunications

AI for telecom fraud detection (SIM swap, IRSF and Wangiri)

AI that protects the operator's own network, revenue and numbers from fraud: it watches call, messaging, roaming and account activity to detect SIM swap and port out takeovers, international revenue share fraud (IRSF) and Wangiri one ring scams, blocks or flags them in real time, and shares risk signals with banks and other businesses that rely on the phone number for security. Scam calls aimed at subscribers are handled by call blocking.

Deployments
4 public, best grade B
Reported detection improvement
30%
Vodafone, organization claim
BankingPayments and cards

AI scam intervention for instant payments

AI that talks to the customer when they are about to authorise an instant payment that looks like a scam: it combines the payee check and the risk score, asks targeted questions about the payment in plain language, explains the specific scam pattern, and holds, delays or escalates the payment to a human specialist when the risk stays high. Unlike fraud scoring, which stops payments the customer did not make, it protects customers from payments they are being manipulated into making.

Deployments
6 public, best grade B
Reported detection improvement
300%
Starling Bank, vendor claim
BankingPayments and cards

AI for money mule account and network detection

Graph and behavioural machine learning that finds money mule accounts and the networks around them, such as circular flows, layering chains and clusters of newly linked accounts, and supports investigators in tracing scam proceeds and restricting accounts before the money is gone.

Deployments
3 public, best grade B
Autonomy
Copilot
BankingPayments and cards

AI agent for card dispute intake

A customer facing AI agent that handles the "I do not recognise this charge" moment: it finds the transaction, separates suspected fraud from merchant disputes and simple confusion, explains the customer's rights and timelines, collects the details and evidence the rules require, and opens a correctly classified dispute case for the operations team.

Deployments
3 public, best grade B
Autonomy
Supervised agent