Risk
Case Studies
ANZ bank identifies high risk loans and predicts customer defaults with deep learning
ANZ bank has collaborated with Nvidia and Monash University researchers to develop deep learning technology. The neural network was trained on customer credit card data and is able to assess risk on a much more frequent basis than current practices and predict the client who are likely to default on payments. The system is currently a proof of concept as the bank has stated that it needs to fully understand how it works before it commercialises it.
Linked use case: Evaluate customer credit risk using application and other relevant data for faster and more efficient decisions
Adobe is developing technology to detect manipulation in images using a deep neural network
Adobe revealed that it is currently working on developing an AI system that spots whether an image has been altered artificially. It does so using a deep neural network which is trained on existing data sets of altered images.
Linked use case: Detect manipulated or falsified media
American Express Australia used machine learning to identify 24% of customer accounts that would close within four months allowing them to take preventative save actions
American Express has over 100 million credit card customers globally representing over $1 trillion in annual charge volume. The Australian company used advanced data analytics and machine learning to analyse historical transactions along with 115 variables to forecast customer churn. They were able to identify 24% of accounts that would close within four months allowing them to take preventative save actions.
Linked use case: Predict and drive customer retention and churn management
American Express identifies $2 billion in potential annual incremental fraud incidents with machine learning
American Express has over 100 million credit card customers globally representing over $1 trillion in annual charge volume. Amex used machine learning models based on an understanding of customer spending patterns alongside normal merchant purchase information. This allowed real-time detection of potential fraud resulting in an estimate $2b in potential annual incremental fraud incidents.
Linked use case: Identify fraudulent activity using unusual payment transaction patterns and other data
An Anonymous UK Insurance firm demonstrates that it can better assess the risk of insuring companies by automatically determining relevant UK court cases with deep neural networks
A UK insurance firm wanted to improve its risk assessment of insured companies by automatically reviewing court case listings that might involve their clients. This would help better align risk premiums at the time of renewal or claims. Using EvolutionAI natural language processing they were able to scan court documents and identify relevant companies and cases even if the company name spellings and use were inconsistent and varied.
Linked use case: Lower the loss ratio for insurance companies through portfolio management
Baltimore Gas and Electric generated $2.8 million in economic benefit from identifying fraud and unbilled energy usage with machine learning
Baltimore Gas and Electric Company (BGE) is leveraging machine learning to identify and tackle unbilled energy usage. In doing so, the company has generated $2.8m in economic benefit and is expecting its annual economic benefit to reach $20 million. The company is using C3 IOT''s solutions, such as C3 Revenue Protection™ and C3 AMI Operations™ to improve the operation of its advanced metering infrastructure (AMI) network.
Linked use case: Identify potential fraud from utility consumers
Beijing Police identify people by their walking gait from up to 50 metres away
Beijing Police are using gait recognition technology to identify individuals based on their physical features when walking. The benefit of this technology is that it can be used from further away than facial recognition technology. One of the vendors of this software is Watrix, although the technology does not work yet in real-time.
Linked use case: Identify security-related individual targets from range of data including sensors, camera feeds and suspect activity
Berkeley Lab to use deep learning to analyse alternate data sources for more accurate predictions of air quality
The US Department of Energy’s Lawrence Berkeley National Laboratory (Berkeley Lab) are using deep learning methods to analyse additional information, such as satellite images and cellphone data, to more accurately predict air quality levels.
Linked use case: Accelerate data integration from multiple sources
CCC Information Services predicts repair requirements from vehicle collision photos using AI
In December 2018, CCC Information Services launched an automated estimation tool for insurers. The tool, called Smart Estimate, leverages AI to analyse collision photos. Based on analytics, the system is able to automatically assess the damage and generate estimates for cost and any parts that would need replacing. The company''s Insurance customers that already use CCC ONE® Platform services can have easy access to the tool for virtual inspections.
Linked use case: Recognise and categorise type of damage to predict insurance claim valuation through accessing photographic data
Chime decreases basis point loss by 40% using a machine learning fraud detection platform
Chime online banking implemented Simility''s machine learning fraud detection platform to identify fraudulent users and payments more quickly based on historic and live data, reducing loss basis points by 40%.
Linked use case: Identify fraudulent activity using unusual payment transaction patterns and other data
China''s Hangzhou No. 11 Middle School monitors students through facial recognition cameras
The Hangzhou No. 11 Middle School has installed a “smart classroom behavior management system”, which captures students’ expressions and movements, analyzing them with big data to make sure they’re paying attention.
Chinese authorities trial facial recognition technology to identify parallel traders at Hong Kong-Shenzhen border
The Chinese General Administration of Customs has announced that a facial recognition technology system is now in use at the Shenzhen Bay and Lo Wu checkpoints. It aims to streamline travellers'' checking process at the border while primarily to identify parallel traders. No further details have been given as to what will happen when a traveller is flagged as a potential parallel trader based on the database which real-time footage is compared against.
Linked use case: Identify persons of interest to law enforcement through facial recognition
Choozle announces partnership with Oracle''s Grapeshot to determine optimal online ad placement and ensure brand safety using AI
Programmatic advertising platform Choozle is implementing Grapeshot''s AI to determine optimal ad placement locations while being brand safe and avoiding negative associations.
Linked use case: Evaluate where brand is appearing to check for inappropriate usage and placement
Cincinnati Children’s Hospital Medical Centre predicted at expert-level 91% accuracy which students are at higher risk of perpetrating school violence using machine learning on interview scores
A pilot study indicates that AI may be useful in predicting which students are at higher risk of perpetrating school violence. At 91% accuracy it is as accurate as a team of child and adolescent psychiatrists, including a forensic psychiatrist.
Linked use case: Predict likelihood of recividism or criminal activity on an individual basis
Citi reduced cost, time and reliability issues in the delivery of the complex stress test models needed for regulatory compliance through deploying ML to support the process
Citi reduced cost, time and reliability issues in its delivery of the complex stress test models needed for regulatory compliance. Citi tackled the stress testing using model acceleration and improvement technology from Ayasdi.
Linked use case: Determine treasury or currency risk situation
Clearbank combats fraud and money laundering with the use of machine learning
ClearBank has deployed machine learning technology to fight fraud and optimise its anti-money laundering (AML) operations. The U.K''s new clearing bank is using Featurespace''s ARIC™ platform that develops individual behavioural profiles and detects abnormal activity in real time, using machine learning.
Linked use case: Identify fraudulent activity using unusual payment transaction patterns and other data
Danish Danske Bank increases payment fraud detection by 60% and reduces false positives by 50% with machine learning
Danish Danske Bank had a problem with false positive that was approaching 99.5%. Using advanced machine learning they were able to increases fraud detection in real-time by 60% and reduce false positives by 50% .
Linked use case: Identify fraudulent activity using unusual payment transaction patterns and other data
Dankse Bank identifies fraudulent online banking customers with 99.7% accuracy in a pilot with the BehavioSec behavioural biometrics system
Danske Bank piloted Behaviosec''s BehavioWeb product for fraud detection. BehavioWeb analyses customer actions when they log in to online banking and perform transactions during a session and compares them with a threshold for determined by machine learning in an effort to reduce false positives and false rejections for fraud detection. The pilot program had a 99.7% success rate of correctly identifying fraudulent users with legitimate ones.
Linked use case: Accelerate identity verification for new and existing customers
Danske Bank prevents card fraud with the use of machine learning
Danske Bank has leveraged machine learning technology to fight fraud across all its payment channels. Featurespace''s ARIC Fraud Hub prevents fraud by detecting anomalies in a user''s behavioural profile, in real time. The bank is mitigating the risk while also eliminating false positives to offer greater customer experience.
Linked use case: Identify fraudulent activity using unusual payment transaction patterns and other data
Direct Line Insurance uses machine learning to automatically flag fraudulent claims in automotive section with 75% hit rate
Direct Line Group, a Spanish insurance company, is using Shift Technology''s Force platform to better detect fraudulent claims. Assessing the claims, the platform alerts human analysts if it identifies patterns of fraud in the claim. The system flags 1000s of claims every year. Automating the process has sped up claim processing time for all customers.
Linked use case: Enhance insurance customer fraud detection by assessing behavioural data
Drivers demonstrate Tesla''s Autopilot limitations in recreation of fatal Model X crash at exact same location
Two Tesla drivers have almost recreated the fatal accident of March 23rd in Mountain View, California. The uploaded YouTube videos show them driver their Tesla cars at the exact same location using the Autopilot feature, which fails to identify faded lane markings.
Linked use case: Visualise and recognise road layout, other vehicles and pedestrians and potential risks to empower predictive driving
Dutch police plan to use machine learning to detect illegal substances sent in the mail based on X-rays
The Dutch police are planning to implement an AI system which analyses X-ray scans to detect illegal substances, such as drugs, contained in letters and parcels without opening them. The pilot will begin in January 2019.
Linked use case: Identify physical properties of scanned images
Earthport Payment Network reduces false positives of automated suspicious transaction detection using AML risk data in real-time
Earthport implemented ComplyAdvantage''s Transaction Monitoring platform to analyse data in real time to identify and report suspicious payment behaviour which may be linked to money laundering or terrorism-financing. Minimising false positives for this behaviour reduces the amount of human intervention needed making the process more efficient.
Linked use case: Detect potentially fraudulent or nefarious users
Elections Canada plans to use AI to identify electoral misinformation campaigns on social media
Elections Canada plans to use AI to monitor conversations on social media with regards to elections. The intent is to identify misinformation campaigns before they spread as a safeguard.
Linked use case: Censor user generated content on social media and other platforms
Enel improved the average energy recovered per non-technical loss inspection by 70% in Italy and more than 300% in Spain using machine learning
Enel leverages C3 IoT to identify electricity theft (non-technical loss) and recover unbilled energy. The solution applies machine learning and analytics to calculate the probability of fraud for each customer meter using data from seven Enel source systems. The company managed to improve the average energy recovered per inspection in both Italy and Spain.
Linked use case: Identify potential fraud from utility consumers