M...
Highlighting Fraud with Artificial Intelligence
We have taken very careful design decisions all with the aim of enabling customers to address any type of fraud they face. That means we can ingest any data, generate any feature and provide you with the best machine learning models for identifying fraud and any rule. We have state of the art in-memory databases, algorithms, and low level code in our solution which makes it lightening fast whilst keeping the costs down. So what does a next generation AI solution look like?
Greg Hancell
Current state of play
Weโre in a transition period where society is almost expecting Artificial Intelligence services as part of ย their daily lives. There is great value shown in uses such as the large language models of ChatGPT4.0. However in Fraud Prevention solutions there is hesitancy to trust machine learning models, due to many reasons:
- Lack of understandingย in how the model works
- Lack of trustย in the scores provided by the model
- Lack of dataย available to train the model
- Bad historical performanceย prior generations of fraud solutions using โAIโ but with poor performance
- Misconceptionsย generally around how Machine Learning works and its limitations
These are very valid reasons and you the reader are right to be cautious in using AI in fraud prevention solutions.
Today I aim to provide information that will help answer some of the questions you might have about AI and build trust in Lynx Fraud Prevention.
First let me explain that not all solutions that โutilise AIโ or โuse Machine Learningโ actually use it, or at least use it in a meaningful way.
Ultimately, any solution that makes use of AI should be able to confidently perform a proof-of-concept and outperform any rules based solution. To give you an idea, Lynx Fraud Prevention outperforms known leading rules based solutions by a factor of up to 100.
That means 10 times less alerts with 10 time better accuracy, meaning we give you less alerts and catch more fraud. We will gladly prove this to you in a POC, as will any vendor that is truly using AI in their solution, as itโs simple to do.
So how do we utilize AI in the Lynx Fraud solution?
Next Generation Fraud Prevention
We have taken very careful design decisions all with the aim of enabling customers to address any type of fraud they face. That means we can ingest any data, generate any feature and provide you with the best machine learning models for identifying fraud and any rule.
We have state of the art in-memory databases, algorithms, and low level code in our solution which makes it lightening fast whilst keeping the costs down.
So what does a next generation AI solution look like?
- Data agnostic approachย meaning we can ingest any type of event, transaction, file you need to protect
- Build any featureย (intelligence points used by a model) meaning we have veryย comprehensive intelligence
- Auto feature generationย means we automatically classify data and generate relevant features.
- Realtime in memory feature calculationย meaning our models always have up to date data
- Daily Adaptive Modelsย meaning the models do not drift and stay relevant to new attacks and products.
- Monitor the performanceย of the model to bring you confidence that it is outperforming the competition and massively outperforming any rules based solution.
- Apply supervised learning,ย which outperforms unsupervised learning by up to a factor of 10.
- Genuinely multichannelย meaning we are not blindsided by cross channel attacks and are able to identify attacks traversing multiple channels.
Not only do we do this, we also return a result to the calling application in just 25 milliseconds and enable you to blend this score, if you wish into a rule.
This means you can use the score on its own, as many of our customers do. And/or if you want to, you can blend the score with rule logic to reduce the amount of false positives received from the rule. Remember this will reduce false positives by a factor of up to 100. Thatโs up to 100 times less work for you and your team for the same amount of fraud identified.
Legacy Solutions
What does a legacy AI solution look like?
- Claim to use AIย however they do not use labelled data, remember supervised learning can reduce false alerts by a factor of 10, versus unsupervised learning which doesnโt use labelled data. So, if you see another solution utilizes unsupervised learning, thatโs up to 10 times more work for you and your team for the same amount of fraud. Unsupervised learning has itโs benefits for example in facial recognition, image recognitition, anomaly detection, nlp and other types of tasks where identifying data types and classifying data is onerous. However it should not be used as a main approach in fraud prevention, as it is inefficient as the main approach to identify fraud. Instead it can compliment a performant solution with additional atypical interaction and transaction identification.
- Claim to train their models regularly but do not and need a data science team.ย We train our models every day and have been doing so with top financial institutions automatically for years. We do this without the need for a data scientist on your side. This means that we not only outperform other solutions claiming to use AI, but we you will be able to identify solutions that do not truly know how to apply machine learning i.e. those that require you to uplift your data every quarter and manually retrain your models.
- Professional Services costs are kept lowย as you donโt have the professional services overhead or the data science cost on your side to keep the model relevant.
- Rules based.ย This means that every new attack needs to be studied and identify how it beats the current rule set. Not only that but attackers are able to interpret rules and logic based on machine learning attacks and beat the rule set. Ultimately a rule set is not adequate to defend against todays advanced attacks.
- Claim to continuously learn, yet when you review the data that the models are using you realize there is a delay of a day. Meaning you and your customers are vulnerable to attacks until the new data is available.
Attacks we can identify
Lynx fraud prevention can identify and stop the following attacks:
- Merchant Fraud
- Account Takeover
- Authorised Push Payment Fraud (APPF)
- Social engineering based attacks
- Phishing victims
- Skimmed cards
- Replayed transactions
- Stolen identity
- And more
We do this by understanding your users better than the competition. We deeply understand the device, user behaviour, locations, travel, spend, patterns of interaction, their associated beneficiaries, how much money they typically transfer/ spend and when.
Why donโt you give us a try?
We were born out of data and data science. We live and breathe data, algorithms, insight and intelligence. We ensure that:
- Our models are the best in the business
- We reduce your costs by reducing false positives
- We reduce fraud
- We reduce the complexity of rule building
- We improve job satisfaction and alert fatigue by giving you meaningful alerts
- We continuously learn to changing attacks and new products / customer behaviour
- We ensure our models are drift and attack resistant
Weโre confident that weโre able to stop the attacks you face and have been doing so for the last 20 years. Donโt take our word for it have a look at other financial institutions that love our solution.
Weโre the AI Solution youโve been patiently waiting for since they came onto the scene.
So why donโt you reach out and ask for a proof of concept today, you wonโt be disappointed.