Although the term Machine Learning has been around for decades, since self-learning algorithms have became applicable in business, it’s being frequently abused — just like some other terms of certain marketing power e.g. Cloud, Omnichannel or Big Data. Fraud prevention is at the heart of this trend. Most providers of fraud detection solutions talk about intelligent algorithms and Artificial Intelligence, but hardly anyone explains what in their particular case these terms really stand for. We have collected some common myths about Machine Learning in the fraud prevention industry and…dispelled them one by one to make the matter as clear as possible.
Myth #1: Machine Learning is something you just switch on and within seconds benefit from using it.
The more popular the term gets, the vaguer it becomes. Some providers of anti-fraud solutions describe Machine Learning as if it was a standardized tool one can just use or not, suggesting that using it automatically makes fraud prevention more effective. Let’s make it clear: Machine Learning can make risk management far more effective, yet it must be applied wisely.
Machine Learning is a field of computer science that includes a multitude of technologies, approaches and methods. In fact, this term has a different meaning in each single case and thus cannot be considered a feature that makes one or another solution better or worse just by its presence.
Broadly understood Machine Learning means much more than just a set of technologies leveraged to bring a certain effect. It also includes people standing behind it, their expertise and experience.
Each company has its own specificity, collects a different range and volume of data, targets different segments of customers, operates on different markets.
In order to fight fraud with Machine Learning, one needs business-savvy Data Scientists who in cooperation with companies’ business decision makers could skillfully select the right data, variables, methods and measures for the particular business case. Self-learning algorithms must be thus appropriately designed, implemented, “told” what to learn and how to learn it.
At Nethone, we create Machine Learning models for each and every organization individually. This allows us to provide the best possible solution to fight fraud.
Myth #2: Self-learning algorithms will produce satisfactory results from any data.
People tend to overvalue technology and underestimate the role of the human element. This way of thinking is tricky, specially when it comes to Machine Learning. Even the most sophisticated self-learning algorithm won’t help much without appropriate data to analyze.
Stating that Machine Learning is as effective as high the quality of data it operates on is, would be, however, only partly-true.
One can have petabytes of high quality data but Machine Learning won’t produce satisfying results unless — once again — the people responsible for creating models really understand the data that machines are meant to analyze.
Simply put, AI, Machine Learning or whatever name you choose, won’t produce results from poor quality data, though data of top quality does not guarantee spectacular results either.
The key to success is also to know what data to collect, how to combine different kinds of data, and what are the best predictors of fraud that should be taken into account. Historical data such as a register of transactions and some chargeback reports should be enough to start with.
Myth #3: Most merchants don’t have the necessary data to use Machine Learning for fraud prevention as one needs to collect and store it specially for that purpose.
As the quality of data is so important, some merchants might think that using Machine Learning for fraud prevention is unavailable for them since they don’t collect and store petabytes of data to be used by algorithms.
Nothing can be further from the truth!
First of all, the anti-fraud solution should collect the necessary data on its own and so it does once properly implemented. Merchants should not be expected to have their own rich databases prior to implementing a fraud prevention system.
Secondly, as far as the merchant keeps a register of transactions and stores historical chargeback reports, probably there is enough information to effectively start with. In other words, **online businesses of any size, with any data resources can use fraud prevention systems based on self-learning algorithms. **It’s just the matter of finding the right provider.
Myth #4: Machine Learning is for big players only as it’s expensive, difficult to implement, and solves problems encountered only by NASDAQ-100 giants.
This point actually includes 3 common myths: expensiveness, difficulty and size-as-a-condition.
First of all, AI-based fraud prevention is not expensive anymore as computing gets much more affordable.
The approximate cost of 1 GFLOPS, has decreased from around $50 in 2007 to no more than $0,08 in 2015. Same with data storage, where the $/GB relation is predicted to reach 10–2 by 2020. This has a direct impact on prices of services based on Machine Learning, including fraud prevention.
Secondly, since anti-fraud systems are offered as SaaS (Software as a Service) and in most cases integrated via APIs and plug-ins, the implementation is really simple and requires little toil from merchants. Nethone, for instance, is a SaaS solution, offering clear and concise API that holds the REST rules in utmost regard. It means that even a single developer can handle the integration process on the merchant side within a few days.
Some merchants, mostly SMEs, think that fraud prevention is necessary only for big players, with multi-million revenues and Walmart-like volumes of transactions.
One cannot be more wrong.
Fraudsters are targeting merchants of any size — **specially mid-sized businesses**and startups which are freshly on the market and haven’t had enough time to learn how to fight fraud. E-tailing giants leverage multilayer security solutions, cutting-edge anti-fraud systems and employ teams of top-grade professionals responsible for protecting their bottom lines. Therefore, for criminals it is much easier to attack a large number of SMEs who hardly use any fraud prevention mechanisms, than merchants from NASDAQ-100.
TL;DR? Key take-aways:
- Machine Learning is neither an app you can just run to get results, nor a feature that would define the quality of this or another anti-fraud solution by being used or not. It is in fact the combination of technology, humans’ skills and knowledge.
- Self-learning algorithms require data of at least decent quality to bring results but having high quality data does not equal spectacular results.
- Merchants don’t need to have petabytes of data to start protecting their bottom lines with AI-based anti-fraud systems but they should choose a provider who knows what data to collect and how to do it.
- As fraudsters poach easy money rather than cut through thick walls of big players’ security strongholds, SME merchants become their targets more and more frequently these days. Therefore, even companies with relatively low turnovers should seriously consider getting a modern anti-fraud system before it’s too late.