E-commerce scams are exploding in Singapore - here's how AI can combat this

Scams are exploding in Singapore. E-commerce scams in particular - one of the most prevalent types of scams in recent years - have doubled just this past year - going from 4,762 cases in 2022 to a whopping 9,783 cases in 2023.

This is an incredibly worrying trend, especially as online scamming has become the theft technique du jour for criminals around the world. It's clear why this is the case.

Most people’s digital footprint is outpacing their physical one by a long shot and criminals want to capitalise on this. Plus, these criminals don’t need to put themselves in mortal danger when stealing in cyberspace; and when they do, avenues of redress for victims are essentially non-existent.

Given such a skewed incentive structure, why wouldn’t they try to part unsuspecting internet users from their hard-earned money? This problem is exacerbated by most people’s relatively novice understanding of cyberspace and how to protect themselves online.

But we can’t entirely blame ourselves.

We’re not evolved to handle this newfangled virtual world that a few dozen engineers in Silicon Valley cobbled together over the past few decades.

While millennia of human evolution have made us relatively adept at avoiding physical dangers in this natural world we’ve inherited, we’re still coming to grips with this new, pixelated world we’ve breathed life into.

So naturally, many are faltering while navigating this uncharted territory, especially those who are older and aren’t as internet savvy as most millennials and Gen Z-ers are. This is why the Singaporean government’s newly released E-commerce Marketplace Transaction Safety Ratings (TSR) are a step in the right direction.

Singapore’s marketplace ratings

Published by the Inter-Ministry Committee on Scams (IMCS), TSR rates different e-commerce platforms based on the anti-scam measures they have in place using the metrics below:

1. User authenticity;

2. Transaction safety;

3. Availability of loss remediation channels for consumers; and

4. Effectiveness of their anti-scam measures.

Based on these metrics, here’s how the most popular e-commerce platforms in Singapore stack up:

Sure, these ratings help but more can and should be done. This is where AI can be especially useful.

AI’s potential to fight scammers

Artificial Intelligence (AI) is going through its Cambrian explosion - a period of “weeks where decades happen”. It’s poised to be an overwhelming force for good. While many who fear AI are well-intentioned, they are over-indexing its potential to wreak havoc and under-indexing our ability to harness its immense potential for human flourishing.

After all, we have mastered every technology before us, and by doing so, have turned barren deserts into vibrant cities, and transformed ourselves from hunter-gatherers into space farers in just a few thousand years - a minuscule amount of time on a cosmic timescale.

Technology is a force for monumental good. Similarly, I’m confident AI - one of the most transformative technologies of our age - will be a force for monumental good as well. One major immediate good it could do is greatly reduce e-commerce scams.

The most prevalent types of e-commerce scams are:

1. Phishing and stolen credit cards. This accounts for 30 per cent of all fraud attacks on online merchants;

2. Friendly fraud by customers requesting chargebacks for items they ordered and received. This accounts for 40% of all fraud attacks on online merchants; and

3. Merchant fraud where scammers clone legitimate e-commerce stores and masquerade as them.

To incentivise e-commerce platforms to develop AI capabilities that will help prevent such scams, ICMS should include the development of novel scam-prevention AI in its evaluation metrics.

So how can e-commerce platforms step up their AI game to combat fraud? Let’s look at three ways this can be done.

Anomaly detection using machine learning

This method of fraud detection and prevention works best when applied to frequent customers.

Once someone has transacted on an e-commerce site a few times, the AI embedded in the e-commerce site can study their online behavioural patterns. Using its machine learning algorithm, it can come up with a range of behaviours considered “normal” for this user.

If their usage deviates from what is deemed “normal” based on past behaviour, the e-commerce site could throw up a verification check to ensure no fraudulent party has assumed control of the user’s account.

Examples of this would be:

1. User who has a history of purchasing low-value items is attempting to purchase a high-value item;

2. User changes their address after consistently ordering items to a different address in the past; and

3. User’s click patterns and page navigation patterns are incongruous with their past usage.

Biometric authentication using computer vision

Every transaction on an e-commerce site could require facial recognition or fingerprint recognition-based authentication.

This isn’t a wild suggestion, as every iPhone user will be able to testify. Apple has already implemented biometric verification (face ID or touch ID) for all Apple Store purchases.

So why can’t e-commerce sites implement this, especially since phones are fast becoming most people’s primary computing device?

This simple AI verification layer will ensure scammers can’t use stolen credit cards or phones to authorise fraudulent transactions. This technique will entirely neuter identity theft - one of the most prevalent forms of online fraud.

This technique would also eliminate chargeback disputes - a form of “friendly fraud” where a customer alleges that an order was fraudulent once they receive the item, triggering a refund - the modern version of trying to have the cake and eat it.

Storefront likeness detection using machine learning

Triangulation fraud - when a scammer clones a popular, existing e-commerce store - is mushrooming.

These fraudulent stores then buy the corresponding product from the actual store and ship it to the customer. But in the process, they steal the customer’s card and personal data.

To combat this, AI algorithms that detect store likeness, recency of store setup, legitimacy of store owner’s profile and other relevant metrics should be developed.

Stores that seem fraudulent should be temporarily frozen, pending a more thorough review, thus ensuring users aren’t scammed in the interim.

Sure, implementing all these AI algorithms won’t entirely eradicate e-commerce scams, but, if they are honed over years and trained on an extensive enough dataset, I am confident a majority of these scams will soon become yesteryear concerns.

Kathirgugan is an inventor and food robotics pioneer who has worked in Silicon Valley, Shenzhen and Singapore. He believes in the power of technology and capitalism to make the world a better place. All views expressed are the writer's own.