Picture: ISTOCK
Picture: ISTOCK

As companies  around the world recalibrate their businesses to keep up with the shift from physical retail to
e-commerce, the spectre of innovative digital scammers looms ever large.

It’s not just companies that stand to lose. Users of platforms such as online classifieds are also at risk of being exposed to fraudsters and dangerous criminals.

The sheer volume of advertisements placed online presents fertile ground for criminals who play the numbers game, lying in wait for unsuspecting victims.  

It has become clear that human moderation of every online interaction, every time, is near-impossible. Far from the Terminator movies’ doomsday scenario with Skynet trying to kill us, it is glaringly obvious that artificial intelligence (AI), and machine learning, provide the best chance businesses have in winning the fight against online scammers.

Thought leaders and conference speakers have been waxing lyrical about the potential of AI for some time already, but just how important is AI for the world of business?

Alistair Hofert, PwC cognitive and iIntelligent automation lead, says:  “Our PwC global study shows that AI will have a global GDP economic impact, with an uplift of $15.7-trillion by 2030.

“We see AI as being integral to consumers’ everyday lives and an integral part of the way businesses operate globally. Significant investment in technology, science and research and development have made AI one of the top investment priorities of companies worldwide.”

Hofert defines AI as the theory and development of systems that sense the environment, make decisions and act in a way that would normally require human intelligence. If machine learning and AI are used to support and augment human intelligence, how does this look in the real world of online classifieds?

A recent interaction with a supposed seller on the online classifieds website OLX resulted in an e-mail being sent to the buyer — who had initiated contact with the seller shortly before — at 8.47am on Friday, October 15. The e-mail read: “Suspicious user detected. This person has been removed from OLX. Please don’t communicate with them or exchange any money or items. Contact us for more support.”

The e-mail from OLX was a breath of fresh air, as it arrived in the inbox less than 12 hours after a scammer from another online advertisements platform tried to extort money from the same buyer. “Listen, there are a lot of people interested in this car. A guy called me now and asked me for my banking details, he is very serious about the car. I don’t know if you want to beat him to it?” 

The request was an instant money transfer that night without seeing the vehicle. This WhatsApp voice note landed at 9.52pm on Thursday, October 14, after extensive interaction between the buyer and the supposed seller of a pristine Hyundai Getz “in massive demand”.

Mark Kritzinger, GM of OLX SA says: “Scammers and fraudsters are getting smarter. They are learning from each other and in some cases, building global networks and fraud rings to defraud unsuspecting users. These attacks are not restricted to OLX alone, or even to the online classifieds industry.

“As we scale, it is impossible for us to constantly review our ads using manual methods. The integration of machine learning in our moderation process is essential for success.  OLX is investing heavily in big data expertise within the organisation, hiring top data scientists from around the globe, as well as incorporating machine-learning technology in our products. We make sure that everything we do is data driven, especially in our techniques and approach to fighting fraud.”

Kritzinger says that since the beginning of 2018 OLX has removed more than 250,000 “suspicious users” from its platform. In total 1.3-million ads have been removed. “This is primarily through a combination of human and automated detection. Most of these ads are flagged through automated detection. However, in a few cases, we still rely on human moderation to review and validate the decision before taking action,” he says.

Hofert says that in order for fraud to be detected successfully, one would need to train machine-learning algorithms on large data sets from prior transactions and use a human in the loop method to train the algorithms on variations or exceptions.

“This has been successfully deployed in the insurance industry and this has been exhibited in multiple geographies globally,” he says.

“Patterns in behaviour are detected using a combination of robotic process automation, machine learning, natural language processing, computer vision for images and virtual agents. These form an ecosystem that learns over time, based on the transactions conducted online and will predict and alert humans to suspicious activity.”

Kevin Derman, CEO at Johannesburg-based Kaskade.cloud, a managed services provider that utilises Amazon Web Services, says that SA is very active in the adoption of AI.

“The South African AI market is extremely active. A large number of start-ups are incorporating elements of AI into their solutions, which has been made possible by the availability of AI technology by the mega cloud providers. This year saw the inaugural event of the AI Expo Africa in Cape Town. It had over 400 people attend and was held over two days,” Derman says.

When it comes to preventing scammers using online advertisements, Derman says there are two areas where it needs to be tackled: ascertaining the likelihood of the seller being a scammer and the second is where the seller gets a response to an advertisement.  

OLX uses contextual massaging, where it notifies users in chat messages in specific situations “where we detect unusual behaviour”. In addition to this, OLX uses “a user's online device identity to monitor their footprint across the internet for suspicious patterns”, and registration or login captchas to authenticate users that are signing in from unrecognised devices or locations, among other interventions.

Derman says that while it is more complicated to protect a seller once the calls start flooding in for their product — unless the platform acts as an intermediary and collects data from the possible purchaser too — ultimately, human common sense is priceless.

“The best type of intelligence here is ‘HI’ or human intelligence. If it sounds too good to be true it probably is. Take cash only and only meet in public places. This usually scares off most of the scammers.”