Interviews, insight & analysis on digital media & marketing

How AI is solving a headache for ecommerce platforms

By Chris Downie, CEO at Pasabi

“With great power comes great responsibility”, advice that now applies equally to ecommerce platforms as it does to superheroes! Throughout the pandemic, our reliance on online marketplaces and ecommerce has grown at a fantastic rate. In 2021, ecommerce accounted for an estimated 19.6% of retail sales worldwide. With the prediction that this will rise to 24.5% by 2025, ecommerce platforms are likely to continue enjoying a period of sustained growth. Success, however, comes with increased risk and responsibility – to provide customers with a safe, fraud-free online experience.

As customers shift to buying online more frequently and in greater numbers, the risk of fraudulent behaviour also increases. From fake listings to counterfeit products and fake reviews, fraudsters are good at spotting opportunities to capitalise on businesses’ growth and success.

Add to that the new legal responsibility that legislation brings in the form of the forthcoming EU Digital Services Act, and the US Shop SAFE Act, and ecommerce platforms need to employ strong risk mitigation strategies to combat fakes and frauds.

AI provides a more strategic approach

The ‘whack-a-mole’ tactic has never been enough. Human experts can spot individual infringements, but the task of analysing thousands of new profiles, reviews and posts every day is impossible. Given the volume of data produced by e-commerce platforms, and the considerable growth to come, an effective strategy requires the right technology.

Thankfully, artificial intelligence (AI) loves big data. A combination of AI and behavioural analytics techniques now provides forward-thinking platform owners with a new and unprecedented level of insight into the behaviours of fraudsters. This allows them to deploy a more strategic approach to fighting fraud at scale.

How AI & behavioural analytics can help

AI handles large datasets very efficiently and can process millions of profiles, customer reviews, and product listings across platforms’ entire data sets. It can also look at data from external sources, such as social media and third-party websites, for a better comparative view.  Even the most sophisticated fraudsters leave behind digital fingerprints that technology can be trained to spot. If bad actors are infringing on one platform, they’re likely to be doing the same elsewhere too!

Sophisticated models, in the form of machine learning classifiers, categorise data into different types of information. The results are scored to look for specified patterns indicating the likelihood of a review, profile or product being fake, and are used to build a watch list of suspicious profiles.

Crucially, behavioural analytics goes further by providing insights into the actions of platform users. By combining all the data points gathered, cluster technology and behavioural analytics can uncover the negative actions of groups of individuals working as one organisation, or a series of ‘bots’ (automated systems using fake accounts) being deployed alongside real users. This approach focuses on the behaviours of bad actors rather than posts in isolation. It’s only through this kind of sophisticated technique that platforms can uncover and tackle the scale of the hidden problems within their data.

In the case of fake reviews, analysis could surface clusters of suspicious activity where third-party sellers have bought reviews from review sellers. These ‘paid for’ fake reviews could be positively-biased to boost their business profile or negatively-biased to harm competitors.

With counterfeits, similar connections could be uncovered to find covert groups of fraudsters selling fake goods. The ability for people alone to analyse the vast volumes of data and identify these connections on the scale required is unfeasible.

Insights inform actions

We all work with limitations. Limits on funds or human resources mean that we need to focus our efforts to ensure we stay on top of our challenges and remain as effective as we can.

By combining the power and scope of AI with human expertise and specific market knowledge, the outcome is detailed analysis regarding the scope and threat of risks involved and, most importantly, the identification of the biggest threats for action. Enforcement teams can be much more effective, directing resources where they can have the biggest impact. Simply put, it can make a seemingly impossible task possible.

It often feels like fraudsters are continually one step ahead with their ever-evolving tactics, which has traditionally made it a significant challenge for platforms to keep up. Thankfully, the technology required to tackle fraudsters is also evolving. With regulation sharpening its focus on online content, greater responsibility means the need for more powerful technology.