5 key trends in the application of AI in retail
By Anais Masetti - 25 July 2018
We live in exciting times: artificial intelligence (AI) is no longer only a thing of science fiction as it’s finally entering the real world. With the emergence of tangible applications, the global AI market is booming and is projected to be worth 7,35 billion U.S. dollars by the end of this year.
For the moment, machine learning (ML) is primarily seen as a way to make sense of huge data sets and optimise recurring tasks. In fact, one industry where this technology could have a big impact is that of retail and e-commerce.
With hundreds of European start-ups rated in this sector, Early Metrics has built a large database of qualitative and quantitative data on early-stage ventures. Here are the 5 main trends we have identified that are affecting AI innovation and adoption in the retail space.
1. Logistics jobs are increasingly threatened by robots
Currently AI is perfectly positioned to take over low to mid-level employee positions in logistics departments. Indeed, these generally consist of very repetitive and predictable tasks so they could easily be automated and carried out by robots. The Chinese e-commerce giant Alibaba is among the first to have implemented this and its Huiyang warehouse is now run by more robots than humans, moving goods across the warehouse floor and reducing the human manual work by 70%. Of course, this evolution raises serious concerns about job security: a report by PwC estimates that automation could make 44% of jobs in the retail sector redundant by 2030.
2. Fraud detection can be greatly improved by AI
Another thing AI is great at is spotting patterns and this ability can be very useful for detecting fraud on e-commerce sites. Indeed, applying ML to an algorithm means it will learn and self-update by analysing data, getting better at detecting fraud over time. Moreover, AI algorithms are built to react and adapt much faster than a human ever could. Among the successful applications, Ocado stands out. The online grocery shop, has developed its own fraud detection system, combining its own algorithm with open-source software Tensorflow and running in the Google Cloud. According to the company, its ML system has improved its precision of fraud detecting by a factor of 15x. Another application worth noting is the use of facial recognition in video surveillance to prevent shoplifting.
3. Retailers are fast adopting automatised customer service and target marketing
Many retailers are also experimenting with AI-assisted customer service and target marketing. The latter is again made possible by the capacity of AI systems to make sense of big sets of customer data and tailor marketing messages (e.g. gift recommendation) in a timely and personable fashion. Then, certain chatbots are becoming increasingly efficient when it comes to replying to recurring questions from customers, saving precious time for employees. Natural language processing is still far from perfection but digital assistants are getting better at contextualising our requests and responding closely to how a human would.
4. Retailers have thrown their customers into the uncanny valley
As mentioned in the previous point, digital assistants are pretty good at acting like humans but the little step that separates them from real human interactions have a huge impact on customer satisfaction and consequently mass adoption. When we feel like there is something off with the customer service officer we’re chatting to and we realise “he” is an “it”, we generally get very ill at ease and become much less forgiving of mistakes. This is commonly referred to as the uncanny valley. By wanting to innovate fast, some retailers have forced their customers deep into this valley. This was perfectly illustrated earlier this year by the failure of Fabio, the friendly shopbot who was fired after week because Scottish customers found it “creepy”. The only way out of the canyon is to either stop anthropomorphizing AI behaviour or make them so efficient, they could pass the Turing test.
Have a look at what went wrong with Fabio:
5. Data privacy regulation and concerns are major hurdles for AI adoption
The biggest hurdle for the wide adoption of AI in retail remains privacy regulation. By now you will surely be aware of the newly introduced GDPR but it’s still unclear how data processing by ML algorithms sits with this regulation. Moreover, in the light of the recent Cambridge Analytica scandal, consumers are much more wary of allowing companies to use their personal data. This could therefore slow down the adoption of new solutions by retailers as they grow reticent to testing tech that could alienate their customers.
So we can agree that whilst AI is still not truly “intelligent” there are interesting applications being carried out in the retail and e-commerce sector, especially when it comes to optimising warehouse logistics. The large amount of data produced by customers is also making the intervention of AI increasingly relevant for customer service and target marketing. However, there are major hurdles to the wide adoption of such technology like privacy regulation and the uncanny valley created by customer-facing humanoid systems.
Early Metrics is constantly analysing these developments thanks its close relationship with innovative ventures and traditional retailers so we will be sure to watch out for further progress in this space – and for creepy shopbots.