Algorithms determine the future: Why they work more objectively and are more scalable than humans, but will never entirely replace humans.
Whether to handle customer enquiries, recommend movies, or grant loans: machine learning algorithms are penetrating more and more areas of our everyday lives. In fact, it is not exaggerated to say that algorithms have life-saving capabilities. For instance, algorithms already detect malignant melanomas, known as black skin cancer, better than most doctors. But algorithms are not only helping doctors to diagnose diseases, they are also helping retailers to better understand and predict customer behavior. Let’s take a closer look.
Machine learning algorithms process huge amounts of data – keyword “big data” – and do so much faster and more accurately than any human ever could. In retail, machine learning algorithms are becoming more and more common. This development is triggered by online pure players such as Amazon and Zalando, who use artificial intelligence to optimize their merchandise flow and automatically set prices according to consumer demand.
Algorithms for optimized pricing
When deciding the right price for a product, subjective criteria always play a role for humans. Say a particular swimsuit is not being purchased as often as it should in order to be sold off by the end of the summer season, the responsible merchandising team may be prone to say something like “It’s just because of the recent bad weather spell, this will turn around soon.” Or the new floral dress that was supposed to be all the rave, but hasn’t captured consumers’ interest yet: “The trend will come!”
Thus, retailers sometimes delay a price markdown for too long because they hope to sell a higher number of articles at the original price and profit from the higher margin. The result: merchandise sits more or less untouched on the shelves for weeks on end, until at the last minute retailers realize the need to drastically slash prices.
The better approach: data-driven pricing. Algorithms used to predict future demand and calculate optimal prices are purely data-driven and therefore one hundred percent objective. In addition, they take a huge number of influencing factors into account – in real time:
- product availability
- competitive prices
- current events, such as Valentine’s Day & Halloween
- current marketing campaigns and promotions
- the weather
- location
- purchase history and visit history of customers
- customer traffic in stores or on the website
And the list goes on…
Taking all these factors into account for each of the thousands of SKUs that a retailer carries, is clearly not humanly possible. And certainly not at the level of speed and accuracy demanded in today’s dynamic market.
And if that doesn’t convince you, then the numbers will. AI-driven pricing has been shown to yield an impressive 3-6% increase in sales and a 2-8% increase in profits.
Artificial intelligence needs human intelligence
Clearly, AI has many practical applications and the power to create a significant positive impact. Like with any technology though, the use of AI can also lead to undesired results. After all, machine learning algorithms are only as good as the data they are fed with and the specifications according to which they are programmed. This means setting the right “rules” for the algorithm to follow and building in appropriate sanity-checks. And behind this lies the number one source of danger: humans.
The AI-controlled chatbot “Tay”, which caused a stir on Twitter in March 2016, gave us a first taste of what can go wrong. The aim of the experiment was for the bot, which was programmed by Microsoft, to interact with other users and learn from them. And he did – but not as expected. “Tay” took on the racist and sexist statements of the other users without reflection and had to be taken off the net within a day.
Even Amazon is not immune to such AI “errors”. This is how a book about flies (Peter Lawrence’s The Making of a Fly) came to be priced at a whopping $23,698,655.93 in April of 2011. Clearly this price is not an accurate reflection of consumers’ actual willingness to pay. So how exactly did this happen? Two Amazon retailers were offering this book, and both were using algorithmic pricing, i.e. algorithms were automatically setting the prices. One of the two retailers had set their pricing strategy to always price slightly higher than the competitor (relying on better customer reviews to attract buyers). With no humans checking the prices, and no built-in upper limit, prices quickly spiraled out of control.
Ultimately, it is always the human being who bears the responsibility for the result of the algorithms. The desired results can only be achieved if humans input the right data and set the right strategies and business rules. AI must therefore not be a black box, the calculations and recommended actions must be displayed transparently and continuously checked by humans.
In addition, humans possess numerous abilities such as creativity, innovation, or the ability to independently make ethical decisions, which AI does not (yet) possess. Let’s go back to the swimsuit example. While an AI application is perfectly suited to calculating the swimsuit’s optimal price, and in which sizes, colors and quantities it should be distributed to which stores, that is neither the start nor the end of the process. Humans are still needed to design the swimsuit, to market it, and to ultimately advise consumers in store as to which size and color suits them best.
Working hand in hand to create amplified intelligence
Many media outlets like to pose provocative questions along the lines of “Will AI take over?” But when it comes to AI versus humans, we don’t believe it has to be an either-or question. Whether in retail or in medicine, in delivering movie recommendations or driving cars, the answer is AI AND humans. When it comes to sorting through huge quantities of data and discovering patterns, AI is unquestionably far superior to human intelligence. But when it comes to being creative and adapting, AI is no match for the versatile human mind. Combining the two yields what we term “amplified intelligence”.
In the case of aifora, our AI platform optimizes and automates important retail processes by making data-driven decisions. Carefully configured business rules prevent errors like those of books priced at $24 million to happen. Moreover, a transparent web interface gives users the opportunity to review the decisions made by “the machine”. Thus, humans no longer need to calculate and manage prices for thousands of SKUs – a time-consuming and error-prone process. Freeing them from this repetitive manual task enables them to put their retail expertise to better use, for example to make the customer journey more appealing. We think that’s a pretty intelligent move.
