How to predict what a person will want in five years
Using neural networks is one of the ways to change
Sales history is used for analysis, typestore, its assortment. American retailer Macy's uses similar mechanisms. Each user interaction with the site updates the array of data about the person, and machine algorithms respond faster to fresh information than real employees. Nike has built whole new Nike Live stores around the idea of personalization, with which the customer interacts only with the installed application - this helps him become part of the community and receive the most personalized offers, as well as monthly gifts from the brand. With personalization, Nike increased the likelihood of purchasing its products by 40 times.
Neural networks are capable of predicting not only the effectfrom promotions. On the websites of online stores, they analyze a person's previous purchases and conclude, for example, that the sugar purchased a month ago should run out in the coming days. So it's time to offer a person to renew his reserves.
Chatbot development is another applicationneural networks. Virtual assistants eliminate the need for a large call center staff, while working quite efficiently. They give out more detailed information at a faster rate than a real person, and answer any question about a product or service - right up to the nearest outlet address.
In online stores, neural networks are able to createpersonal recommendations not only based on what a person has viewed recently, but also taking into account his portrait (gender, age, nationality and other parameters).
Analysts predict explosive growth in investment inAI projects after the pandemic. Among IT startups, there are more and more projects based on artificial intelligence and ML, because there is a demand. Retail is increasingly using AI: to select assortments in stores, develop promotions, predict prices and demand for goods. Full-fledged stores working on neural networks appear - Amazon Go, Pro Market in Skolkovo. Big Data analysis and processing by neural networks allow, for example, to see that users who tweet with the # sneakers tag also often attach #ASICS or #Nike tags. This signals the retailer which products to include in their ad campaigns more often.
At Amazon, AI makes product selections thata person can add a cart right now. To do this, the cohorts of users of the site or mobile application are analyzed, information about what these users like and what not, what other people (similar to the one for which the product is currently being selected) have looked and purchased. Conventionally, in December, an American woman will be offered goods for Christmas, and a Russian woman will be offered something related to the New Year. Thanks to neural network-based recommendation engines, Amazon generates 55% of its sales. The company says it predicts user behavior even five years into the future.
In 2016, Amazon provided access to the originalcode of his smart recommendation algorithm, and also invited other players to integrate these mechanisms to themselves. In a recent report from the US House of Representatives, Amazon was accused of monopoly (in the e-commerce segment) and of using data from competing sellers for its own benefit. And according to the Wall Street Journal, Amazon employees are analyzing third-party sales data to work on their Amazon-branded products.
How advertising works that knows everything about you
Marketers are actively using pixels fromFacebook and Google. A big plus of such codes is that the more different companies use them, the wider the base becomes and the more efficient is the analysis of the data obtained. And the more often the user visits the site, the more actively his ID base (personal folder with information) grows.
The pixel collects more than just static information(for example, IP, which allows us to understand the user's location), but also dynamic - the actions of a person on the site. Conventionally, if he looks at two shirts in the catalog of an online store, the neural network can offer him to get acquainted with other similar models or pick up parts for a complete ensemble: trousers, a jacket, accessories.
Depending on how exactly it is implementedpixel in the page code, the moment of information collection is determined. It can be configured to define targeted actions that are not related to website reloads and page changes - for example, a user likes a product or marks it with an asterisk to be placed in a wishlist. Also, the pixel is configured to analyze information about page reloads: this allows you to analyze exactly where a person is visiting. The third option is to implement a pixel for clicks on links, including affiliate ones. Due to this, you can track the third-party interests of a person. For example, on the website of chandeliers, he sees a proposal for a new collection of porcelain stoneware from a partner and goes there.
Technologies work not only in a straightforward manner:if a person is actively studying baby carriages on different sites, the neural network will show him an offer from a reproductive medicine center or a crib manufacturer. Because the algorithms have already considered this person a parent and are ready to submit several relevant proposals at once.
Companies are actively purchasing data on typicalpatterns (patterns) of behavior of various categories of customers, can exchange pixels with partners and multiply the base. If we consider the Facebook Pixel, then the person's Facebook account, the changes that have occurred in him (divorced, changed jobs, etc.), the actions taken from him, up to stopping attention on the ad (even if you do not click on it), also provide additional information.
The pixel works in conjunction with cookies:these are data files that are located on the user's device and are an information source for marketers. This is a login in social networks, products for online stores selected in the basket, search queries and much more. Collecting this data isn't just for marketers: it makes life easier for the users themselves. For example, a person is logged into Facebook and goes to different pages. He does not need to re-enter his login and password every time he reboots - the site that saved the cookies did it for him. The fact that the browser has memorized the geolocation and does not try to suggest either Dubai or Morocco on every page is also a merit of cookies.
Neural networks of recommendation services
How far did recommendation services go?See the example of a virtual assistant created by Macy’s in conjunction with the Watson Marketing platform. Neural networks track the history of a visitor's purchases on a website or application, analyze his geo-location, as well as the behavior of similar customers. After that, the virtual assistant offers goods that are suitable for a person not only on the basis of his previous purchases (conditionally the fifth white sneakers), but also taking into account his mentality and other national characteristics. For example, a dedicated animal advocate in the recommendations will definitely not receive either a fur coat made of natural fur or a bag made of calfskin.
Amazon is also developing anotherrecommendation service based on neural networks: now smart algorithms analyze which products the user of the site has liked, and offer products that are relevant to him. Moreover, tips can be issued already at the first visit to the store: it is enough to choose the ones you like from the proposed options (random selections of the day on Pinterest work in a similar way). The neural network will process the data and provide relevant offers. The idea is intended to solve the question “I don’t know what I want” among site visitors. According to Amazon, this is a step towards innovative shopping: the ability to receive only useful recommendations, without having looked at a million products. The tool works not only on the website, but also in the mobile application.
In addition, Amazon began to train a neural network.study the strategies of customer behavior, taking into account the length of the search query, the purchase price and the relationship between the goods already purchased (placed in the basket). It is assumed that people who drive in too long or too short queries are more flexible in their choice and it is easier to interest them in something that they did not initially plan to buy.
However, recommendation systems based onNeural networks are not only in retail: a similar product has been developed by the streaming service Netflix. The system takes into account standard criteria such as browsing history, ratings, favorite actors and genres, as well as the time of day of logging into the service, used for this device, the preferences of other users with a similar "profile". Interestingly, personalization even goes as far as choosing a cover for a specific user of the service: previously, the viewer was shown the one that was viewed more often. And now each person sees an image selected for him.
Taking into account the speed of development of neural networks, alsoIncreased by the pandemic, tools that allow companies to achieve even greater personalization will be in increasing demand and thus transform. It is highly likely that predictive mechanisms that work more efficiently than any person will come to the fore. And if today the store does not offer a mink coat to a convinced follower of Greenpeace, it is possible that tomorrow the car will feel the intention of a person to become a zoo activist even before this decision is made in his head.
The first accurate map of the world was created. What's wrong with everyone else?
Uranus has received the status of the strangest planet in the solar system. Why?
NASA told how they will deliver samples of Mars to Earth