Currently, IoT consists of around 30 billion devices used in various industries around the world, and it is expected to double in 2024.
At Netguru, we are happy to be a part of this transition. We have been building solutions for retail services for over a decade now, and we could see this transformation in practice many times. In this article, we are going to show how IoT paired with Data Engineering, Machine Learning and Business Intelligence, has been transforming the retail world.
What is the Internet of Things (IoT)?
First, let’s take a step back and explain what exactly is the Internet of Things (IoT) we’re talking about here. IoT is a network of many devices, distributed within a certain location, which measure, process, and transfer data based on particular events. Be it temperature, humidity, or any physical phenomenon, like pressure, or strain. It could also be a video, if we treat cameras as sensors. In fact, any device that captures information can be used as a “thing” in IoT.
I came across the Internet of Things concept for the first time when I was at the University, doing my Computer Science studies in 2016. At that time, the Internet of Things was mainly used for simple projects, such as home automation, irrigation systems for houseplants, or household temperature monitoring. This was a prolific period of maturing IoT technologies, such as MQTT (the protocol for Internet communication in IoT), or Node-Red (Open Source protocol prototyping tool), or Data Engineering technologies (Apache Kafka, Apache Airflow).
At the university, we would build simple IoT projects using hardware platforms like Arduino or Raspberry Pi, and it was a lot of fun! Today, IoT has gone to a completely new level. Thanks to the development of Data Engineering, Machine Learning, hardware, and broadband networks, IoT can soon become a key technology across many sectors. For retail, it is becoming very apparent.
Saving time and money with retail data analytics
As with many things in life, the more advanced we are in technology, the more demanding we become. In the case of retail, a major factor that contributes to a business’s success is TIME. In the ever-accelerating world, people do not want to wait to be served. People have started to move to online shopping, because they do not want to waste time going to a supermarket, waiting in queues, or being surprised with products going out of stock. There’s good news, though! All those issues can largely be addressed with IoT.
There is just one thing we should bear in mind: using many devices – dispersed, but connected by one network – generates copious amounts of data. With retail data analytics, we can optimize processes across the whole chain and gain insights that will enable us to make strategic decisions in order to increase business value. This is the whole point of IoT adoption in retail.
Standard, well-known approaches are being replaced with data-driven processes, which helps customers get what they want and businesses save money as well as attract more customers. IoT already accomplishes that, and the odds are the benefits from IoT will continue to grow, as there are few major factors that will make IoT even better at the job.
Factors speeding up IoT development
Firstly, the increase in the computational power of IoT devices is progressing very fast. Currently, major companies like nVidia and AWS are shipping advanced edge computing devices that help solve complex tasks. It is not just predicting the humidity, or calculating the distance to detect a passing customer anymore. There are specialized devices with the compute power comparable to that of a PC GPU, which speed up the process of creating more advanced applications for IoT.
Secondly, the broad adoption of 5G networks can greatly accelerate the technological development in IoT. The current 4G technology can only provide download speeds up to 100Mbps, whereas the download speed of 5G can go up to 10Gbps! 100 times more than 4G! This means faster communication, more advanced algorithms that could be deployed on devices, and better real-time insights!
These two factors combined present a huge opportunity for IoT development within the next few years. But why do we need more computational power for devices? And why do we need that power in retail stores – can’t we just perform computations in the cloud?
Why does retail need more computational power for devices?
To leverage the data captured to the maximum, we need to deploy more advanced Machine Learning algorithms. That’s where Deep Learning comes in. Currently, Deep Learning is the most active, and the most developed branch of Machine Learning. If you have recently watched a video or read a blog post about another major breakthrough in AI, it will most likely have featured Deep Learning.
Actually, Deep Learning is already addressing many challenges. In retail, for instance, it helps detect queues, customers, or out-of-stock products, or even produce complex heatmaps of supermarket journeys, which wouldn’t be possible if it wasn’t for Deep Learning.
DL solutions analyze videos in search of people waiting in a queue. If the system detects people waiting in a queue, it notifies the store staff. Deep Learning algorithms can also help find empty shelves via video surveillance. An algorithm will check which product is missing from a shelf and notify the staff that it is out-of-stock.
But nothing comes without a cost – Deep Learning needs significant computational resources, such as GPUs. On top of that, a device, having a higher computational power, can host more advanced Machine Learning models, performing more advanced tasks. In the ML team at Netguru we constantly check and research which approaches work best. I can tell from experience that to increase the performance we almost always need more power.
In the past years, there were no devices that had built-in GPUs to be used in stores. But with technological advancement, it has changed, the GPUs are now shipped in small devices that can be deployed in stores. The most common devices are nVidia Jetson Nanos, which can process videos even in real time due to their very efficient GPUs, but there are many more devices to choose from.
Edge computing vs. Cloud in retail
Why deploy Deep Learning solutions in retail rather than store data in the Cloud? Imagine a retail brand that has 1000 stores, and each store is IoT-driven. All these stores send the data to the cloud to perform computations. Consequently, we get massive amounts of data that need to be first preprocessed, cleaned, and finally run through the algorithms. And Big Data needs significant computational power to handle it.
Alternatively, we can have a separate “cloud” in each store. The data is preprocessed, cleaned, and run through algorithms in each store on its own, and only the final outcome is sent to the cloud. The cloud deals only with data analytics and insights, and leaves the hard work to the stores. This concept is known as edge computing, where the necessary computations are moved away from the central server performing all the transformations and transferred to the premise, where computations are done in place. Once the computations are done, the final message is sent to the server (it can be a hosted service in the cloud).
Of course, at the beginning, we need a much bigger financial investment because the purchase of devices is more expensive than “renting devices in the cloud”. But in the end, after having paid off the devices, we save more.
Data Science-driven IoT examples in retail
IoT is already widely used together with Data Science, and in the near future, companies using those technologies will most likely outperform stores still using the traditional business model. Below you can find some current technologies already adopted in retail.
This category is so broad that it would need a separate article to be covered completely. Video data analysis offers unlimited possibilities when it comes to process optimization. That’s why video analysis has already been widely implemented in many retail chains across the world.
The obvious application of video data analytics in the retail industry would be queue detection. Hikvision, Prolux, Axxonsoft, or Easyflow provide solutions for analyzing the number of people waiting in queues. They also give insights on how queues change over the course of the day. All these solutions use Convolutional Neural Networks, which are specialized neural networks in Deep Learning. Those algorithms are synced with shop cameras to analyze videos in real-time. Whenever they detect a person, the event is recorded and added to the overall shop traffic data.
Another useful integration would be detection of out-of-stock-products. Panasonic offers a solution for it. Walmart has a few supermarkets that have a built-in system informing store employees about shortages of products on shelves.
Then there are also heat maps that provide insights about the interest in different products on shelves. A comprehensive data analysis of heatmaps can help understand which products are of interest to the customer, and which are not.
Video analytics could also measure how much time a customer spends in a shop. Using personal identification, the system could develop strategies for shop management to optimize customers’ shopping time. Of course, taking into account whether the customer wants fast service, or they’d rather stroll along the aisles.
On top of that, there is a whole new approach to shopping called “just-walk-out”, introduced first by Amazon Go supermarkets. These stores do not have registers or cashiers – they are entirely self-service. Customers enter a store using a palm scanner as an identification method assigned to their card, take what they need, and the purchase is executed as soon as they leave the store. The whole system is based on video analysis that tracks the customer’s shopping cart. A flavor of the experience is shown in Amazon’s promo video.
Finally, video analytics could be also used for outdoor monitoring in the retail industry. The system could detect how many potential customers are passing by the store, and thus define hot hours. This type of retail data analytics provides information on how busy the area where a store is located actually is. And based on the geographical locations of supermarkets and their workload, optimal locations for new supermarkets could be found.
Smart labelers are yet another approach to out-of-stock detection. Customers can manually inform the supermarket about the shortage of the given product on a shelf. Once the label is clicked, the staff receives information about replenishing the missing product.
In fact the labels are even smarter now. Smartphone applications enable scanning QR codes of the labels, and if they are replenished, the customer receives a message that a product is available again.
Movement and Distance Sensors
Ever had trouble picking up a cart or a shopping basket? By using smart sensors placed in appropriate locations, staff can be automatically notified about shortages, and bring more carts or baskets.
Analogically, distance sensors do the job for out-of-stock detection. They can be mounted inside a shopping shelf and detect if there are any products of a particular type left on the shelf or not.
Finally, distance sensors are sometimes used for queue management, although video analytics seems to be a more robust solution for that.
Transforming retail with IoT
The Internet of Things is truly changing retail. And IoT together with Data Science give the business all the tools to grow faster than ever. We can already observe big retail chains – such as Amazon or Walmart – adopting IoT on a large scale, and other brands will soon follow. The question is not if, but rather when the adoption of IoT will be the standard when running a retail business.
More posts by this author