MCON data team got the opportunity to assist to Big Data & Analytics Innovation Summit Shanghai 2017 held in Shanghai on 6th and 7th of September and organized by Innovation Enterprise. Among the highlights of the conference was Mobike showing the impact they are causing in the way citizens commute daily and how they manage the huge amount of data they collect from the trips and how they use it to increase the operation’s efficiency; Airbnb sharing about how they use data to make decisions at scale; ZF displaying its outstanding implementations with Industry 4.0 in the automotive industry and of course the very interesting introduction into deep reinforcement learning from a senior researcher in Google DeepMind.

What are AI, ML, Deep Learning, Big Data and IoT?

Artificial intelligence, machine learning, internet of things, big data, etc. are very trendy topics now days, several times are wrongly used or used interchangeably; here is a brief description of each:

Artificial Intelligence (AI), this is the broader concept that somehow involves all the other topics, we could think of it as a machine that performs tasks in a way that can be defined as intelligent and simulating human decision making processes.

It is often categorized into two groups – General and Narrow (also known as Applied); General AI would have all of the features of human intelligence: recognizing sounds and objects, understanding language, planning, learning and problem solving. Narrow AI displays some aspects of human intelligence, and can do that aspect very well, but is lacking in other areas. A machine that’s very good at identifying images, but nothing else, would be an example of narrow AI.

As it might be expected General AI is more difficult to achieve and is still a field in which researchers are primarily working; Narrow AI is a lot more common now days.

Machine Learning (ML), it is essentially a way of achieving AI; it does its finding patterns in big data and learning from it. One of the earliest definitions of ML was made by Arthur Samuel in 1959:”the ability to learn without being explicitly programmed”.

ML algorithms are often classified as supervised, semi-supervised or unsupervised. Supervised algorithms involve humans to provide feedback about the accuracy of predictions or models. Unsupervised algorithms don’t need any human involvement. They use an iterative approach to review inputs and make deductions.

The most common algorithms for supervised learning are Decision Trees, Naive Bayes Classification, Ordinary Least Squares Regression, Logistic Regression and Support Vector Machines.

Some of the applications that can be achieved by this kind of ML are:

  • Credit scoring
  • Predicting the success rates of marketing campaigns
  • Predicting the revenues of a certain product

For unsupervised algorithms the most common are Clustering, Principal Component Analysis, Singular Value Decomposition and Independent Component Analysis.

Deep Learning is one of many approaches to machine learning based on learning data representations, as opposed to task-specific algorithms; it was inspired by the structure and function of the human brain, especially the interconnecting of neurons. Deep learning algorithms are implemented in Artificial Neural Networks (ANNs) which is a computational model based on the structure and functions of biological neural networks, an ANN has nodes or “neurons” which are distributed in layers and have connections to other “neurons”. Each layer takes care of a specific feature to learn. It’s because of this layering that deep learning gets its name since the layering has a depth in contrast with other methods that use a single layer in the learning process.

Big Data, the term is quite intuitive, it refers to big amounts of data that started to be more available with the arriving of Internet and later the Internet Of Things (IoT); however it is not only about the size of the data, usually 3Vs define big data – Volume, Velocity, Variety.

Most often the data is not structured; it can be logs, images, text files, etc. Traditional data storage and processing software like RDBMS are not suited to handle the huge quantities of data. Therefore it requires different storage and computing paradigm to handle big data.

Internet of Things (IoT), is an ecosystem of interconnected computing devices, mechanical and digital machines, objects, or even animals or people that are provided with unique identifiers and the ability to transfer data over a network without requiring human-to-human or human-to-computer interaction.[1] It is of course a huge source of data that needs of Big Data paradigm and Machine Learning technics in order to leverage all of its potential.

The eye catching trends are in fields like computer vision, speech recognition, natural language processing, audio recognition, social network filtering, machine translation, etc. which produces amazing applications that appear in every newspapers’ first page as autonomous cars, cellphones that are able to unlock with face recognition, automated call centers, headsets with real time language translations etc.

Nevertheless the regular retail businesses are finding troubles in how to apply some of this flood of new applications and technologies and are feeling like left behind in the new digital race which could lead in many cases to wrongly adoptions of these technologies.

To see clearer the possibilities it is necessary to take a step behind and see the whole picture, the entry point of AI into the businesses mainstream is not going to be with state of the art applications, but it will be with sound ML algorithms that has been successfully proofed already.

The opportunity to use these knowledge and technologies is bigger than ever. Social networks and IoT have enabled access to an amount of data as never before and the increasing computational capacity of current computers makes the embracing of these technologies very feasible.

Also the availability of cloud-based IT services (AWS, Azure, Google Cloud Platform) and the so called algorithm economy, retail businesses will have easy access to Machine Learning applications, which was until now only a dream.

At MCON, we find some very good opportunities of implementing AI applications for our clients in following fields:

Credit scoring OEM Financial Services for the automotive industry has grown considerably in last years; there is a lot of historical data available which will be very useful when it comes to implement Machine Learning models that calculate credit scoring and that predict the possible outcomes of a loan.

New pricing models The big amount of available historical transactional information can be used by retailers to adjust the pricing of their different products in real time using predictive models applications.

Prediction of equipment failure After-sales services in the automotive industry is becoming more and more important in the Chinese market; with the expansion of aftermarket and the stable margins from selling new vehicles, dealers will change their major business from selling new vehicle to providing after-sales service.[2]

Every OEM and dealership has big amounts of historical aftersales transactions in their dealer management systems (DMS), this will be very useful to build Machine Learning algorithms that can predict when a part or equipment of a car will fail, and this would certainly be very useful for retailers to offer an after sales service at exactly the right time.

Churn prediction Identifying customers before they become lapsed is key for the automotive after sales business; our experience has shown that the way it has been done within OEMs is not very data driven; usually the identification is done following few metrics, perhaps only ownership age and last service record. The reason to do it that way is mainly to follow HQ standard metrics that not necessarily apply the same for every market.

We can use Machine Learning to customize the rules to the Chinese market bringing in a lot more of information like customer demographics, vehicle characteristics, aftersales behavior, etc. to be able to predict in a much customized way when a particular customer will become lapsed.

The boom of vehicle sales in the Chinese market occurred mainly after 2010. OEMs have more and more customers that finished their customer cycle. Using this ML models will allows us to tailor actions to keep customers actively participating with the dealers.

Would you like to know more about MCON’s best practice on utilizing IoT and big data? We can talk anytime!


  2. PwC Automotive Industry Bluebook (2017 Edition) China Automotive Market: Witnessing the Transformation