ESADEd3 Conference AI-driven enterprise

Page 1

The journey to the AI-driven enterprise: Machine learning automation for executives

ESADEd3 Institute for Data-Driven Decisions

www.esade.edu/d3 esaded3@esade.edu

1


The journey to the AI-driven enterprise: Machine learning automation for executives (This article has been written based on the talks by Manu Carricano, Director of the Institute for Data-Driven Decisions of ESADE, and John Boersma, Director of Educational Services of DataRobot, at the conference at ESADE on December 12, 2018)

From information systems to decision systems (Manu Carricano) We are living in an era of exponential creation of data. As a consequence, artificial intelligence (AI) and Machine learning are expanding dramatically, not only in the world of computer sciences, but in the society in general and - more specifically - in the management and operations of companies.

Data science is changing the rules of the company for strategy, leadership, skills and process improvement What if companies could run on autopilot? How far are we from automating and optimizing the decision making processes? Can machines and models make decisions already? We are living a paradigm shift where we do not have to talk about information systems anymore but about decision systems. Data science is changing the rules of the game of the company in terms of strategy, leadership, skills, process improvement and in other strategic and operational fields of management. All these changes can be summarized in five concepts that reveal the basics of the transformation: 1. Science When we try to solve business problems with models, a scientific process is applied. An algorithm is nothing more than an opinion, so no one can say that the algorithm is better than the decision-making process. The benefit of using an algorithm is the process: we ask the right question and we get rid of what is not relevant; we automate and optimize procedures as much as possible. Optimization is fundamental but automation is even more powerful. 2


2. Skepticism Evidence has to be brought to the table and we need to be able to extract the models we are using from it. Experimentation is one of the basic pillars to deliver better information, for example using A/B testing. 3. Serendipity The case where sometimes discovery comes by surprise. The more exposed to data we are, the more possibilities we have to find better processes and results in the decision-making process. 4. Strategy It is all about change management and how the transformation is measured out and managed internally (e.g. top-down or bottom-up). The transformation can be carried out in small parts (pieces of the process), repeating or replicating them in several departments to thus automate the process in small sections. 5. Scalability Taking the positive findings from experimentation to revolutionize the production chain, where we can accelerate the release of models, new cases, business solutions, etc.

There is a need to democratize and accelerate the transformation to datadriven companies Transformation should be seen as a quick access or pilot with a narrow team thinking about new cases driven by business, different case scenarios, models, architecture, infrastructure, etc. New users should be brought on board, moving from the center of excellence in analytics to a broader audience in the business functions.

3


The implementation and leadership role in Machine learning (John Boersma) Four or five years ago Machine learning was an optional issue; there were other measures to improve the efficiency of the company. This technology has now become central. Five or ten years from now, there will be a landscape of companies littered with organizations that have not figured out the importance of Machine learning and AI.

In five or ten years, in terms of implementation of AI and Machine learning, there will be winners and losers How do competitors take business away using AI? A fundamental strategy that some companies are carrying out is customizing the organization behavior down to the individual customer and transaction. A good example are insurance companies. Some of these companies are pricing according to a pool of the segments they have. Other companies are pricing to the individual level through AI instead of pooling the prices. These companies will attract the best customers and will leave the worst opportunities to those companies using pool rating.

What is Machine learning and AI? Machine learning is any technology that learns from already existing examples from the past. By using certain computer algorithms, we are able to come up with better rule systems than with using Expert systems, for example. Deep learning is a mathematical approach that shows that a computer is better than a person at recognizing specified information in a mass of data – for example, whether there is a cat in a picture. Deep learning is a very good tool for highly unstructured data like image or audio files. Nevertheless, for some types of practical business problems other Machine learning paths can work significantly better: for example, if we want to know if a customer is going to leave us. 4


One of the reasons why we call it Machine learning is because the more predictors we add to the model, the better the prediction of the model will be. In every model there is going to be an unexplained variation. Adding more variables will reduce this variation, thus giving us a better prediction. However, we should demystify the fact that sometimes, simple models work. The systems are never going to be perfect because we will never have all the predictors necessary to reduce the variations to zero.

AI is not going to make perfect decisions for us Machine learning is a learning process and the result of that process is an AI system. It is a system created to perform tasks that ordinarily require human intelligence. AI is used for three potential reasons: -

Automate processes Optimize processes Produce actionable insights

How do we look for opportunities to use AI? Our company may have the best data scientists, the best infrastructure and the best employees. Even so, in order to identify the best opportunities we need all the people who understand our business. Every manager in the organization should know what Machine learning is and how it works. We are always going to start with a business problem or an opportunity and we will have to figure out what we will predict. In order to solve the problem or foster the opportunity, several actions have to be undertaken. Since this is a broad issue, it is very important that we do not miss the opportunities, but it is important to be well aware of three things before starting: 5


-

-

-

We need to narrow the problem and define exactly what we are going to predict: Is the customer going to purchase another service from us? Or we can predict a number - what is Apple stock price going to be in the next month? We need to estimate the value of the project. Is it worthwhile for the organization? Is the cost of implementing a Machine learning solution greater than the benefits that you we going to obtain? How many departments are going to be involved or affected? We need to identify the competitive advantage in Machine learning and it basically comes from understanding data better than the competitors do.

The implementation of Machine learning cannot be just because ‘it is important’ Some examples of applications of Machine learning could be: automating the digital marketing funnel using AI, Internet advertising, optimizing online advertising, providing customers with liquidity forecasts in banking, predicting the probability of there being an outage in the system in order to avoid the cost of being without internet connection, how to price agreements by the consultant companies, reducing the high cost of raw materials in the manufacturing industry, optimizing the staff in a large hospital system given the expected patients per department or how many nurses per hour are going to be needed.

Organizational approaches to Machine learning 1. Experiment with data science. There are some managers that can get some interesting results from introducing themselves in the Machine learning and AI world. The problem they will probably face is that they do not really have organizational level sponsorship. 2. Centralized analytics and build a data science team: a. Pros: We have organizational level sponsorship so we start giving organizational level results. b. Cons: We build a staff of data scientists that maybe is too isolated of the line units they have to serve. There are many platforms that allow democratizing data science and can bring Machine learning to all the departments of the company. 3. Critical roles 1. Executive sponsor. Key responsibilities: a. Owning the prediction target. b. Thinking through implementation. c. Establishing very tight interconnections among all the people involved in the project. 2. Business champion. This probably is the most important role because the executive sponsor is not going to be in the operational implementation. The place where the competitive 6


advantage is, it is going to come from understanding the business, understanding the problem and understanding the data better than anybody else. 3. Technical staff. These are people who will get the data, prepare it, build the models, deploy it, etc.

When implementing a Machine learning project, nobody can be left behind Change can be painful. Sentences such as ‘We do not do it that way here’ or ‘I do not need AI, I am an expert’ are very common when these kind of projects are implemented. It is fundamental to do it properly using the resources needed and involving everyone, making everyone feel as part of the change.

7


Turn static files into dynamic content formats.

Create a flipbook
Issuu converts static files into: digital portfolios, online yearbooks, online catalogs, digital photo albums and more. Sign up and create your flipbook.