Week 6

Management Summary MobileIron

Initial situation

During the six weeks of our international campus experience we were employed at MobileIron, who is localized in Mountain View. MobileIron plays a major role in the B2B market by providing enterprise mobility management (EMM) solutions, for small to large enterprises. Main goal of their product offering is to manage and secure mobile devices for organizations and to provide secure access for their internal resources. MobileIron is in a very competitive market with rapidly changing technologies and customer preferences. MobileIron engaged us to find out the potential of machine learning and artificial intelligence (AI) in enterprise mobile solutions. The goal of our engagement is to identify:

“How could machine learning be applied to enterprise mobility to create high-value use cases for an enterprise?”

In the past few years, the hype around “AI, machine learning or deep learning” was often mentioned in media, product and service offering presentations. Think of deep learning, machine learning and AI as a set of Russian Matryoshka dolls nested within each other; deep learning is a subset of machine learning, which is a subset of AI. There are huge expectations behind those technologies but there aren’t any proper solutions out there which are capable of acting or behaving like a real human. Other areas of machine learning are around video and image analytics which are much more sophisticated. MobileIron has the ambition to think outside the box, which challenges us to not only focus on what is possible now but rather on what it will be possible in five years.

Solution

Simplified view of MobileIron’s mobility solution is illustrated in the big picture (Figure 2). List of use cases for every aspect of their daily business has been identified and marked in the big picture. The outcome affects the following domains:

• sales and marketing

• product portfolio

• mobile devices

• customer support

Each use case has been evaluated and ranked based on their first impression and long-term value into a positioning map. There were three use cases shining out. We focused our work on deeper analysis of those three use cases. This deeper analysis contains suggested machine learning model, required data, UML diagrams, mockups and feasibility analysis.

Best practice recommendation:

MobileIron supports a high number of configuration and policy options, which creates a learning curve for new customers.The best practice recommendation helps to find industry standards and best practice setup for each customer based on defined metrics.

User Profiling:

The system analyzes the behavior of the user including positions/angle of the phone (gyroscope), location, speed and roughness of a user’s typing style. This data can help to protect the user from fraudsters as it knows the differences of the actual user’s behavior compared to a new behavior. The system can recognize, lock and alert the user when someone tries to use the phone.

Enterprise immune system:

The fundamental technology is powered by advanced, unsupervised machine learning, which is capable of learning and detecting normal and abnormal activities inside a network on an evolving basis. This allows it to detect cyber-attacks that may not have been observed before. Legacy approaches is to describe what an attack looks like and then to look for a match to that description.

Outlook

The most difficult part when working with machine learning is data analysis, usually this stage is executed by highly qualified data scientists. This concept paper provides suggestions on which machine learning model can be applied based on domain knowledge, the suggestions are built on how similar problems were solved with machine learning. To help data scientists and developers to push the use cases forward into working products in the future, UML diagrams like sequence and activity diagrams are designed. In addition, the required data for each use case is defined, and separated into already available and additionally required data for MobileIron. As a next step, the software has to be developed. Most of the codes for such models are typically predefined by frameworks.