More...More...More...More...More...More...AIoT Framework

AIoT combines two of the most important technology paradigms of the 2020s: Artificial Intelligence (AI) and Internet of Things (IoT). In order to best understand AIoT from all relevant perspectives, we will start by looking at the why, what, who and how perspectives, inspired by the work of Simon Sinek [1] as well as the St. Gallen IoT Lab [2]:

  • Why: Better understand and articulate the purpose and AIoT-enabled business outcomes
  • What: What can be achieved with AIoT in terms of smart, connected products and solutions
  • Who: Roles and responsibilities in the context of an AIoT initiative
  • How: Project blueprint for AIoT execution and delivery

While Simon Sinek suggests to Start with Why, we will first look at the what to provide some context, before discussing why you should consider it.

What: Smart, connected products and solutions with AIoT

The smartness of an AIoT-enabled product or solution is usually either related to an individual physical product/asset ("product/asset intelligence"), or to a group/fleet of assets ("swarm intelligence"). Technically, asset intelligence is enabled via edge computing, while swarm intelligence is enabled via cloud computing. Asset intelligence is applying AI-algorithms to data which is locally captured and processed (via sensors), while swarm intelligence is applying AI-algorithms to data which is captured from multiple assets via IoT-technologies in the cloud.

For AIoT systems with a high level of complexity, it can make sense to apply Digital Twin concepts to create a digital representation of the physical entities. The Digital Twin concept can help with managing complexity and establishing a semantic layer on top of the more technical layers.

AIoT Intro

AI (or Machine Learning, one of the most important sub-sets of AI) makes use of different types of methods. The three main methods include supervised, unsupervised and reinforcement learning. They are explained in the AIoT 101. Different, highly specialized AI/ML methods support a wide range of use cases. AIoT is focusing on those use cases which are most relevant when dealing with physical products or assets. To mention just one example from the figure shown here, supervised learning can be used for image classification, which plays an important role in optical inspection in manufacturing. While the adoption of AI and ML has become already mainstream in some areas like social media or smart phones, for many AIoT use cases this is still not the case. There is a famous quote from James Bell at Dow Jones, which says that "Machine Learning is done in Python, AI in PowerPoint.". The goal of the AIoT Playbook is to explore and enable use cases which make use of real AI in the context of IoT, mainly utilizing supervised, unsupervised and reinforcement learning.

AIoT Use Case Patterns

An important differentiation which we are making in the AIoT Playbook is between smart connected products and smart connected solutions. Smart, connected products are often very highly standardized, feature-rich and well rounded. Smart, connected solutions on the other hand are often more custom, ad-hoc solutions. They are often designed to solve a specific problem, e.g. for a particular production site, a particular energy grid, etc. Obviously, this is not a black and white differentiation. There are also often cases that are a bit of both product and solution.

As will be discussed in more detail later, smart connected products are manufactured and sold by a Digital OEM, while smart connected solutions are usually acquired and operated by a Digital Equipment Operator.

What: Product vs. Solution

Why: Purpose and AIoT-enabled business outcomes

While AI and IoT are exciting technical enablers, anybody embarking on the AIoT journey should always start by looking at the why: What is the purpose? And what are the expected business outcomes?

From a strategic (and emotional) point of view, the purpose of the AIoT initiative should be clearly articulated: What is the belief? The mission? Why is this really done?

For the business sponsors, the expected business outcomes must be clearly defined as well. As was discussed in the what section, most AIoT initiatives are either focusing on products or solutions. Depending on the nature of your initiative, the KPIs will differ: for AIoT-enabled products, they tend to focus more on the customer acceptance and revenue side, while for AIoT-enabled solutions they tend to focus more on efficiency and optimization.

Why: Product vs Solution

How: Getting things (and AI) done

Smart connected products and solutions usually make use of AI and IoT in different ways. This must be taken into consideration when looking at the how. Smart products often rely on AI that was specifically developed for them using a Data Science approach. The goal often is to create new Intellectual Property which helps differentiate the product. For solutions, this often looks different: here the goal is to minimize development costs, e.g. by re-using existing AI algorithms and models. From the IoT point of view, products and solutions also differ: products usually have built-in connectivity capabilities (line fit), while solutions usually have this capability retrofitted. This is especially important for operators looking at heterogeneous fleets of assets or equipment.

How: Product vs. Solution

It is important to understand which capabilities are required for implementing AIoT. The AI side usually requires Data Science and AI Engineering capabilities, as well as AI/ML Ops capabilities (required for managing the AI/ML development process).

The IoT side usually requires generic cloud and edge development capabilities, as well as DevOps supporting both cloud and edge (which usually means support for OTA, or Over-the-Air-Updates of software deployed to assets in the field).

The third key element is the physical product or asset. For the Digital OEM, it will be vital to manage the combination of physical and digital features and their individual life cycles. For the physical product, this will also need to include manufacturing, as well as field support services.

AIoT Overview

Who: AIoT roles and responsibilities

The Who perspective must address the roles and responsibilities required for successfully delivering your AIoT initiative. These will partially be different for product- vs. solution-centric initiatives, as we will discuss later. It is important to have a holistic view on stakeholder management, including internal and external stakeholders.

AIoT - Who?

External stakeholders can include investors, users of the product of TDB: solution, partner, and suppliers. In a larger organization, internal stakeholders will include business sponsors, senior management, compliance and auditing, legal and tax, global procurement, central IT security, central IT operations, HR, marketing, communication, and sales. And finally, one should not forget about the stakeholders within its own organization, including developers, technology experts, AI experts, and potentially HW/manufacturing (in the case of the Digital OEM).

As indicated by the diagram following, the AIoT Playbook primarily addresses middle management, including product/solution managers, project/program managers, development/engineering managers, product/solution architects, security/safety managers, and procurement managers. Ideally, the Playbook should enable these key people to create a common vision and language that enables them to integrate all the other stakeholders.

Who: Roles and Responsibilities

References

  1. Start with Why: How great leaders inspire everyone to take action, Simon Sinek, 2010
  2. The Business Model Navigator: 55 Models That Will Revolutionise Your Business, Oliver Gassmann, Karolin Frankenberger, Michaela Csik, 2014