Skip to Content

Artificial Intelligence: With great power comes great responsibility – Part 2

February 22, 2021

As concluded in our previous article, Artificial Intelligence is a game changer.  It is hard to find a sector or organization that would not benefit from AI. At the very least AI can be instrumental in optimizing existing business processes, saving costs and increasing revenues. But it can also be used to come up with new business models. Uber, and Deliveroo are some of the companies that have redefined their sector by using AI. We now have the means to continue on the promising AI journey, and have to follow the right approach to change the game. But what is the best way to start and to accelerate? In this article will deep dive into the factors that will enable an organization to implement and prosper from having an AI function.

Purpose, commitment & maturity

Starting point of the AI journey is to clearly understand the purpose of AI in your organization: what is the reason you want to apply AI? What is the intention with AI you abide and commit to? What is the long-term goal that is both personally and organizationally meaningful, and makes a positive mark on your customers and the world? Time and again during your journey this will both be your compass and benchmark by which you navigate your efforts.

Important in the AI journey is that it has to be continuously supported by C-level. This is not only essential during setup, but also to continue and scale the AI effort. C-level commitment ensures proper attention and funding, and it can prove to be instrumental in overcoming resistance to change and surviving the occasional road bump.

It must be clear by now that here is no successful AI without properly involving people. An important part of the organizational change is about making sure that your employees accept and trust the changes involved so they are better able to deal with this. Another part is the reskilling and upskilling of employees so that they are better able to play their part in the AI journey.

While understanding the goal, securing C-level commitment and addressing organizational change are obvious components of a successful AI journey, using AI maturity models is as important. Using a maturity model helps you understand your current position. In doing so it clarifies the challenges ahead and the next steps. To put it different: it guides your ambition. Before you can run, you need to be able to walk first. Understanding your starting position helps in finding and applying the best practices from organizations that are leading the AI journey – for your current maturity level. Best practices are not a standard recipe that you just need to follow, but it will tell you some of the essential ingredients needed for your own version of AI success.

In general maturity models distinguish five phases. When you focus on organizations that have actually started on the AI journey and disregard the last phase (Phase 5 in which organizations continuously reinvent and reorganize themselves based on AI outcomes – most would have not already achieved this), it comes down to the following phases:

  1. Orienting, learning on AI and how to start
  2. Pilots/Proof of Concepts launched, but not yet deployed in production
  3. Few use cases deployed in production, but on a limited scale
  4. Successful deployment in production and continue to scale

A survey by Capgemini Research Institute shows that during 2017-2020 the number of organizations moving beyond pilots and Proof of Concepts increased from 36% to 53%. The same study shows that sector wise the life sciences, retail, consumer products and automotive are ahead in terms of AI implementation with a range of 17 to 27%. This outperforms the global numbers where 13% of organizations have AI in production and scaling, 40% have implemented some limited AI applications and the remaining 47% have yet to leave the pilot/PoC phase.

How to move ahead

AI is a process and it needs to be handled as such. The technical part of AI is important, but by no means is it the complete picture. Take into account that crossing the chasm from AI prototype to AI at scale is a common challenge, so prepare bringing ideas to production from the first step in your AI innovation processes.

When moving ahead, one important decision is how to realize your AI capabilities: build or buy. Do you have (or make) the time to build inhouse capabilities and putting together teams of data scientist, ML Engineers and others or are you opting for a quick start by purchasing solutions in the market.  Also consider working with your partners on AI and leveraging each other’s strong points. One consideration worth making is setting up a hybrid AI ecosystem where you use external data and AI models to enrich your internal data and AI models.

Adopting a cloud first approach to AI is quickly becoming the way to go, even when requirements are still unclear. Cloud Machine Learning (CML) platforms such as AWS ML, Azure ML or Google Cloud ML (TensorFlow) can power the ML models that you are creating. There are also many AI cloud services available that you can tap into. In addition to the extensive availability of rich AI functionality it brings, cloud is the only sensible way to scale.

Infusing AI while the store stays open

Any innovation with AI should have a positive business case, either quantitative or qualitative. Driving AI by means of a business case ensures that it does not fade away as an expensive hobby or solutionism, where it’s solely about having an AI project without any concrete game changing factor. Consider applying a self-funding approach: reinvest benefits achieved by applying AI in the further innovation with AI.

When applying AI to improve existing business processes, ensure that both the data and models are treated as the enterprise assets they are. Areas to take care of are at least: clear ownership and governance, version management and automated testing and deployment. AIOps/MLOps – inspired by the successful DevOps paradigm – have proven to be a valuable approach in managing AI.

Trusted. Data.

AI is powerful, yet also surrounded by controversy. This implies that AI has to be used in a responsible and ethical manner, where explainability and fairness have to be paramount to profits and growth, and because you want to prevent unintended consequences or failures. And maybe even more importantly, you want to avoid AI rejection by important stakeholders like users, clients, citizens, employees, sponsors, shareholders and regulators. The assessment list for trustworthy artificial intelligence (ALTAI) provided by the European Union can act as a guideline for self-assessment in this area. Ethical principles have to be woven into the AI fabric of your organization. Not only to ensure the application of AI adheres to these principles, but also as a message to the outside world that you understand the concerns and powers surrounding AI and are taking it very seriously The goal here is build an ethical AI capability, supported by frameworks and tools. In many sectors and organizations Code of Ethics are put together that support and strengthen the emergence of ethical AI.

There is no intelligence without data at scale: models need data for food, lots of it. Data is both required to train your AI models and to feed them while they are used in production. This means data quality is a critical factor in the performance of your model. Yet data can bring about another risk: bias. If you train your models with biased data – this can already be the case by just using your historic data – there is a fair chance that the outcomes of the models are also biased, resulting in reasoning errors and unintended consequences. If you have lots of personal data, privacy considerations can limit the use of data and working with synthetic data or masked data can help.

AI benefits from a data centric architecture, where data has been organized in its own right and not as a byproduct of applications. In such an architecture AI can access all required data without friction, and there is a continuous flow of data to AI models like water running from a tap. Looking beyond the walls of your own data to the outside world is another must-do. It contains a huge wealth of data that can be leveraged to boost the value of your AI models, and ultimately boost the value for your stakeholders.


The journey towards successful implementation of AI is not easy, but it can be highly rewarding. Starting it off with purpose, commitment and a business case focus is essential. Having a solid and trusted AI innovation and operating model is important to continue to channel the energy in the chosen direction while continuously building on the support of both the governance and the workforce of the organization. In a subsequent article we will discuss proven AI patterns that can be leveraged to accelerate the value and scale of AI, without reinventing the wheel.

This blog has been co-authored by Erwin Vorwerk and Paul van der Linden. Please reach out to the authors for more information.