How to succeed in AI?

A bit of a worn-out title, but let's see if I can get a new angle to it.

I have written this article with my own little hands. The AI was on a coffee break this time.

I recently read a recent Harvard Business Review article "What Companies Succeeding with AI Do Differently" by Bruce Lawler, Vijay D'Silva and Vivek Arora. It also made me think about it through my own experience with AI (17+ years and about 100 projects). Will we finally find the final list of things where success will enable us to succeed in artificial intelligence? The Holy Grail of making everything work by fiddling?

The article reviewed a survey conducted by MIT and McKinsey (two rounds, the first in 2021 and the second in 2023 after the breakthrough of GenAI) that had interviewed more than a hundred companies about their AI utilization. The aim of the study was to identify factors that distinguish organizations that have achieved excellent results with AI from the rest of the crowd.

In the 2021 round, the wheat (top performers) stands out from the chaff (not so well performers) in five different areas:

• AI governance models
• Deployment policies
• Partnerships
• People and skills
• Data availability

The main findings of the second round of interviews conducted in 2023 were the following developments:

• The gap between wheat and chaff has grown considerably – the financial performance of AI-rival organizations had increased 3.8 times compared to chaff. It was in the order of 2.5 in 2021.
• The payback times of AI projects of organizations competing with AI have improved considerably, typically between 6 and 12 months.

One of the reasons behind the success was better AI governance models, which help organizations find the best use cases for them – sufficient business benefits and sensible challenges in implementation. Another (natural) explanation was the higher quality of the available data. The third element identified was the wider ecosystem of AI off-the-shelf software available.

What, then, according to this study, is the anatomy of success in AI? The study highlights four critical success factors:

1. Top management's commitment to AI development

In a 2023 survey, more than three-quarters of AI development executives had a C-level sponsor — most often a CEO or board of directors. Determining the return on investment (ROI) of AI initiatives is often difficult: The savings may not be immediate, and some benefits, such as freeing up employees' time for more value-added work, are indirect. It takes leadership to decide to move forward and direct resources to projects with the greatest potential, but calculating payback isn't straightforward.

2. Network of partners

Internal capabilities are essential; Nearly 90% of survey leaders in the 2023 survey said they use internal resources to develop AI solutions. But they are often not enough: Two-thirds of managers also had external partners, typically filling skills gaps and accelerating development and payback. The composition of partnerships changed significantly over the years. In the 2021 survey, universities and startups were the most common partners. Two years later, the respondents named consultants and technology houses as their most important partners. From this, one could conclude that AI has matured enough for practical approaches to be most appreciated.

The importance of information sharing was also highlighted. Both intra-industry networks and cross-industry networks were seen as important channels for gathering experiences before making large investments themselves. So, calling a friend is a good option.

3. Interdepartmental collaboration

Organizations succeeding in artificial intelligence saw smooth cooperation between IT and operations. One common approach was to create a "Center of Excellence" (AI CoE), an internal organization between silos staffed by employees with skills suitable for data science. The CoE will ensure that AI projects are implemented efficiently and deliver value, while ensuring data protection/security, data quality and compliance.

CoE activities also often create standardized processes, and this can also be used to manage the development of the necessary skills (training and recruitment). An alternative approach is to create multidisciplinary teams consisting of data scientists, engineers, architects and operations experts within business units.

As a personal observation, the leader of the AI CoE should report directly to the CEO of the organization. Not to the CIO. AI is not about IT. 

4. Data management under control

Without data, there is no artificial intelligence. To utilize AI in their operations, companies need high-quality, well-organized data. However, it is often observed that the data needed for analysis is not collected or is of such poor quality and poorly managed that it is not useful. Organizations that succeed in AI invest in systems and data management processes to ensure that this does not happen.

Was there a Holy Grail effect or were there any surprises for an experienced consultant in the field?

1. Top management's commitment to AI development
2. Network of partners
3. Interdepartmental collaboration
4. Data management under control

Not really. The study highlights important focus areas for success in artificial intelligence. Succeeding in this already creates quite good conditions for success. Personally, however, I would like to add a few other aspects to the list.

Success in AI requires an organization-wide basic understanding of AI. A lack of basic understanding leads to, for example, starting to pursue overly ambitious goals in relation to one's own level of competence – the probability of failure is high. A lack of basic understanding also leads to failure to identify use cases for AI. Not everyone needs to be an in-depth expert in artificial intelligence, but a basic understanding is worthwhile.

Developing AI solutions is an iterative process. It is not known in advance whether the available data has explanatory power in relation to the problem to be solved. It is good to have a cost-effective testing environment, i.e. a sandbox, for this experimental phase. The sandbox can be, for example, an AI PC or a trial environment built in the cloud. Or something else. The main thing is that hypotheses can be tested with a low threshold. Turning successful experiments into production solutions is a separate chapter – what architecture is suitable for us, how to influence operating methods and processes, and how to involve people?

Do AI projects somehow complete themselves? No, they don’t. Managing AI projects is an art form in its own and requires experience. An SAP project in a CV led with a waterfall model does not make anyone a good leader of AI projects. An SAP project led with the Scrum method already provides slightly better tools for leading AI projects. AI is not about IT, AI is business - business-oriented thinking, understanding of data and knowledge of the iterative project models already provide quite good prerequisites for success.

At the bottom of my list, I leave the most important one. It may or may not be included in the success factor point 1 of the study (Top management's commitment to AI development): Defining a target state and communicating with the rest of the organization – what we want to achieve with data and artificial intelligence. Once the direction is clear, the machines can be started.

Vamos!