8 points that every AI strategy should include

GM, GM 🤙

Todays topic is not about a specific use case.

It’s about how you can create your AI-Strategy for your company.

Thanks to the rise of ChatGPT, all the hype is about AI.

In many organizations, there are increasing efforts to use artificial intelligence (AI) in initial proof of concepts (PoC) and in production operations. Often, these individual initiatives are driven by different departments.

It is also not uncommon for similar or even the same topics to be worked on across the organization without the project teams knowing about each other.

To channel this AI diffusion and raise it to a strategically important level in the organization, it is essential to develop an AI strategy. But what do we need to do to develop a comprehensive AI strategy?

Below are the essential elements that I believe should be included.

Business strategy

Before starting with an AI strategy, a clear business strategy must be in place. If this strategy is already 2-3 years old, it should be reviewed and critically examined in order to update it if necessary.

The strategy is the foundation for all business activities, including initiatives related to artificial intelligence.

Strategically important use cases

Only with an up-to-date corporate strategy can you identify and describe the strategically important use cases for the application of AI. You also need to understand how AI will contribute to your business goals, such as optimizing your processes or introducing an innovative product.

The first step is to select 3-5 use cases. More than that is usually not useful. You need to focus at the beginning of your AI journey.

To select the first use cases, you can use the following evaluation points

• Customer value

• Lighthouse character (for internal and external marketing)

• Return on investment (ROI)

Data strategy

There is no artificial intelligence without data. In this step, you should ask yourself the following questions

• What data do we need for our use cases?

• What data do we have?

• What data sources do we have?

• Do we have the right data?

• Do we have enough data?

• What data are we missing?

• How do we get the data we need?

• How do we store the data?

• Who needs access to the data?

You should visualize this information and your data flows. This will give you a complete overview of your data landscape.

An example of a visualization is a radar or matrix view.

Culture and employee skills

In addition to data, people are essential to the implementation of AI projects.

• Do you have early adopters with experience in data science or machine learning?

• Do you need new roles in your organization?

• Do you need new hires? How do you train your existing team?

• What is the buy-in from subject matter experts in each department? This is essential to ensure smooth collaboration.

• Do you need to bring in outside expertise to strengthen your team around AI or bring in additional knowledge?

A cultural change in an organization will not happen overnight with the entire workforce, as AI systems often face low acceptance among employees.

The key is to find a part of the organization that is the quickest and most efficient to start with, but also large enough to start the chain reaction across the organization.

Technology and Infrastructure

Clearly define which technologies you will and will not use in your organization. Typically, the technologies emerge from the use cases defined above. Computer vision or natural language processing, for example.

It is also important to consider the conditions and requirements under which each technology is used. In the case of computer vision, for example, there may be a requirement that recognized persons be immediately made unrecognizable so as not to violate privacy rights.

Don't reinvent the wheel! Open source frameworks and transfer learning based on open source models are essential for efficient development of machine learning solutions. If processing in the cloud is possible, hyperscalers also offer a variety of services to get you there quickly, like the OpenAI LLM on Azure. (AWS and Google have similar LLM models in place. ;-))

Compare the different services in a matrix to find the ideal solution for your use case.

For example, you can evaluate AI components based on the following aspects

• Tasks/use case

• Scalability

• On-premises vs. public cloud

• Cost

• Training with custom data

• Out-of-the-box services

There may be other issues to evaluate in your organization.

Legal and Ethical Issues and Bias

The use of artificial intelligence always raises ethical and moral questions. It is imperative that these are considered when deploying AI. Ask yourself the following questions

• How do we protect individual privacy and personal data?

• How do we ensure that the AI is not biased in its decisions and does not disadvantage anyone?

• Are there legal restrictions on our use cases?

• Do we need consent to collect data? If so, from whom and for what?

Avoid bias in the AI due to biased data or unconscious societal biases influencing the data (selection).

Implementation

Before you begin implementation, you should fully analyze the status quo in your organization or summarize and evaluate it from the previous topic blocks:

• What AI projects and initiatives are already underway?

• What data is available and in what form?

• Do we have qualified employees?

• Do we master the required technologies?

Then compare the status quo with your vision to get a clear picture of what you're missing on the way to your vision (gap analysis).

To close these gaps, you plan appropriate actions in your roadmap. The roadmap is completed with the first use cases and their rough planning.

Change Management

No matter how good an implementation plan is, if people do not understand the big picture, the plan will fail. It is therefore very important to take all employees on the journey, not just those who will be directly affected.

Proactive communication is also needed with those who are not (yet) affected. In particular, fears, concerns and reservations about moral issues and possible job losses should be explicitly considered and discussed in advance.

However, active communication must also be directed at the client, usually the management or the board of directors. Are expectations clearly communicated and realistic goals clearly described? Experience shows that top management often expects very quick results, which is rarely the case with AI initiatives. Typically, these projects take longer than expected because of the amount of foundational work that needs to be done in many organizations.

An AI system that deeply interferes with a company's processes is less a software procurement process than a reflection and learning process about one's own company.

Conclusion

I am a friend of the agile way of working, also for strategic issues. That is, you start with the strategic considerations and develop 1-2 PoCs in parallel.

This way you can prove or disprove the hypotheses of the strategy development. Furthermore, quick wins are important to win over stakeholders and minimize risks.

Only the combination of quick and scalable PoCs together with a clear overarching strategy will bring sustainable success to your AI initiatives.

I first wrote this article in 2021 for my then employer, Cloudflight.

I've adapted and updated it a bit, but the basic concepts are still true and need to be addressed in any AI strategy.

Hope you liked this other format as well.
Just hit reply and let me know!

Have a great week until next time! ☀️

Florian