What is Article 4 of the AI Act?

Article 4 of the EU AI Act, set to take effect on February 2, 2025, requires that organizations ensure their employees possess a sufficient level of AI literacy. This means that all personnel involved in the deployment or operation of AI systems must have the skills, knowledge, and understanding necessary to work effectively with AI, in accordance with the regulations. In short, this provision aims to make sure that employees are not only capable of using AI systems but are also aware of the risks and ethical considerations associated with these technologies.

According to the AI Act, “AI literacy” includes “skills, knowledge, and understanding that allow providers, deployers, and affected persons […] to make an informed deployment of AI systems, as well as to gain awareness about the opportunities and risks of AI and possible harm it can cause.” (Article 3 (56) AI Act) This definition underscores the dual responsibility of understanding both the technical and ethical dimensions of AI systems.

Why Organizations Should Care

While the Act mandates AI literacy, the enforcement mechanisms surrounding Article 4 remain somewhat ambiguous. It’s not yet clear how strict the penalties for non-compliance will be or how authorities will monitor compliance. However, failure to meet these requirements could result in legal consequences for organizations, especially those deploying high-risk AI systems, as oversight mechanisms may evolve over time.

But rather than viewing this obligation solely as a compliance burden, organizations should consider it a significant opportunity. By fostering AI literacy across their workforce, companies can not only avoid potential penalties but also accelerate AI adoption and drive innovation internally. AI literacy is essential for developing a deeper, more strategic understanding of AI’s capabilities — thus enabling businesses to harness its full potential.

Practical Steps to Achieve AI Literacy Compliance

1. Assess Existing AI Knowledge
Before diving into training, companies need to evaluate their employees’ current AI literacy levels. This can involve creating a detailed profile of each employee’s technical skills, experience, and familiarity with AI systems. One effective method could be deploying an internal survey or diagnostic tool that assesses the different dimensions of AI literacy, such as understanding of machine learning concepts, awareness of ethical implications, and experience with AI-based tools. This assessment can help identify gaps in knowledge and set benchmarks for progress. This profiling is also critical because it allows organizations to take a targeted approach rather than a generalized training, which might not address specific needs effectively. Often, organizations start by hosting an event where they introduce AI concepts, inspire employees, and allow for open discussions and feedback. Such events are excellent opportunities for employees to share their experiences and concerns, which can help kickstart AI initiatives in a positive and inclusive manner.

2. Tailored Training Programs
Once the assessment is complete, training should be customized to the specific needs of each group. For instance, employees with more technical roles might need in-depth training on data handling, AI model development, and the AI system lifecycle, while those in non-technical roles may require a broader understanding of AI’s business applications, potential biases, and ethical considerations. Training should cover both the legal aspects of the AI Act — such as understanding how AI can and cannot be used — as well as hands-on experience with the technologies they will interact with. Practical exercises, such as workshops, case studies, and scenario-based learning, can greatly enhance the effectiveness of these programs. By splitting employees into segments based on their roles and familiarity with AI, companies can ensure each person receives the right level of instruction to build confidence in working with these technologies.

3. Theoretical and Practical Education
A combination of theoretical knowledge and practical skills is crucial. Employees must understand not only the core principles of AI — such as how AI makes decisions, what constitutes an AI model, and the basics of machine learning — but also how to apply these principles in real-world scenarios. This helps employees become capable of bridging theory and practice, ultimately improving the organization’s capacity to work with AI effectively. For instance, practical training might involve sessions where employees interact with actual AI systems, understand their functionality, and troubleshoot simple issues that may arise. The goal is to create not only a technically competent workforce but also individuals who can critically assess AI outcomes, identify biases, and contribute to ethical deployment.

It is important to have different kinds of training, ranging from technical to non-technical, as well as training at the tool level. Training that goes more in the direction of prompt engineering, for example, can significantly boost productivity. Additionally, it is crucial to understand that when rolling out a new AI tool, employees should be trained on its use, much like they would be trained on how to use a new machine. Ensuring proper training for new AI tools can lead to more effective integration and utilization of these technologies in day-to-day operations.

4. Developing Internal Guidelines and Standards
Organizations should also consider developing internal codes of conduct or standards for AI usage. These internal frameworks help employees adhere to the necessary legal standards while providing a clear structure for the responsible use of AI technologies. Such guidelines could outline best practices for using AI responsibly, avoiding biases, and ensuring transparency in AI-driven decisions. For example, they might specify under what conditions employees are allowed to use AI-driven automation or what checks must be made before relying on AI-generated data. By establishing these internal standards, companies create a cohesive environment where AI is used consistently and ethically across different teams.

It is also quite common for organizations to develop guidelines that address how AI might impact employees, their jobs, or working conditions. These guidelines are often used as a commitment to AI literacy and to support the workforce rather than replace them. This approach can be a great way to communicate the organization’s AI ambition and set the pace for AI initiatives. The recommendation is to develop these guidelines together with the organization’s workers’ council, if one is in place, to ensure alignment and foster a collaborative approach.

5. Ongoing Education and Certification
Given how rapidly AI evolves, continuous education and certification programs are essential. These can help ensure that employees stay updated on the latest AI developments and remain compliant with evolving regulations. Partnering with recognized institutions to provide certifications can also validate the organization’s commitment to maintaining a high level of competence among its workforce. Certifications serve as a formal acknowledgment of an individual’s capabilities in handling AI systems and can be particularly useful for roles involving higher accountability, such as AI officers or project managers. Additionally, establishing an internal culture of continuous learning — such as encouraging participation in workshops, online courses, and professional AI forums — ensures that employees maintain their AI literacy in the face of rapid technological changes.

Even though it can make a lot of sense to partner with external experts, not all training has to be external. Quite often, organizations educate their employees by doing “roadshows” where they showcase AI use cases they have developed and built, share their learnings, and also discuss their failures. Such roadshows can be an effective way to inspire employees and foster a culture of AI literacy within the organization. A good mix of internal and external training approaches can make a significant impact.

6. Appointment of an AI Officer
For larger organizations or those heavily reliant on AI, appointing an AI officer or manager can be a valuable step. Although not a formal requirement under the AI Act, having a dedicated individual to oversee AI projects, ensure compliance, and drive AI literacy within the company can prove highly beneficial. The AI officer acts as a bridge between technical teams, management, and compliance officers, ensuring that all AI initiatives are aligned with company objectives and regulatory requirements. The role involves not just overseeing current projects but also planning for future AI needs, identifying training requirements, and establishing a roadmap for compliance. This position becomes increasingly crucial for companies that use high-risk AI systems, as it ensures there is consistent oversight and an established point of accountability within the organization. This approach is not only relevant for larger organizations; many companies start with so-called AI ambassador programs to identify potential individuals who could take on this role and gradually skill them up. Larger organizations often have multiple AI ambassadors and consolidate these competencies in an AI Center of Excellence.

7. Establishing an AI Center of Excellence
In many larger organizations, an AI Center of Excellence (CoE) serves as a central hub for coordinating AI activities, providing training, and ensuring consistency in AI practices. The AI CoE is often responsible for overseeing the appointment and training of AI ambassadors, assessing skill gaps across departments, and developing tailored programs to address these needs. Additionally, the CoE plays a key role in communicating AI strategies and best practices throughout the organization, ensuring alignment between AI initiatives and business objectives. By centralizing expertise and resources, an AI CoE helps to drive AI literacy and supports the organization in moving forward with AI in a cohesive, strategic manner.

A Strategic Opportunity for Growth

While Article 4 presents an additional regulatory layer for companies, it’s also a unique opportunity. Organizations that proactively build AI literacy will find themselves better equipped to innovate, scale their AI initiatives, and remain competitive in an increasingly AI-driven world. The ability to effectively work with AI — from understanding its potential to mitigating its risks — gives companies a strategic advantage in the market. By treating AI literacy as a strategic initiative rather than a checkbox for compliance, companies can develop a culture that not only embraces AI but also uses it to drive long-term value.

In summary, while enforcement may still be unclear, the benefits of fostering AI literacy far outweigh the risks of non-compliance. Now is the time for companies to act, not only to meet legal requirements but to position themselves at the forefront of AI innovation in the coming years.

At the AI Strategy Institute, we empower individuals and organizations to become responsible leaders, drive innovation, and create sustainable value in the Age of AI. As a unique combination of think tank, consultancy, and academy, we provide the tools, insights, and expertise needed to navigate and lead in an AI-driven world.

If you want to turn AI literacy from a compliance checkbox into a strategic advantage, check out our Academy’s AI Strategy Fundamentals program. It can help you not only meet the requirements of Article 4 but also leverage AI literacy to drive transformation and innovation within your organization.

Discover how we can support your AI journey today.

Navigating AI Literacy Under the EU AI Act: What Article 4 Means for Organizations was originally published in AI Strategy Institute on Medium, where people are continuing the conversation by highlighting and responding to this story.

There’s an ancient legend about the person who invented the game of chess, a wise man who asked for an unusual payment. When asked by the emperor what he wanted as a reward for his invention, he requested one grain of wheat for the first square of the chessboard, two grains for the second square, and so on, doubling the amount of wheat for each subsequent square. The emperor readily agreed, failing to grasp what an enormous sum would result from this exponential growth. By the time they reached the second half of the chessboard, the amount of wheat was so vast that fulfilling the request became impossible.

In technology strategy, the “second half of the chessboard” is a phrase coined by futurist Ray Kurzweil, referring to the point at which exponential growth becomes so significant that it leads to rapid, transformative changes. This may impact various aspects of our lives, such as the economy, society, or the job market. As Kurzweil notes in “The Law of Accelerating Returns,” though initial returns may be small, it is during the later phases of exponential growth that the effects become extraordinary and beyond all human common-sense models.

This concept illustrates the profound effects of exponential growth, which is especially relevant in the age of artificial intelligence (AI). This field has been making incremental progress for decades, but we are on the cusp of a new era. As we stand at or near this inflection point, understanding its implications and bracing for the rapid changes ahead is essential.

People’s ability to comprehend exponential growth is limited. In fact, the 20th-century physicist Albert Allen Bartlett once said, “The greatest shortcoming of the human race is the inability to understand the exponential function.” He observed that while the world was rapidly transitioning from linear to exponential growth, most people failed to grasp the consequences of the shift. We often envision trends as linear because most trends we encounter in daily life appear that way. Linear trends are easy to recognize, such as steady population growth or incremental price increases, and they make intuitive sense to us.

Evolution has optimized the human brain for dealing with linear functions because they have generally been far more critical to our immediate survival. However, we struggle to understand exponential trends, as there are few situations in nature where these apply. As we navigate the second half of the chessboard, recognizing and adapting to exponential growth will be crucial for constructively shaping the future.

AI’s Exponential Progress

Over the past several years, AI has experienced exponential growth in many areas, such as in the realm of language models like OpenAI’s GPT series. These models have seen rapid improvements in their capabilities, allowing them to perform tasks that were once exclusive to humans, such as generating human-like text, translating languages, and even coding. There has been similar progress in the creation of AI-generated images, a development that has upended the professional artistic community. The progression of AI is intrinsically linked to Moore’s Law, an observation stating that the number of transistors on a microchip doubles approximately every two years, leading to exponential increases in computing power. Moore’s Law has been a driving force of the technological and social change, productivity, and economic growth that are hallmarks of the late 20th and early 21st centuries.

Recent advances, provisional though they are, have already disrupted business models in creative fields once thought safe from automation. Looking ahead, we can only speculate what decades of exponential improvement in the efficiency of AI algorithms will bring to the future. Major breakthroughs in recent years include algorithms such as AlphaGo, GPT-3, and AlphaFold, which have pushed the boundaries of what seemed possible and defined the history of AI.

We are not quite in the second half of the chessboard yet, as anyone who has used an AI-driven conversational assistant like Alexa might attest. There is still no substitute for human creativity and judgment (yet). Furthermore, despite advancements in self-driving car technology, many people still rely on human drivers for transportation.

Nevertheless, when referring to AI’s exponential growth, we mean the growing range of applications and capabilities seen in the field. Recent advances in computer vision, autonomous vehicles, and machine learning algorithms capable of detecting cancer through image analysis or predicting financial markets with increasing accuracy are prime examples of this growth.

The Implications of Rapid Change

As we reach the second half of the chessboard, changes will happen faster than we can imagine, making it difficult to understand their impact on humanity. The labor market is already feeling the effects of AI, as evidenced by a recent paper by Tyna Eloundou, Sam Manning, Pamela Mishkin, and Daniel Rock (2023) titled “GPTs are GPTs: An Early Look at the Labor Market Impact Potential of Large Language Models.”

According to this paper, approximately 80% of the U.S. workforce may have at least 10% of their work tasks affected by the introduction of GPTs; 19% of workers may see at least 50% of their tasks impacted. The influence spans all wage levels, with higher-income jobs potentially facing even greater exposure. Notably, the impact is not limited to industries with higher recent productivity growth.

As new algorithms are developed, AI startups emerge, and new models are deployed, it becomes increasingly difficult to keep up with the rapid changes in the AI landscape. The truth is, AI has transitioned from research to practical applications used by companies and individuals alike.

Facing the Future with Caution

As AI continues to advance at an exponential pace, it is crucial for individuals, organizations, and governments to remain vigilant and prepare appropriately for the challenges that lie ahead. The rapid changes brought about by AI will necessitate more than the usual response to innovation. They will require a reevaluation of our societal structures and prompt us to develop entirely new strategies for education, employment, and economic growth.

Moreover, it is essential to consider the ethical implications of AI at every step, to ensure that its development and deployment are guided by principles that prioritize the well-being of all people. Any innovation with the potential to impact 80% of the workforce needs to be carefully assessed. Disruption can be planned for, if we are able to think outside the box. By recognizing the potential risks and opportunities that lie in the second half of the chessboard, we can hope to harness the power of AI for the betterment of humanity while mitigating its potential negative consequences.

Some people have suggested that AI may be the last invention that humanity will ever make. This idea stems from the belief that once AI reaches a certain level of sophistication, it will be capable of designing and inventing new technologies on its own, without human intervention. However, it is crucial to note that we are still far from achieving true artificial general intelligence (AGI), a level of AI that possesses human-like cognitive abilities and would be able to perform intellectual tasks in wide variety of domains without requiring special training. Nevertheless, a recent paper by Microsoft Researchers working claims to note first sparks of AGI in an early version of ChatGPT. This caused the Future of Life institute to issue an open letter, together with people like Elon Musk, Yoshua Benigno, Stuart Russell, and Yuval Noah Harrai, with the request to pause giant AI experiments that work on things more advanced than GPT-4 for 6 months.

The rapid developments in AI technology are closely linked to the future of humanity. Whether this impact will ultimately be positive or negative remains to be seen. The onus is on us to shape AI’s development in a way that benefits society as a whole. The future may be unclear, but by maintaining a cautious and responsible approach, and keeping an eye on that chessboard, we can strive to create a world where AI serves as a powerful tool for the betterment of all.

During my years working with AI, I’ve seen many people who misunderstand what it takes to build a successful AI startup. In particular, they often overestimate the role AI plays. As a result, they tend to focus too much on the AI component and ignore the fact that it’s just one small part of a much greater effort. Building an AI startup is a lot like trying to land a man on the moon. It’s a massive undertaking that requires meticulous planning, cutting-edge technology, and a team of experts working together to achieve a common goal. And just like the moon landing, success isn’t just about one individual effort; it’s about the entire package.

The AI — the astronaut that pilots the rocket.

I like to think of the AI part as analogous to an astronaut. Training an AI model (or models) is akin to training an astronaut for a space mission. It’s an essential component that requires a significant investment of time and resources. For the moon mission, of course, the highly-trained astronauts were critical to the mission. While the pilots of Apollo 11 rightly deserve a place in our history books, I believe that the role the astronauts played in the bigger picture of the moon landing is usually overestimated. A lot of work went into that mission behind the scenes. And just like the astronaut, the AI is only one part of the equation — a highly visible and compelling part, but only one part just the same.

What’s more, the field of AI is continually evolving. What was once groundbreaking can quickly become commonplace. In just a couple of years, today’s top-of-the-line models may well be surpassed by new, even more advanced algorithms. And just like astronauts, AI systems can become outdated and need to be retrained or replaced if they are no longer up to the job.

The product/service — the rocket that brings you to the moon (and hopefully back!)

To get an astronaut to the moon, you obviously need a rocket, and building that rocket usually takes a lot more time and effort than training the astronaut. You can think of the rocket as being your product or service. You may have the most cutting-edge AI with the best performance available, but if you are not able to build a product or service around it, that AI becomes useless. What is an astronaut without a rocket? The quality of the rocket, or the overall package, is what sets an AI startup apart from the rest.

Just like a rocket, once an AI startup has achieved a good product-market fit, the product will change very little. The people and AI systems involved, however, are likely to change over time. The ability to build strong foundations, find and train the right people, and deliver high-quality products and services is what sets successful AI startups apart from the rest. It is also what wins the race to the moon.

The team — it’s not rocket science! (Not all of it, anyway)

Think of all the people that it took to land the first man on the moon. The people who selected astronaut candidates and trained them; the rocket scientists and engineers who designed and built the rockets; the people at Mission Control who planned the flight — many people were involved. Likewise, the competitive advantage of an AI startup often doesn’t come from the AI itself. Instead, it comes from the team’s ability to build a robust foundation, create a compelling product or service, and find the best people to work with. It’s about building the rocket, finding the right pilots, and training them to become astronauts faster and better than your competitors — all while plotting a course ahead and keeping a close eye on the mission.

The value for the end user — one small step, one giant leap

Never lose sight of your goal. The moon landing was a landmark achievement for humankind and has had a profound impact on our society and our understanding of the universe. The value of the moon landing can be measured in several ways. There is, of course, its scientific, technological, and cultural significance. Beyond that, however, in the decades since the moon landing, the legacy of this achievement has continued to inspire new generations of scientists, engineers, and explorers. The moon landing has shown us what is possible when we dream big and work towards a common goal, and it has reminded us of the power of science and technology to transform our world for the better.

Ultimately, all that counts is the value you can deliver for your end users. If you are not able to build something that someone wants to purchase, you will not be able to build a successful startup, and it will be very difficult to find anyone crazy enough to invest in the project. At the end of the day, it does not matter whether you are using AI, rule-based algorithms, machine learning, fancy convoluted deep neural networks, or actual magic. If you cannot deliver value to your customers, you are doomed to fail before you get off the ground.

Building an AI startup is a complex undertaking that requires a significant investment of time, resources, and expertise. While the AI component is important, it’s essential to recognize that it’s not the only thing you need to succeed. Success requires building a strong foundation, creating compelling products and services, and finding (and training) the right people to bring it all together to deliver value to your end users. So, the next time you think about AI startups, remember that it’s not just about the AI; it’s about the entire package behind it. Plan accordingly so you can successfully reach for the stars!

Tristan Post is a founder, experienced AI expert and lecturer with a proven track record of developing strategies, partnerships, and solutions centered around AI for startups, global companies, and government institutions. He currently serves as the AI Lead at AI Founders, one of the biggest AI startup incubators in the world. He is a lecturer for AI for Innovation and Entrepreneurship at the Technische Universität München and a lecturer for AI for Business at the Munich Business School. If you like what you read please clapp and follow me for more.

Revolutionizing business operations and providing a competitive edge, Artificial Intelligence (AI) is transforming the way companies function. AI can be used for a wide range of tasks, such as automating repetitive tasks, analyzing vast swathes of data, and providing insights that would otherwise be difficult to obtain. However, many companies still struggle to apply AI successfully into their business — particularly small and medium-sized enterprises (SMEs). Startups are not an exception, especially those not widely considered “deep tech” companies. However, AI does not have to be part of your core offerings in order to make use of it in your business.

Think of the AI boom as similar to the early adoption of the computer, which had a profound impact on business. Some of the biggest companies today were among the first to use computers in the 50s and 60s. In the late 1950s, American Airlines pioneered the use of a computer reservations system, and in 1963 American installed its Semi-Automated Business Environment, allowing the company to grow its business and become one of the largest airlines in the world.

Another example is Walmart, which was one of the first retailers to adopt computers, after the CEO Sam Walton attended an IBM computer class for executives in 1968 to learn how to use this new technology in operating his business. Walmart used the computers to track sales and inventories. This helped them to improve their supply chain and increase efficiency. Walmart is now one of the biggest retailers in the world.

Whilst many SMEs and startups might lack a thorough understanding of AI, have limited resources, and a lack of in-house expertise, applying AI has never been easier. This is especially true for startups, which are by nature more agile and flexible, and therefore can more swiftly and comprehensively incorporate AI into their business practices. In the following blog post I shall use my own startup, Yoga Cat, as an example of how a business might employ AI in their operations.

Yoga Cat is a startup that is focused on developing and providing high-quality, ethical, and sustainably made yoga products. We started by developing the first deep cleaner for yoga and sports mats that we launched successfully into the market in 2022. Working closely together with many startups I often notice that many startups only focus on their products and emphasize operations only much later. However, at Yoga Cat we have always been concerned with both product development and operations — neglecting operations can lead to a variety of issues that can hinder the growth and success of our young company. In order to master operations and save resources, we rely heavily on the use of AI.

For marketing we use AI to advertise, generate new ideas, and create new content targeted towards our specific audiences. With AI, we track the performance of our marketing campaigns — we even use AI to select the right influencers. AI helps us to excel in customer success, clustering incoming and sometimes even answering incoming mails from customers. Using AI we keep track of our inventory and receive alerts when we need to restock.

AI assists us in managing our suppliers and analyzing sales. We have used AI to structure our company; AI has supported us in writing our strategy and our employee manifesto. Yoga Cat is not perfect and we probably could automate much more with AI. But applying AI is a journey and, believe it or not, we have not written one single line of code thus far in our business journey. There are no expensive data scientists, nor any AI or data engineers on our payroll (yet). We just use what is readily available in the market and there is a lot out there. So how did we do it?

Cut through the noise

One of the best ways for small and medium-sized enterprises (SMEs) and startups to adopt AI is to cut through the noise and gain a clear understanding of what AI can do for their business. While AI is still a buzzword, encompassing many different technologies, it is important to remember that it is not rocket science. There are countless resources available to help you gain a better understanding.

One key piece of advice is to focus on understanding how AI can be applied to your business and what it can do, rather than trying to become an expert in programming or technical aspects of AI. There is no need to become a programmer or data scientist in order to implement AI, as there are many tools and resources available that make it easy to get started.

In my case this was rather easy, as I actually am an expert in AI. I have worked with international companies on their AI strategy and have assisted in building the competences to apply AI in their organization. Also, I am a lecturer for the application of AI at the Technical University of Munich, as well as the Munich Business School. Furthermore, I manage one of the largest startup incubators for AI in Europe. Trust me when I tell you that applying AI is not rocket science — you do not need years of formal training or the ability to write or read code. However, some basic understanding does not hurt. Here are some free resources and online courses worth reading:

• Introduction to AI and Entrepreneurship | AI Campus (ki-campus.org)
• A free online introduction to artificial intelligence for non-experts (elementsofai.com)

Think about the role AI might play for your company

AI can be used as a core technology, meaning it is the main product or service that your company offers. An example of this is an AI-powered chatbot for customer service. In this case, the AI technology is the main offering and the primary value proposition for the company. On the other hand, AI can also be used as an enabling technology, which means it is used to support and enhance existing products or services and/or operations. An example of this would be using AI to optimize supply chain management or to analyze customer data to improve targeted marketing efforts.

This distinction — whether AI is a core or enabling technology for your company — is of vital importance because this will determine how much resources and focus should be allocated towards developing and implementing AI solutions. It will also impact the overall strategy and direction of the company. For example, a company that is using AI as a core technology will likely prioritize investment in research and development and may have a more technical team and focus. A company that is using AI as an enabling technology may prioritize partnerships and integrations with existing technology platforms and tools.

As previously mentioned, Yoga Cat is not a deep tech startup. It is also very unlikely, for now, that we will build products that use AI. Yoga Cat, therefore, uses AI as an enabling technology. Namely, we mainly use AI to improve our existing processes. As a startup, it is important to make the most of our resources and use them efficiently.

Define a vision for how AI can be applied at your company

When looking to adopt AI you need to have a clear vision for how it will be implemented within the organization. This vision should be based on a thorough understanding of the current status quo and capabilities within the company, as well as an understanding of how competitors are applying AI and how the industry may be impacted by AI. Additionally, involving key stakeholders such as customers, employees, and partners to gain full awareness of their expectations will help to ensure the implementation of AI aligns with the overall goals and objectives of the company.

At Yoga Cat, we understand that as a company that does not have any technical products, there may be limited opportunities to incorporate AI directly into our products. However, as we operate mainly in the e-commerce business, we have recognized the potential for AI to be applied in our operations and marketing efforts.

In the e-commerce industry, customers expect quick response times and fast delivery — this is where AI can be of great benefit.

By using data and AI, we are able to create data-driven insights and strive towards operational excellence. We are able to analyze large amounts of data, identify patterns and trends, and use this information to make informed and data-driven decisions. This helps us to optimize our processes, increase efficiency and ultimately, deliver a better experience for our customers.

Furthermore, we use AI to automate repetitive tasks, improve our inventory management, and personalize our marketing efforts to better target our audience. We also employ AI to analyze and improve our customer service, so that our customers have a better shopping experience and are more satisfied. All of these efforts are aimed at saving time and resources, thereby allowing us to focus on our core competency, which is creating high-quality, sustainable yoga products.

Understand the enabling factors

In order to successfully implement AI in any organization, there are several key enabling factors that must be considered. These include:

• Expertise: Having a team with the necessary expertise to understand AI and how to apply it is crucial for the success of any AI initiative. This includes having individuals with knowledge of both the technical and business aspects of AI.

At Yoga Cat, we recognize the importance of expertise when it comes to understanding and applying AI. We have a tech-savvy team that possesses an excellent understanding of how to integrate new AI-enabled tools into our processes. Additionally, we prioritize data literacy when hiring new talent. This allows us to stay up-to-date with the latest developments in the field of AI and make informed decisions about which tools and solutions will be the most beneficial for our business. Furthermore, we also invest in training and development for our team, thereby continually improving our expertise in AI.

• Organization: It is important to clearly define who within the organization is responsible for AI and how these individuals are organized within the company. This will ensure that there is a clear structure and chain of command for decision making and the implementation of AI initiatives.

At Yoga Cat, we understand the importance of having a clear organization when it comes to implementing AI initiatives. As a small startup, we currently do not have a large team, which means that organizational setup to drive AI is less relevant. However, we are aware that as we grow, we may need to re-evaluate our organizational structure to ensure a clear chain of command for decision making and implementation of AI initiatives. For now, I, as the founder, am responsible for driving our AI strategy with the support of the whole team, ensuring that everyone is aligned with the AI vision and strategy.

• Culture: Having a culture that is open to embracing AI is essential for the successful implementation of AI. This includes encouraging experimentation, learning from failures and promoting a culture of continuous improvement.

At Yoga Cat, we encourage a culture based on experimentation and embracing failure. We understand the application of AI as a journey and we are constantly trying to incorporate AI into more and more of our processes. We believe that by encouraging experimentation and learning from failures, we can foster a culture of continuous improvement, which is essential for the successful implementation of AI. Additionally, we also promote a culture of open communication, where ideas and feedback are welcomed. This helps to ensure that everyone is on board with the AI vision and strategy and that we, as a whole, are moving in the right direction.

• Technology infrastructure: A strong technology infrastructure is necessary to support the processing and storage of large amounts of data, as well as the running of AI models.

At Yoga Cat, we do not have a dedicated technology infrastructure for AI development (for now) as we are primarily focused on buying pre-existing AI-enabled tools and apps that can be integrated into our existing systems. However, we do have a robust technology infrastructure that is necessary to support the processing and storage of large amounts of data, as well as the running of AI models. We utilize multiple platforms and tools that allow us to gather data and easily incorporate third-party tools and AI models. This ensures that we can use the most advanced technology available and stay competitive in the market.

• Data: Having high-quality data is crucial for AI to function properly. This includes having accurate, relevant and up-to-date data.

At Yoga Cat, the importance of data is well understood. We collect data at each of our processes to ensure that our decisions are data-driven. We place a high emphasis on understanding our data and using the right key performance indicators (KPIs). This enables us to make informed decisions and optimize our processes. To achieve this, we have various dashboards that we track on a regular basis to monitor our performance, identify opportunities for improvement, and make data-driven decisions. Additionally, we also ensure that our data is accurate, relevant, and up-to-date. This is crucial for the proper functioning of AI and ensuring that the results we get from our AI models are reliable.

• Ecosystem: It is important to stay connected to the AI ecosystem, and to be aware of the latest developments and solutions that are coming out on a daily basis. This will allow companies to stay ahead of the curve and take advantage of new technologies as they become available.

At Yoga Cat, we understand the importance of staying connected to the AI ecosystem and keeping up with the latest developments and solutions. We strive to be an active part of the AI and e-commerce community. We regularly attend events related to e-commerce and AI, where we network with other professionals, exchange tips, best practices, and recommendations on which tools to use. This allows us to stay ahead of the curve and take advantage of new technologies as they become available. Additionally, we also engage with other startups and SMEs, to understand their AI journey and learn from their experience. This allows us to be better prepared for the future and ensure that we are not missing any opportunities in the field of AI.

Own your processes

It is vitally important for companies to understand their processes in order to identify areas where AI can be applied to improve efficiency and increase competitiveness. One way to do this is by creating process maps to visualize the steps and tasks that make up one’s business. This can help identify areas where there is a lot of potential for AI to be applied, as well as where there is repetitive work or other pain points.

Additionally, companies should also make sure to collect data at every point along the process map. This data can be used to gain better insights into their processes and inform data-driven decision making. It can also be used to train AI models and optimize processes. By understanding their processes and collecting data, companies will be able to identify where AI can be applied to improve efficiency and increase competitiveness.

At Yoga Cat, we realize the importance of understanding our processes in order to identify areas where AI can be applied to improve efficiency and increase competitiveness. We actively create process maps to visualize the steps and tasks that make up our business. This helps us identify areas where there is potential for AI to be applied, as well as where there is repetitive work or other pain points.

Additionally, we also make sure to collect data at every point along the process map, which is used to gain better insights into our processes and inform data-driven decision making. This data is also used to train AI models and optimize processes. By understanding our processes and collecting data, we are able to identify where AI can be applied to improve efficiency and increase competitiveness. Once we identify the opportunities, we actively try to source the right tools that can best help us to achieve our goals and improve our processes.

Make or Buy

When it comes to applying AI to a business, one important decision to make is whether to build the tools in-house or to purchase them from a third party provider. This decision will depend on the resources and capabilities of the company. For some companies, building AI tools in-house may be the best option, while for others, buying pre-existing tools may be more cost-effective and efficient.

At Yoga Cat, as an e-commerce company, we decided to buy AI-enabled tools and apps that can be integrated into our existing systems. This decision was made because we do not have the capabilities or resources to build AI tools in-house — such a task would also be prohibitively expensive in the early stages of our journey. Moreover, the e-commerce market offers a vast array of different tools and apps that leverage AI and can be integrated in our shop.

However, with so many different options available, it can be difficult to keep track of which tools work well and which may not be suitable. This is why Yoga Cat adopts an open approach to trying new things out and experimenting. We also rely heavily on our community and ecosystem to share which tools work well and which might not be suitable.

In summary, whether to build or buy AI-enabled tools is an important decision that companies must make when implementing AI. Factors to consider include the company’s resources and capabilities, the cost-effectiveness of the solution, and the availability of tools in the market.

The adoption of AI in small and medium-sized enterprises (SMEs) and startups is not rocket science. Whilst this process takes some thought and planning, it can deliver great benefits, especially for early adopters. As with the adoption of computers and the internet in past decades, companies that adopt AI early on can gain a significant competitive advantage.

However, a basic understanding of AI and its applications is vital. This knowledge can be used to formulate a vision that serves as a “north star” and is aligned with the overall strategy of the company. Additionally, understanding the enabling factors to apply AI is crucial for successful implementation.

Understanding processes and identifying areas where AI can be used to improve efficiency is an important step towards successful implementation. By understanding these key concepts and being able to execute on the strategy, companies will be able to leverage the power of AI and improve their operations, thereby ensuring business success and a competitive edge.

This is Part 3 of my series of blog posts on how to invest in AI. In my job at building and leading one of the biggest AI Startup Incubators in Europe, working closely with various founders, and teaching a course on AI for innovation and entrepreneurship at the Technical University of Munich, I screen a lot of pitch decks on AI and am privy to a great deal of AI startups as they come and go. Frequently I am asked by investors and VCs to examine AI startups and help them to make their investment decisions. In this series of posts, I wish to share my investment hypotheses and explain what I look for in evaluating and investing into AI startups.

How to Invest in AI — Part 1
How to Invest in AI — Part 2

Step 5 — Do the Due Diligence

There is no way of avoiding due diligence when it comes to investing in AI startups. It is important to speak with the founding team in order to gain a more comprehensive understanding of the startup’s business model, technology, and growth prospects. Such a discussion can provide valuable insights into the startup’s vision, strategy, and execution capabilities, as well as its approach to risk management and compliance. This can also help investors to identify potential red flags or areas of concern that may impact the long-term viability of the startup. When I speak with teams, and when I have more time than just 30 minutes, I try structure my questions along the machine learning life cycle.

Scoping

Product-Market fit

To overcome the challenge of finding the right product market fit, an AI startup may need to take the following steps:

• Define the target market: Identify the specific group of customers the startup’s products or services are intended to serve and understand their needs and preferences.
• Research the competition: Analyse the competitive landscape and understand how the startup’s products or services compare to those of its competitors.
• Test and validate the product or service: Use market research, customer feedback, and pilot tests to validate the value and viability of the startup’s product or service.
• Adjust the product or service as needed: Based on the results of testing and validation, make any necessary adjustments to the product or service to improve its fit with the target market.

During the due diligence process, investors may want to ask the founding team the following questions to assess the startup’s ability to find the right product market fit:

1. Who is the target market for the startup’s products or services, and how is the startup identifying and targeting this market?
2. What is the competitive landscape, and how does the startup’s product or service differentiate itself from its competitors?
3. How has the startup validated the value and viability of its product or service through market research and customer feedback?
4. What steps has the startup taken to adjust its product or service to better meet the needs of its target market?

By answering these questions, the founding team can provide insight into their approach to finding the right product market fit and their ability to adapt and evolve their products or services as needed.

Scalability

To overcome the challenge of scaling, an AI startup may need to take the following steps:

• Define the growth strategy: Identify the specific goals and milestones that the startup aims to achieve as it scales and develop a plan to achieve them.
• Build a scalable business model: Identify the key drivers of growth and develop a business model that is scalable and can support that growth.
• Secure the necessary resources: Identify the resources (e.g., funding, talent, infrastructure) that the startup will need to scale and develop a plan to secure them.
• Invest in automation and optimization: Implement automation and optimization technologies and processes to improve efficiency and reduce the workload as the startup scales.

During the due diligence process, investors may want to ask the founding team the following questions to assess the startup’s ability to scale:

1. What is the startup’s growth strategy, and how does it plan to achieve its scaling goals?
2. What is the startup’s business model, and how is it designed to support growth?
3. What resources will the startup need to scale, and how does it plan to secure them?
4. How is the startup addressing the challenge of scaling with heterogeneous data, and how does it plan to handle and make use of this type of data as it grows?

By answering these questions, the founding team can provide insight into their approach to scaling and their ability to secure the necessary resources and implement automation and optimization technologies as needed.

Long lasting USP

To overcome the challenge of creating a long-lasting unique selling point (USP), an AI startup may need to take the following steps:

• Identify the target market: Define the specific group of customers that the startup’s products or services are intended to serve and understand their needs and preferences.
• Analyse the competition: Research the competitive landscape and understand how the startup’s products or services compare to those of its competitors.
• Develop a unique value proposition: Identify a unique feature or benefit that sets the startup’s products or services apart from its competitors and makes them attractive to customers.
• Validate the USP: Use market research, customer feedback, and pilot tests to validate the value and viability of the startup’s USP.
• Protect the USP: Implement strategies to protect the startup’s intellectual property and ensure that its unique value proposition is not easily copied or replicated by competitors.

During the due diligence process, investors may want to ask the founding team the following questions to assess the startup’s ability to create a long-lasting USP:

1. What is the target market for the startup’s products or services, and how is the startup identifying and targeting this market?
2. What is the competitive landscape, and how does the startup’s product or service differentiate itself from its competitors?
3. How has the startup validated the value and viability of its unique value proposition through market research and customer feedback?
4. How is the startup using data to create a long-lasting USP? For example, is the startup using data to deliver personalised or relevant experiences to customers, or is it using data to identify unique patterns or trends that can inform its product or service offerings?

By answering these questions, the founding team can provide insight into their approach to creating a long-lasting USP and the role that data plays in this process.

Data Engineering

Data

When performing due diligence on an AI startup, it is important to understand what kind of data goes into a model and the nature of the output. This can impact the value proposition, scalability, and long-term viability of the startup. Understanding the data inputs and outputs of an AI model can help investors assess the potential risks and opportunities associated with the investment, as well as the reliability and accuracy of the model.

It is also important to make sure that the training data used to build an AI model is the same as the data used in production as this can impact the performance and reliability of the model. If the data used in production differs significantly from the training data, the model may not perform as expected or may produce inaccurate or biased results.

To find out what kind of data goes into a model and what is the output, as well as to assess the quality and consistency of the training and production data, an investor could ask the following questions:

1. What kind of data does the startup use to train its AI models, and how does it gather and pre-process this data?
2. What is the output of the startup’s AI models, and how is this output used by the customer or end user?
3. How does the startup ensure the quality and relevance of the data used to train its AI models, and what measures is it taking to prevent bias or errors in the data?
4. How does the startup ensure that the training data used to build its AI models is the same as the data used in production, and what measures is it taking to prevent data drift or other discrepancies between the two datasets?

By answering these questions, the founding team can provide insight into the data inputs and outputs of their AI models, as well as their approach to data quality and consistency. This information can be valuable for investors as they consider whether to invest in the startup.

Data Centric AI

One way that AI startups can create a long-lasting unique selling point (USP) is by focusing on data-centric AI. Data-centric AI refers to AI solutions that are built around the collection, analysis, and use of data to improve performance and drive business value. In the case of AI startups, this can involve developing proprietary data sets, building expertise in data management and analysis, or creating AI solutions that rely on data to drive their value proposition.

Focusing on data-centric AI can help AI startups to create a long-lasting USP because:

• Data can be a competitive advantage: Access to large amounts of high-quality data is often essential for training and improving AI models. By developing proprietary data sets or building expertise in data management and analysis, AI startups can differentiate themselves from their competitors and create a unique selling point.
• Data-driven solutions can be more valuable: AI solutions that are built around data and rely on data to drive their value proposition can be more valuable to customers because they are able to deliver more relevant and personalised experiences. This can help AI startups to create a long-lasting USP.
• Data-centric AI can be more resilient to technological change: As mentioned above, the field of AI is rapidly evolving, and what may be considered innovative or unique today may quickly become commonplace or obsolete in the future. By focusing on data-centric AI, startups can create solutions that are less reliant on specific technologies or approaches and are more resilient to technological change.

Creating a long-lasting USP is a critical challenge for AI startups, as this can impact their ability to attract and retain customers, generate revenue, and achieve profitability. By focusing on data-centric AI, startups can create unique and valuable solutions that are able to adapt and evolve over time.

To better understand if an AI startup is using data-centric AI to create a long-lasting unique selling point (USP), an investor could ask the following questions during the due diligence process:

1. What is the source, type, and quality of the data that the startup is using, and how does it plan to acquire and manage this data over time?
2. How is the startup using data to create a unique value proposition for its products or services? For example, is the startup using data to deliver personalised or relevant experiences to customers, or is it using data to identify unique patterns or trends that can inform its product or service offerings?
3. How does the startup’s use of data-centric AI differentiate it from its competitors, and how does it plan to protect this differentiation over time?
4. How has the startup validated the value and viability of its data-centric AI through market research and customer feedback, and what steps has it taken to adjust its approach as needed?

By answering these questions, the founding team can provide insight into their use of data-centric AI to create a long-lasting USP and their ability to protect and maintain their unique value proposition over time.

Virtuous Cycle of AI

When doing due diligence on an AI startup, it is important to understand how the startup can make use of the virtuous cycle of AI to create value and growth because this can impact the value proposition, scalability, and long-term viability of the startup. The virtuous cycle of AI refers to the feedback loop that occurs when an AI system is used to improve the data that it is trained on, which in turn improves the performance of the system, which in turn leads to further improvements in the data. This cycle can be beneficial because it can help to drive continuous improvement and value creation in an AI system.

To find out how an AI startup can make use of the virtuous cycle of AI to create value, an investor could ask the following questions:

1. How does the startup’s AI solution make use of the virtuous cycle of AI to drive continuous improvement and value creation?
2. How does the startup plan to acquire and manage data to support the virtuous cycle of AI, and what measures is it taking to ensure the quality and relevance of this data?
3. What benefits does the startup expect to derive from the virtuous cycle of AI, and how does it plan to measure and track the impact of this approach?
4. How does the startup’s use of the virtuous cycle of AI differentiate it from its competitors, and how does it plan to protect this differentiation over time?

How is Data Used to Create Future Value

It can be important to understand how an AI startup uses data to create new use cases in the future because this can impact the value proposition, scalability, and long-term viability of the startup. Identifying new use cases for an AI solution can help to drive growth and value creation for the startup. Understanding how the startup plans to do this can help investors assess the potential risks and opportunities associated with the investment.

To find out how an AI startup uses data to create new use cases in the future, an investor could ask the following questions:

1. How does the startup use data to identify new use cases for its AI solution, and what methods does it use to gather and analyse data for this purpose?
2. What steps does the startup take to validate and prioritise new use cases identified through data analysis, and how does it plan to bring these use cases to market?
3. How does the startup plan to expand its customer base or target market through the development of new use cases, and how does this align with its overall growth strategy?
4. How does the startup’s use of data to create new use cases differentiate it from its competitors, and how does it plan to protect this differentiation over time?

By answering these questions, the founding team can provide insight into their approach to using data to create new use cases and the benefits and risks associated with this approach.

Modelling

Model Performance

It is important to understand the performance of AI models when doing due diligence on an AI startup because this can impact the value proposition, scalability, and long-term viability of the startup. Understanding the performance of AI models can help investors assess the potential risks and opportunities associated with the investment.

To better understand the performance of AI models, an investor could ask the following questions during the due diligence process:

1. What metrics does the startup use to measure the performance of its AI models, and how does it track and monitor these metrics over time?
2. How does the startup ensure the accuracy and reliability of its AI models, and what measures is it taking to prevent bias or errors in the model outputs?
3. How does the startup plan to handle data drift, which can occur when the data used to train an AI model differs significantly from the data used to deploy the model in production?
4. What steps does the startup take to continually improve the performance of its AI models, and how does it plan to adapt to changing data patterns or requirements over time?

By answering these questions, the founding team can provide insight into their approach to measuring and improving the performance of their AI models, as well as their ability to handle data drift and adapt to changing data patterns or requirements.

Production & Maintenance

When doing due diligence on an AI startup, it is important to understand how AI is embedded in their solution because this can impact the value proposition, scalability, and long-term viability of the startup. AI can be used in a variety of ways in a startup’s solution. Understanding how AI is being used can help investors assess the potential risks and opportunities associated with the investment.

To find out how AI is embedded in an AI startup’s solution, an investor could ask the following question:

1. How is AI being used in the startup’s solution, and what specific problems or challenges is it addressing?
2. How is AI embedded in the startups infrastructure?

By asking this question, the investor can gain a better understanding of the role that AI plays in the startup’s solution and how it is being used to solve specific problems or challenges. This information can be valuable for assessing the potential value proposition and scalability of the solution, as well as the expertise and capabilities of the founding team. It can also help the investor to identify potential risks or limitations associated with the use of AI in the solution.

Risks

When performing due diligence on an AI startup, it is important to understand the risks of using AI and how the startup might be affected by regulation such as the AI Act. Such risks and regulations may impact the long-term viability and growth potential of the startup. Risks associated with using AI may include technical challenges, data privacy and security concerns, ethical and societal issues, and regulatory compliance. Understanding how these risks are being addressed by the startup can help investors assess the potential risks and opportunities associated with the investment.

To better understand the risks of using AI and how the startup is addressing them, an investor should ask the following questions:

1. What are the specific risks associated with using AI in the startup’s solution, and how is the startup addressing these risks?
2. How is the startup complying with relevant regulations and standards related to AI, such as the AI Act, and what impact might these regulations have on the startup’s operations and growth potential?
3. How is the startup managing data privacy and security risks associated with using AI, and what measures is it taking to protect customer data and ensure compliance with relevant laws and regulations?
4. What ethical and societal issues may arise in the startup’s use of AI, and how is the startup addressing these issues?

By answering these questions, the founding team can provide insight into their understanding of the risks of using AI and their approach to managing and mitigating these risks.

Human-in-the-loop (HITL)

When doing due diligence on an AI startup, it is important to understand if the startup is using a “human-in-the-loop” approach to create value and mitigate risks as this can impact the value proposition, scalability, and long-term viability of the startup. A human-in-the-loop approach involves the integration of human expertise and oversight into the development and deployment of AI solutions. This can be beneficial because it can help to improve the accuracy, reliability, and transparency of AI systems, as well as address ethical and societal concerns.

To find out if an AI startup is using a human-in-the-loop approach, an investor could ask the following questions:

1. How is the startup integrating human expertise and oversight into the development and deployment of its AI solutions?
2. What benefits does the startup expect to derive from using a human in the loop approach, and how does it plan to measure and track the impact of this approach?
3. How does the startup plan to address ethical and societal concerns related to the use of AI, and what measures is it taking to ensure the transparency and accountability of its AI systems?
4. How does the startup’s use of a human-in-the-loop approach differentiate it from its competitors, and how does it plan to protect this differentiation over time?

By answering these questions, the founding team can provide insight into their use of a human-in-the-loop approach and the benefits and risks associated with this approach.

Look out for Potential Red Flags along the Due Diligence

There are several potential red flags that investors should be aware of when it comes to investing in AI startups. Some of these red flags include:

• Lack of transparency: If the startup is not forthcoming about how it is using AI or is unable to clearly explain the details of its AI solution.
• Lack of expertise: If the founding team lacks relevant AI expertise or has limited experience in the industry.
• Inaccurate or unrealistic claims: If the startup makes overly optimistic or unrealistic claims about the capabilities of its AI solution.
• Unclear business model: If the startup does not have a clear plan for how it will monetize its AI solution or generate revenue.
• Poor data management practices: If the startup has poor data management practices or is not taking steps to protect customer data.

To find red flags when investing in an AI startup, an investor should ask the following questions:

1. Can the startup clearly explain the details of its AI solution and how it is using AI to solve specific problems or challenges?
2. Does the founding team have relevant AI expertise and experience in the industry, and are they able to demonstrate their capabilities and track record?
3. Are the claims made by the startup about the capabilities of its AI solution accurate and realistic, and can they be supported by data or evidence?
4. Does the startup have a clear plan for monetizing its AI solution or generating revenue, and how does this plan align with its target market and competitive landscape?
5. How is the startup managing and protecting customer data, and what measures is it taking to ensure compliance with relevant laws and regulations related to data privacy and security?

By answering these questions, the founding team can provide insight into their capabilities, business model, and data management practices, which can help investors identify potential red flags and assess the risks and opportunities associated with the investment.

Step 6 — Using outside expertise

It is important to utilise outside expertise when performing due diligence on AI startups, for several reasons. For example, if the startup is using cutting-edge AI technology or AI hardware, it would be prudent to bring in outside experts to navigate the potential risks and opportunities associated with such technologies. This is because these technologies may be complex or unfamiliar to the investor; outside experts can provide valuable insights and perspective on their potential impact and adoption. Such experts may also help to identify potential risks or opportunities related to these developments.

There are several sources of outside expertise that investors may tap into when performing due diligence on AI startups. These sources include:

1. Industry experts: Industry experts can provide valuable insights into the specific AI technologies or hardware being used by the startup, as well as the broader market trends and challenges that may impact the startup’s growth potential.
2. AI researchers: AI researchers can provide valuable insights into the technical capabilities and limitations of the AI technologies or hardware being used by the startup, as well as the potential risks and opportunities associated with these technologies.
3. Legal experts: Legal experts can provide valuable guidance on the regulatory landscape surrounding AI technologies and how such contexts may impact the startup’s operations and growth potential.
4. Consultants: Consultants with expertise in AI or related fields can provide valuable guidance on the strategic, technical, and operational aspects of the startup’s AI solution.

Disclaimer: I have written this article using my own experience and know-how regarding how to approach the evaluation and due diligence of an AI startup. To accelerate the writing I used the help of ChatGPT. However, I manually added examples, interviews, images, and references; I tweaked the inputs and outputs, rewrote, and added paragraphs. Using ChatGPT allowed me to write this series of posts in a couple of days — instead of weeks.

This is part 2 of my series of blog posts on how to invest in AI. In my job at building and leading one of the biggest AI Startup Incubators in Europe, working closely with various founders, and teaching a course on AI for innovation and entrepreneurship at the Technical University of Munich, I screen a lot of pitch decks on AI and am privy to a great deal of AI startups as they come and go. Frequently I am asked by investors and VCs to examine AI startups and help them to make their investment decisions. In this series of posts, I wish to share my investment hypotheses and explain what I look for in evaluating and investing into AI startups.

How to Invest in AI — Part 1
How to Invest in AI — Part 3 (Due Diligence)

Step 2 — Figure out if the startup is using AI as a core technology or as an enabling technology.

Knowing whether a startup is using AI as a core or enabling technology is important as this can impact the potential value of the company and the level of risk involved. If AI is a core technology, the startup may be more reliant on the success of its AI products or services. However, if AI is an enabling technology, the startup may have more diversified sources of revenue. It is important to carefully evaluate the role of AI in the startup’s business model and consider the potential risks and rewards before making an investment decision.

To determine if a startup is using AI as a core technology or an enabling technology, you should consider the following:

1. Product or service: Is the startup’s product or service primarily focused on AI, or is AI being used to support or enhance an existing product or service? If the AI technology is central to the product or service, it is likely a core technology. If it is being used to support or augment an existing product or service, it is likely an enabling technology.
2. Market positioning: How is the startup positioning itself in the market? Is it marketing itself as an AI company, or is it highlighting the use of AI in its products or services?
3. Team expertise: Does the startup have a team with strong expertise in AI, or is the team primarily focused on a different area, such as product development or marketing, and using AI as an enabling technology?
4. Use of data: Does the startup have a large amount of high-quality data, which is essential for training and improving AI models? If so, this indicates that AI is a core technology for the startup.

Step 3 — Try to classify the AI startup and understand how this affects the value proposition.

The value proposition of an AI startup will depend on the specific products or services it offers and the market it is targeting. For example, an AI startup that offers a unique, differentiated AI software product may have a strong value proposition, while an AI startup offering a commodity AI service may have a weaker value proposition. It’s important to carefully evaluate the value proposition of an AI startup when considering an investment.

There are many different types of AI startups. The value proposition of each will depend on the specific products or services they offer and the market they are targeting. Some common types of AI startups include:

1. AI enabling software: These startups develop and sell AI software products, such as machine learning libraries, natural language processing tools, data labelling tools, and no-code solutions. One example is the company Hugging Face which develops tools for building applications using machine learning. It is most notable for its library built for natural language processing applications.
2. AI-powered products or services: These startups offer products or services that use AI to solve specific problems or meet customer needs. Examples include AI-powered customer service chatbots, predictive maintenance tools, and personalised recommendation engines. Examples include self-driving cars, autonomous drones, or the startup Gong which is dedicated to solving the typical problems in the sales and customer area with the help of AI. To help companies improve their customer success and sales performance, Gong tracks every customer interaction, such as sales calls and emails, and evaluates important factors such as conversation topics, and the duration or frequency of customer conversations, using natural language processing and machine learning.
3. AI-enabled products or services: These startups use AI to enhance or support existing products or services, but AI is not the primary focus of the business. Ant Financial uses AI to deliver financial services at almost zero (human) operational costs. For instance, they have automated their loan application process to the point where they claim to have no human involvement when it comes to granting loans. Another example is ByteDance’s TikTok, which provides users around the world with short video clips recommended by an AI based on their preferences, viewing habits and location.
4. AI hardware: These startups develop hardware specifically designed for AI applications, such as specialised chips or sensors. One example is the German startup Konux, which combines machine learning and Industrial IoT to deliver Software-as-a-Service (SaaS) solutions for operation, monitoring, and maintenance process automation. Their sensors are used on railway switches to increase the availability of the track network and thus improve the punctuality of trains.
5. AI consulting: These startups offer consulting services to help businesses integrate AI into their operations or develop custom AI solutions. Usually AI consulting companies such as appliedAI or Merantix would not be the classical VC case as they are (human) resource intensive and therefore do not scale well.

It is important to note that it is not always possible to make a clear distinction between the types of startups. Konux’s IOT devices, for example, comprise both AI hardware as well as an AI-powered product.

Step 4 — Understand the most common challenges of AI startups.

Understanding the most common challenges of AI startups can help if you want to make a quick initial assessment. Oliver Schoppe, early stage investor at UVC Partners, talks in an article about the most common pitfalls of AI startups.

Product Market Fit

Ensuring product market fit is a critical challenge for AI startups, as this can impact the startup’s ability to attract customers, generate revenue, and ultimately achieve profitability. It’s important for AI startups to carefully assess their target market and ensure that their products or services meet the needs and desires of their customers. According to Schoppe: “It’s so tempting to get passionate about beautifully crafted tech. Especially since AI often yields impressive show cases early on. I am the first one to get excited about it. But here’s an ugly truth: no one pays for this. People pay you to solve their problems, that’s all it comes down to. When aiming to build a sustainable business, never start (only) with tech. Start with obsessively investigating your customers and their problems. That is your one and only north star.”

There are several reasons why finding the right product market fit is not easy:

1. Complex technology: AI can be a complex and rapidly evolving field, which can make it challenging for startups to develop products or services that are both innovative and practical.
2. Limited data: Many AI solutions require large amounts of high-quality data to train and improve their performance, which can be a challenge for startups that don’t have access to enough data or the resources to collect it.
3. Limited expertise: AI startups may struggle to find the right talent or expertise to develop and deploy their products or services, particularly if they are operating in a niche market.
4. Regulatory challenges: Some AI solutions, particularly those in industries such as healthcare or finance, may be subject to regulatory requirements that can be difficult for startups to navigate.
5. Competition: The AI startup market is highly competitive, and startups may struggle to stand out and differentiate themselves from their competitors.

Scalability

Scalability is one of the most common challenges faced by AI startups. Scalability refers to the ability of a business to grow and expand its operations without encountering diminishing returns or increasing costs.

“Half of the AI teams I talk to have built great tech to solve a problem for their customers and earned good money while doing so” says Schoppe, “but this early success can be misleading. Developing a useful AI solution for one customer is easy, developing an AI solution that works for 10 or 100 customers without retraining is exponentially more difficult.”

AI startups may face challenges in scaling their businesses for a few reasons:

1. Resource constraints: Many AI solutions require significant computational resources and data to train and operate, which can be a challenge for startups that don’t have access to the necessary infrastructure or funding to scale.
2. Limited data: As mentioned above, many AI solutions require large amounts of high-quality data to improve their performance, which can be a challenge for startups that don’t have enough data or the resources to collect it.
3. Talent: AI startups may struggle to attract and retain the necessary talent to scale their operations, particularly if they are operating in a niche market or are located in an area with limited tech talent.
4. Market demand: Even if a startup is able to overcome the technical and resource challenges of scaling its AI solutions, it may still struggle to scale if there is not enough demand for its products or services in the market.
5. Regulatory challenges: Some AI solutions, particularly those in industries such as healthcare or finance, may be subject to regulatory requirements that can be difficult for startups to navigate as they scale.

One specific challenge that AI startups may encounter when trying to scale is heterogeneous data, which refers to data that is diverse, unstructured, or derived from multiple sources. Heterogeneous data can make it difficult to build and train machine learning models that are accurate and reliable, as it may be more challenging to identify patterns and correlations in the data. This can cause many AI startups to go from pilot to pilot, testing their solutions on small, specific datasets, without being able to create a product that scales well and can handle a wide range of data.

I encountered similar problems when working for an AI startup in the medical sector. We tried to build models that would help radiologists analyse CT scans. However, each scanner generates slightly different images, which requires us to fine tune the models for each new scanner.

Creating a long-lasting unique selling point

It can be challenging for many AI startups to create a long-lasting unique selling point (USP) because the field of AI is rapidly evolving and highly competitive.

“Almost no field of technology develops as fast as AI. The state-of-the-art today is the commodity of tomorrow. And I mean it. If your startup’s sole USP today is your AI solution, you run a very high risk of losing any defensibility within two years. Software eats the world, but tomorrow’s AI will eat today’s AI.” - Schoppe.

A USP is a unique feature or benefit that sets a product or service apart from its competitors and makes it attractive to customers. In the case of AI startups, some potential USPs could include innovative AI technology, access to large amounts of high-quality data, or expertise in a specific domain or application.

However, creating a long-lasting USP can be challenging for AI startups for several reasons:

1. Technological advancements: AI is a rapidly evolving field; what may be considered innovative or unique today may quickly become commonplace or obsolete in the future. This can make it difficult for AI startups to maintain a unique selling point over time.
2. Competition: The AI startup market is highly competitive, and it can be difficult for startups to differentiate themselves from their competitors. Many startups may be working on similar problems or technologies, and it can be challenging to create a truly unique solution.
3. Customer preferences: Customer preferences and needs can change over time; what was once considered a USP may no longer be attractive to customers. This can make it difficult for AI startups to maintain a long-lasting USP.
4. Regulatory challenges: Some AI solutions, particularly those in industries such as healthcare or finance, may be subject to regulatory requirements, such as proposed by the AI Act, that can impact their ability to offer unique features or benefits.

Creating a long-lasting USP is a critical challenge for AI startups, as it can impact their ability to attract and retain customers, generate revenue, and achieve profitability. It is important for AI startups to carefully assess their target market and identify unique features or benefits that will meet the needs and desires of their customers over the long term.

Read on: How to Invest in AI — Part 3 (Due Diligence)

Disclaimer: I have written this article using my own experience and know-how regarding how I approach the evaluation and due diligence of an AI startup. To accelerate the writing I used the help of ChatGPT. However, I manually added examples, interviews, images and references, I tweaked the inputs and outputs, rewrote, and added paragraphs. Using ChatGPT allowed me to write this series of posts in a couple of days — instead of weeks.

The percentage of investments that are AI related has gone up. It almost seems as if every contemporary software startup revolves around the phenomenon of Artificial Intelligence (AI).

In my job at building and leading one of the biggest AI Startup Incubators in Europe, working closely with various founders, and teaching a course on AI for innovation and entrepreneurship at the Technical University of Munich, I screen a lot of pitch decks on AI and am privy to a great deal of AI startups as they come and go. Frequently I am asked by investors and VCs to examine AI startups and help them to make their investment decisions. In this series of posts, I wish to share my investment hypotheses and explain what I look for in evaluating and investing into AI startups.

How to Invest in AI — Part 2
How to Invest in AI — Part 3 (Due Diligence)

The challenge of investing in AI startups.

“I don’t think I’ve seen a PowerPoint without the word AI in the last two years. It is almost like software equals AI.” Brian Ascher, Partner at Venrock.

Investing in AI startups can be challenging. Many startups claim to use artificial intelligence, but in reality, they may not be using AI in a meaningful way, or may not have the technical expertise to effectively develop and deploy AI solutions. This can make it difficult for investors to accurately evaluate the potential of such companies. This situation has created a bubble in the AI startup market, with many companies claiming to use AI in order to attract investment. So-called “snake oil AI startups” are companies that claim to use AI in their products or services. In reality, they may not be using AI at all or may be using it in a superficial or misleading way. These types of startups are often more interested in generating hype and attracting investment than in delivering real value through the use of AI. One study even claims that “forty percent of ‘AI startups’ in Europe don’t actually use AI.“ This can make it difficult for investors to distinguish between legitimate AI startups with real potential and those that are simply riding the hype.

Additionally, in my experience, I have noted that many investors lack a deep understanding of AI and how it works. This lack of sufficient knowledge and experience can make it challenging to accurately assess the viability of an AI startup. The problem is as follows: when investors see large sums of money flowing into AI, they may invest without fully understanding the technology or completing an assessment of whether the investment will actually provide investment returns over time. In order to avoid the bubble and make good investments in AI, VCs and investors should be asking themselves how they can de-risk their investment with an understanding of technology and its potential for future returns. The necessary due diligence must also be comprehensively undertaken.

Here is my guide for the savvy AI investor:

Step 1 — Understanding the difference between AI and ML, and the value startups can create using AI

Artificial intelligence (AI) and machine learning (ML) are often used interchangeably, but they are not the same thing. AI refers to the ability of a machine to perform tasks that would normally require human intelligence, such as recognizing patterns, making decisions, or solving problems. Machine learning, on the other hand, is a subfield of AI that focuses on the development of algorithms and statistical models which allow machines to improve their performance on a particular task over time by learning from data, rather than being explicitly programmed.

One reason these terms are often mixed up is that machine learning is a key component of many AI systems and is used to enable such system to learn and adapt. However, not all AI systems use machine learning, and not all machine learning systems are considered AI.

Mixing up these terms can cause problems because it can lead to misunderstandings or expectations that are not met. For example, an investor may expect an AI startup to have developed advanced, self-learning algorithms when in reality, the startup is simply using basic rule-based systems. Similarly, a customer may expect an AI-powered product to be able to adapt and improve over time, but if the product is not using machine learning, it will not be able to do so.

What is the crucial difference investors need to understand?

In order to avoid misunderstandings and set realistic expectations, it is important to firstly understand the difference between traditional programming and machine learning and secondly, to accurately communicate which technologies a company is using.

There are two different ways of how programming works. Rule-based (traditional) programming and machine learning programming are two different approaches to building software systems.

Rule-based programming involves creating a set of explicit rules or conditions that a software system follows to perform a task. For example, a rule-based system might be programmed to send an email to a customer if certain conditions are met, such as if their account balance falls below a certain threshold. Rule-based systems are generally good at following precise instructions and are easy to debug, but they can be inflexible and may not be able to adapt to changing conditions or new situations.

Machine learning programming involves building algorithms that allow a software system to learn from data and improve its performance on a particular task over time. Machine learning algorithms are trained on a large dataset and use statistical models to find patterns and make predictions or decisions. They do not require explicit rules or conditions to be programmed in advance, but instead learn from the data they are given. Machine learning systems can be more flexible and adaptable than rule-based systems, but they can also be more complex and require more data and computational resources to train and operate.

Both rule-based and machine learning programming approaches have their strengths and limitations, and the appropriate approach will depend on the specific task and the goals of the system.

Understand how AI can deliver value

Understanding the value that AI can create is important because it can one help assess the potential risks and opportunities associated with an AI startup, thereby ensuring the potential investor makes more informed investment decisions. By understanding the ways in which AI can create value, I can evaluate the growth potential and long-term viability of the startup I am looking at.

AI can create value in a number of ways, depending on the specific application and use case. Some of the ways in which AI can create value include:

1. Improving efficiency and productivity: AI can automate certain tasks or processes, enabling organiyations to work more efficiently and effectively.
2. Enhancing decision-making: AI can analyse large amounts of data and provide insights or recommendations that can help organisations make better decisions.
3. Creating new revenue streams: AI can enable organisationsto offer new products or services, or to target new markets, which can help to drive revenue growth.
4. Reducing costs: AI can help organisationsto streamline their operations and reduce costs, which can improve profitability.

Read on: How to Invest in AI — Part 2

Disclaimer: I have written this article using my own experience and know-how regarding I approach the evaluation and due diligence of an AI startup. To accelerate the writing I used the help of ChatGPT. However, I manually added examples, interviews, images and references; I tweaked the inputs and outputs and rewrote and added paragraphs. Using ChatGPT allowed me to write this series of posts in a couple of days — instead of weeks.

An analysis of the history of technology shows that technological change is exponential, contrary to the common-sense “intuitive linear” view. So we won’t experience 100 years of progress in the 21st century — it will be more like 20,000 years of progress (at today’s rate). The “returns,” such as chip speed and cost-effectiveness, also increase exponentially. There’s even exponential growth in the rate of exponential growth. Within a few decades, machine intelligence will surpass human intelligence…

Ray Kurzweil (2001) “The Law of Accelerating Returns”

Technology progresses at an exponential rate. One of the first people to denote this trend was Gordon Moore, co-founder and chairman emeritus of Intel. Moore predicted in 1965 that the number of transistors that fit on a computing chip doubles approximately every two years. Moore’s law not only inspired futurist’s Ray Kurzweil’s Law of Accelerating Returns, but would set the pace for our modern digital revolution, as his observation had widespread impact in many areas of technological progress, including artificial intelligence (AI).

Danny Hernandez and Tom Brown from OpenAI observed that AI models are becoming cheaper to train at an exponential rate faster than Moore’s Law, as algorithmic efficiency and hardware efficiency — two key driving factors for the advance of AI — are growing exponentially.

Looking ahead, we can only speculate as to what decades of exponential improvement in the efficiency of AI algorithms will bring to the future. We have already observed major breakthroughs in the last years, when we witnessed AI algorithms coming at par — and even surpassing — human capabilities. Algorithms such as AlphaGo, GPT-3 or AlphaFold are impressive breakthroughs that pushed the boundaries of what seemed possible and have defined the history of AI. However, we must emphasize that the idea that AI is progressing exponentially is slightly misleading, as anyone who has used a conversational assistant might tell you. My experience repeating impatiently my commands to Alexa, who has been my companion for a couple of years now, sometimes leaves me wondering if there is any progress at all. Similarly, if the AI applications driving self-driving cars would have progressed exponentially, I would not always have to ask my brother to pick me up with his car from the train station whenever I visit home. When I say AI is growing at an exponential rate, I thus refer to the growing range of applications and capabilities that we see in the field of AI.

As developments in AI speed up, organizations must react in order to keep up and prosper. However, many organizations are not adept at changing, especially as they grow in scale. If a company increases in size it also increases in complexity and bureaucracy. We human beings are limited in our ability to organize and handle complexity and scale. According to Charlie Munger, investor and vice chairman of Berkshire Hathaway, the greater the number of people working together in an organization, the more bureaucratic, self-interested and dysfunctional they tend to become:

“They also tend to become somewhat corrupt. In other words, if I’ve got a department and you’ve got a department and we kind of share power running this thing, there’s sort of an unwritten rule: “If you won’t bother me, I won’t bother you and we’re both happy.” So you get layers of management and associated costs that nobody needs. Then, while people are justifying all these layers, it takes forever to get anything done. They’re too slow to make decisions and nimbler people run circles around them.”

Entrenched interests slow organizations down and create resilience. Compared to the exponential rate of change of AI technology, organizations change at a logarithmic rate. The idea that technology is changing faster than organizations is at the center of “Martec’s Law” (which is more of an observation and not really a law) formulated by Scott Brinker, founder and chair of the MarTech Conference. Brinker sees the gap between the rate of change at which technology progresses and the ability of organizations to adapt as the quintessential management challenge of the 21st century

The same challenge can be observed when we look at the struggles of organizations starting to adopt AI: Many still struggle to embrace AI or fail to extract value. One study found that only 6% of companies have adopted AI. A survey from BCG from 2019 found that “many AI initiatives fail. Seven out of 10 companies surveyed report minimal or no impact from AI so far. Among the 90% of companies that have made at least some investment in AI, fewer than 2 out of 5 report obtaining any business gains from AI in the past three years.” Recent research from 2020 found that even though “organizations are using AI as a tool to generate value […] the majority of companies [are] still struggling to capitalize on the technology.”

Even though many still struggle to turn AI into value, they understand the opportunities that AI can bring for their organization. The benefits of adopting AI can be seen from the handful of (primarily tech) companies that have been able to transform their organizations to embrace and adopt AI at scale. These so-called AI-factories have shifted their strategies to put AI in the center of their organizations’ business and operating models. Google became one of the first companies to start this transformation when its CEO Sundar Pichai announced in 2017 that Google would shift from a mobile first world to an AI first world. An AI-first approach allows organizations to automate decision-making, thereby transforming the way they create and capture value. Not only have these companies been able to increase their competitive advantage, but they have been rewarded by skyrocketing stock prices. As of writing this article, all companies valued over 1 trillion have adopted an AI first approach. One of these AI first companies is Tesla, whose market value reached 1.01 trillion USD in October 2021, making Tesla worth as much as the next 10 most valuable global automakers combined, even though it sells only a fraction of the cars of the others.

The faster the field of AI progresses and the slower the rate of adoption the greater the gap between expectations and practice will become. The challenge is that AI requires fundamental organizational change. Most companies, however, are far away from adopting an AI first approach, as this requires a clear vision and a holistic approach to transform the organization. This, in turn, requires rethinking and commitment over all divisions and hierarchies. Historically, there have been especially many large and successful companies that seem to be more likely to struggle when new innovations require them to change their organizational structures and reinvent themselves.

For instance, despite inventing the digital camera, Kodak struggled to transform themselves and eventually filed for bankruptcy in 2012. Blockbuster did not manage to adapt their business model to the internet and so lost out to Netflix, one of the big companies applying AI at scale.

So, what should companies do?

First of all, there are no quick wins. A holistic transformational change is not an easy task and requires substantial resources. However, doing nothing is not an option. McKinsey predicts that AI technologies could lead to a performance gap between front-runners and nonadopters. As with Tesla, front-runners are likely to benefit disproportionately, while late adopters might see their cash flow decrease as they become less competitive and eventually might face similar consequences as Kodak or Blockbuster.

One way to reduce the gap between technological change of AI and organizational change is to become more agile and start setting up the company in a way that enables the adoption of AI. Organizations must make sure they have the right structures in place to adapt, innovate, and take AI to scale. This process starts with acknowledging the value that AI can bring to the organization, defining a vision for AI, working on a strategy for building an AI-powered organization, and enabling data driven decisions. As AI can help to automate decision making it also empowers the whole organization to become more agile and responsive, increasing the rate at which an organization can adapt to change and incorporate technological progress.

Another solution could be an organizational “reset”. Because of their small size and their entrepreneurial can-do attitude, startups often show a high degree of flexibility and innovation. This agility allows them to adapt to change to a greater degree than the more complicated bureaucracies found in large organizations. These startups do not have to rely on old legacy systems and processes that slow them down. At the heart of the organizational reset lies the question: “If we were to build the organization from scratch, what would it look like?” Organizations can achieve a reset through an internal reorganization or by spinning off a new group that can operate with a fresh start, without the inertia of the existing organization. Digital transformation initiatives typically take one of those two approaches to leapfrog to a new baseline on the technology curve. As organizational resets can be extremely disrupting, some larger organizations have created incubator-like structures to systematically spin-up small, intrapreneurial startup teams.

Looking back, we see that most organizations will not be able to adapt. Ultimately, a reset will happen when a company fails and the resources are reallocated to the market. This happens much more often than we do realize. From 1955 to 2014 88% of Fortune 500 companies failed , having undergone ‘creative destruction’. Even Jeff Bezos believes that one day Amazon will face inevitable death: “[…] I predict one day Amazon will fail. Amazon will go bankrupt. If you look at large companies, their lifespans tend to be 30-plus years, not a hundred-plus years.”

AI brings a lot of opportunities. For many existing organizations its adoption will be a challenge. However, it is not too late to start the necessary transformation. Many organizations are still at the beginning of this process. As the technology progresses they will face increasing competition from other players more capable of leveraging AI.

At the AI Strategy Institute, we empower organizations and individuals to lead in the Age of AI. Combining the roles of think tank, consultancy, and academy, we offer insights, strategies, and educational programs to drive responsible AI innovation. Discover our offerings and see how we can help you stay ahead in an AI-driven world.

The Paradox of Applying AI was originally published in AI Strategy Institute on Medium, where people are continuing the conversation by highlighting and responding to this story.

Despite many artificial intelligence (AI) systems matching or surpassing human capabilities in various ways, we are still at an early stage of adopting AI. Only a few corporations have been able to leverage the capabilities of AI and see their market value soar, but the long tail of organizations is still struggling to extract value from the use of the technology. Building an organization that is powered by AI requires fundamental changes to the organizational structure. To drive this change requires a team that has the necessary skills, experience, and know-how.

Many larger organizations, especially those who successfully scale AI, have already established such a unit, often branded as AI Competence Center (CC) or Center of Excellence (CoE) and with it have appointed new roles such as VP of AI, Chief AI Officer, Chief Data Scientist or Director of AI research. To goal of this article is to provide readers with a quick overview and resources that will help decision makers to better understand what it takes to establish such AI units/departments.

What does the core AI team do?

The role of an AI CoE is diverse and slightly depends on the organization’s ambition, size, and AI maturity. For instance, for companies who are just starting with their AI journey, the CoE will not only be smaller but might initially focus more on the AI strategy and project management. While the role of the CoE at more mature organizations focuses more on the execution, professionalization, and scaling of AI. Generally, the AI unit is responsible for a range of tasks that can be clustered into four different categories:

1. (Supporting the) shaping, definition, and execution of the organization’s AI ambition and strategy. The AI ambition sets the high-level goals of the application of AI within a company regarding the fields of application as well as the level of aspiration. It is part of the AI strategy and should be aligned with the overall corporate strategy. The tasks of the AI CoE include creating a vision for AI at the organization, communicating the ambition, and helping to execute the AI strategy.
2. (Supporting the) ideation, assessment, and prioritization of AI use cases in the organization. To successfully do AI, it is important to come up with the use cases that solve a need within the organization. The AI CoE supports in identifying the AI use cases, determining the appropriate level for ambition, managing external innovation by scouting AI technology, applications, and partners.
3. (Supporting the) creation of enabling factors in the organization. Becoming an AI-centered organization often requires fundamental changes to the setup and working of the organization. The AI CoE can facilitate the change management process and take over or support different functions within the organization, such as developing an understanding for AI within the organization, supporting HR to retain and attract AI talent, creating a target data architecture, spreading success stories, and/or developing and maintaining an AI ecosystem.
4. (Supporting the) execution of AI projects within the organization. Obviously, one of the core tasks of an AI CoE is to experiment with AI technologies and applications and obviously to develop, deploy, and maintain AI systems.

From these tasks, it becomes apparent that the AI unit has many different touchpoints in the organization and engages with a lot of different stakeholders. This can be a challenging task especially in larger corporations where the bureaucracies and internal politics can become more complex. Therefore, C-level executives, such as the CIO, have a huge role to play, but they can’t do it on their own because so many of the challenges around AI go beyond their scope of influence. It helps mightily if a critical mass of several top executives, including the CEO, personally commit and drive the rest of the company toward AI maturity.

Excellent resources that give a more detailed description of the varying tasks are:

• Davenport, T., and Dasgupta, S. (2019). How to Set Up an AI Center of Excellence. Harvard Business Review.
• Ramel, D. (2019). How to Create an Enterprise ‘AI Center of Excellence. Pure AI.
• Di Fiore, A., Schneider, S., and Farri, E. (2018). The 5 Things Your AI Unit Needs to Do. Harvard Business Review.

How should the AI team be structured within the organization?

Among leaders and managers, there is wide agreement that certain aspects of AI are best managed centrally (obviously, it also depends on the organization’s size and AI maturity). Central coordination prevents teams from duplicating efforts and ensures collaboration. For example, if one team has successfully solved a problem using AI, there should be a mechanism in place that will enable that team to share that knowledge on a wide scale. To support this collaboration, organizations must centralize certain types of AI expertise, such as machine vision, natural-language processing. Also, central coordination dramatically increases efficiency and effectiveness by:

• Consolidating best practices on AI tools (e.g., ML platforms or container solutions for deploying models)
• Providing benchmark of solutions
• Offering central repositories of training and test data sets, and reference models
• Facilitating easy access to preferred tools

Tradeoffs between centralized and decentralized AI CoE

However, centralization does not offer a complete path to AI success. When organizations centralize AI responsibilities, they can inadvertently build isolated teams of “techies” that develop insular solutions without real business value. Therefore, the most successful AI initiatives set the right balance between centralized and decentralized activities. Use cases should be developed in a bottom-up way from business units where people understand customer needs and pain points in the processes. Leaders need to create enough room for local, decentralized initiatives. These initiatives should be able to draw on centralized know-how and resources to effectively tackle identified challenges. One best practical approach that often can be seen in companies is a hybrid approach. Here, the CoE bundles certain functions and expertise such as data governance, AI recruiting and training strategy, and work with third-party providers of data and AI services and software, while maintaining strong links to decentralized units and the rest of the organization. The AI CoE can be complemented by an integration team responsible for assimilating and scaling PoCs into the organization’s structures. In some cases, this model is known as the “hub and spoke” model. The hybrid approach ensures balance and promotes collaboration between the centralteams and decentralized units, both are key to the successful application of AI.

More on the AI team structure can be found here:

• appliedAI (2020). Building an organization for scaling AI. UnternehmerTUM.
• Fountaine, T., McCarthy, B., and Saleh, T. (2019). Building the AI-powered Organization. Harvard Business Review.
• Hotz, N. (2019). Data Science Team Structure. Data Science Project Management.

How to staff an AI unit?

To staff an AI team, the right mix of talent and skill is needed. This spans various profiles and roles ranging from more technical ones such as Data Scientist, Data Engineer, Software Engineer, and Infrastructure Engineer, to more strategic ones such as User Interface Designer, Business Consultant, (AI) Product Managers and Domain Experts. Ultimately, how to staff the AI unit depends on the structural setup of the AI CoE and where the AI competencies within the organization should be allocated.

However, when thinking about how to staff an AI team, it can make sense to look at the typical lifecycle of an AI project:

Broadly, the profiles that are involved in a typical AI project can be broken down into four distinct core disciplines that come in in different and sometimes multiple stages of the project:

• Experts that understand the organization’s goals and needs, the core and supporting processes, the product and/or service of the organization. They have an understanding of how AI can be applied to address those things. Roles include but are not limited to AI Strategists, Business Analysts, Product Owners, Domain Experts.
• Experts that understand and know how to work with data. People who develop the data architecture, build data pipelines and data lakes, who oversee the data quality and how data flows within an organization. Roles commonly include Data Engineers and Data Analysts.
• AI experts, who have experience with topics such as machine learning, statistics, forecasting, and optimization. These are those who can build, train, and evaluate the AI models but who are also capable of maintaining them. Typical roles are Data Scientists, AI Engineers, Machine Learning Experts, AI Researchers, etc.
• Experts that know how to build the right architecture to embed the AI models and how to scale them throughout the organization. These are those who deploy AI and bring the models into production. Roles vary from DevOps, Software Engineers, Infrastructure Engineers, UI/UX Designer, Product Designers, etc.

It is important to note that all those individuals from different disciplines have to be able to collaborate effectively with each other. Often, organizations are not used to collaborating in this way, and thus fail to bring AI into production. However, looking at the AI lifecycle, it is paramount that, for instance, an engineer can take a machine learning model created by a data scientist, to rewrite the code and bring it into production.

In addition to looking at the AI lifecycle, it is important to include people in the AI team that work on the overall AI strategy and the supporting functions to execute the strategy.

As earlier mentioned, building an AI team needs executive support. Therefore, it is important to have an executive sponsor. A C-Suite sponsor can take an active role and make sure that the AI unit’s efforts are aligned with the company’s strategy. Given the wide range of new expertise required and the extensive demands of the AI transformation, the question arises whether the board should be strengthened via a new dedicated role in charge of the AI journey; the Chief AI and Data Officer (CAIDO). In some ways, this role resembles the Chief Digital Officer at the onset of the Digital Age. But I would argue that the CAIDO role requires even deeper technical (AI) expertise and involves more permanent tasks. The position’s role profile further includes challenging the organization’s status quo from an ‘AI-first’ perspective, driving AI initiatives throughout the organization, and serving as a sparring partner for other board members.

As AI systems can bring their own kind of legal and ethical challenges around issues such as bias, fairness, trust, explainability, and accountability, it needs people who can navigate those challenges. The AI team should have support from legal experts and, in some cases, even AI ethics experts to deal with the regulatory landscape, such as GDPR or the EU regulatory framework on Artificial Intelligence. One measure to build less-biased AI, that has recently gained more and more traction, is the need and push to hire more inclusively. Ideally, AI teams should achieve diversity across race, gender, sexual orientation, age, economic conditions, and more. This will not only help to combat bias but also drives creative thinking.

As there is a high demand for AI talent and a shortage in the supply of experienced talents, finding and hiring the right people to staff the AI CoE can be a challenging and costly task. Because of this existing AI skill gap, staffing and a good recruitment strategy are paramount to bring expertise, knowledge, experience, and manpower into the AI CoE. An alternative can be to offer training to existing employees to fill the positions. With the rise of digital content, including MOOCs (Massive Open Online Courses) such as Coursera, e-books, and YouTube videos there are plenty of resources (freely) available to train large parts of the workforce.

Additionally, an AI unit should have access to sufficient budget and resources to be successful, this is not only limited to paying salaries to talent, which sometimes can demand stellar figures but also includes access to (raw) data, compute resources, and the relevant soft- and hardware necessary to do machine learning and AI.

Further readings on how to staff an AI team and the core roles are;

On the different roles and how to staff the team:

• Kozyrkov, C. (2019). How to staff an AI team: 11 key roles. The Enterprisers Project.
• Beck, M., Davenport, T., and Libert B. (2019). The AI Roles Some Companies Forget to Fill. Harvard Business Review.
• Knopp, E. (2018). Building your AI team: The roles your enterprise needs. IBM.

On the CAIDO:

• appliedAI (2020). Artificial Intelligence for Boards — Gearing up for the Future of Business. UnternehmerTUM.

On more diverse teams:

• Li, M. (2020). To Build Less-Biased AI, Hire a More-Diverse Team. Harvard Business Review.
• Shastri, A. (2020). Diverse Teams Build Better AI. Here’s Why. Forbes.

The AI Research Lab

We see that many organizations are unclear about what constitutes the difference between an AI CoE and an AI Research Lab. The goal of an AI CoE is to leverage the possibilities of AI and apply them to the organization’s processes, products, and services. It involves applied research, which is designed to solve the real-world problems that the organization faces. However, if the organization wants to go beyond doing data science and really push the frontiers of AI (and the industry), they need to establish an AI Research Lab to conduct fundamental research. Because this mainly requires PhD-level experts with backgrounds in computer science, maths, statistics, etc. and often come straight from academia this will bring a very specific dynamic and culture with unique needs into the organization. While this might be a bold (and costly) endeavor, it can generate valuable IP and can give organizations a vast competitive advantage from owning cutting-edge AI solutions in their industry. It is not an easy task to set up an AI research lab, it requires commitment, substantial (long-term) investments, and often creates a chicken and egg problem. Organizations cannot attract top AI researchers because they do not have a research lab, and they do not have a research lab, because they cannot attract top researchers. The big tech companies that follow an AI-first approach like Google (Alphabet), Facebook, Amazon, Microsoft, Netflix, Ali Baba, and Baidu, which account for the majority of money spent on AI research (about $20–30 billion of the $26–39 billion in total private AI investment expenditures worldwide), have invested substantial amounts to build up such AI research labs. Even though smaller organizations have also successfully established such research labs, this undertaking might not be reasonable for every organization.

Further resources are:

• Post, T. (2022). The Difference Between an AI Team and an AI Research Lab. appliedAI.de.
• Metz, C. (2021). Genius Makers: The Mavericks Who Brought AI to Google, Facebook, and the World. Dutton.
• Agrafioti, F. (2018). How to Set Up an AI R&D Lab. Harvard Business Review.

At the AI Strategy Institute, we empower organizations and individuals to lead in the Age of AI. Combining the roles of think tank, consultancy, and academy, we offer insights, strategies, and educational programs to drive responsible AI innovation. Discover our offerings and see how we can help you stay ahead in an AI-driven world.

How To Set Up An AI Unit was originally published in AI Strategy Institute on Medium, where people are continuing the conversation by highlighting and responding to this story.

“The customer is king,’’ says an old adage that highlights the importance of the customer’s experience. But while the adage itself might be old, it remains just as relevant today at a time when it’s clear that good customer service is a decisive factor in the success of an organization.

Today, Artificial Intelligence already plays a major role not only in improving the overall quality of customer service provision but also in the efficiency of how those services can be provided, thus also increasing the satisfaction of customer service employees. To help understand best practices today and identify technologies that will become relevant in the near future, we got together with our partners and experts from EnBW, Telekom, Miele, Google, and IBM.

Customer service has always been an area that has been at the forefront of technological developments. The need for customer service teams was created by the industrial revolution and the arrival of mass production. The invention of the telephone and, shortly after, the switchboard, enabled the development of the first call centres in the 1960s. With the rise of the internet, and technologies such as email, live chat, smartphones, and the rise of social media platforms, customer service teams were faced with new modes of communication. With these came many new challenges: Not only did customer service teams now have to manage multiple communication channels simultaneously, but they also experienced an influx of service tickets that proved to be impossible to handle by hand. Nowadays, with increasing technological capabilities themselves, customers are also expecting a much better customer experience. For instance, many expect the customer service agent to already know who they are before they reach out, or expect service agents to anticipate their needs.

Artificial Intelligence can help to address many of the challenges we see in customer service today. It empowers customers, service agents, and managers and it drives customer success. AI enables improvements in efficiency by subtly shifting parts of the workload from customer service agents to the customers themselves using services such as chatbots, voice authentication, smart search, and intelligent data extraction.

Meanwhile, service agents are supported by AI tools such as root cause analysis, enterprise search, and intelligent call routing, which improve the quality and efficiency of the service they provide. Other AI tools, such as intelligent call analysis, automatic integration into CRM, and trend identification, benefit managers by giving them better insights that can be used to improve products and services. In short, the application of AI in customer service allows organizations to provide a better and more personalized experience that empowers all stakeholders without losing the human touch.

Organizations not wanting to miss out need to integrate AI into their customer service provision, because great customer service pays off. Research shows that customers are willing to spend more with a brand that has outstanding customer service and are more likely to make repeated purchases. Not surprisingly, many organizations already use AI to drive customer success and are continuing to develop, test and deploy new solutions. This also includes our partners with whom we have identified 17 uses of AI and created this infographic of AI in Customer Service:

From the many uses of AI in customer service we shortlisted, our partners highlighted three applications of AI that are high on their priority list:

Intelligent Search

Intelligent search, powered by Artificial Intelligence technology, eliminates data silos and helps employees and customers find the information they need quickly and easily. End-users can use intelligent search to extract information from data located anywhere, both inside or outside the company, regardless of the format: big data in databases, document management systems, digital content, webpages, on paper, wherever. Intelligent search and enterprise search are synonymous with natural language search, AI search (or AI-powered search), and cognitive search. Intelligent search empowers customers to find what they are looking for faster and without resorting to using customer services, thus, reducing the burden that falls on service agents.

Chatbots

A chatbot is an application that uses Artificial Intelligence to simulate human conversation (a chat) with a human user in natural language through voice commands, text chats, or both. It can be used via messaging applications, websites, mobile apps, or the telephone. A chatbot leverages Natural Language Processing (NLP) and deals with two tasks: User request analysis, (i.e. identifying the user’s intent) and returning a correct response.

Chatbot applications streamline interactions between people and services, enhancing customer experience. At the same time, they offer companies new opportunities to both improve the customer’s engagement process and also improve operational efficiency by reducing the typical cost of customer service. With chatbots, it is possible to provide a 24-hour service and be readily available to more customers.

Intelligent Call Analysis

Intelligent Call Analysis or Conversation Intelligence uses AI to react to a conversation in real-time, record calls, transcribe them, enable call playback, and analyse and collect metrics. It is commonly applied in sales calls where the AI employs intelligent real-time analysis of conversations with clients and sends suggestions to guide the conversation efficiently, but the organization can also profit greatly from this application. AI can generate insights from customer interactions with service agents which can then be passed on as feedback to the workstation of a call agent or sales manager for further action. Conversation Intelligence is empowered by speech recognition that converts audio into text and uses natural language processing techniques to analyse the conversation, find patterns, and make predictions.

AI in Customer Service was originally published in appliedai.de on Medium, where people are continuing the conversation by highlighting and responding to this story.