It is nowadays impossible to question the fact that Artificial Intelligence unlocks enormous opportunities for the economy. According to a study by the German digital association Bitkom on the use of AI, which it also repeated in 2021, almost 70% of the 600 companies surveyed see AI as the most important technology of the future. In addition, over 60% see direct opportunities for their own business. Only 11% believe that they will not be influenced at all by AI technologies .
As shown in Figure 1, however, these expectations are not reflected in the companies’ strategies or measures. The proportion of companies that are already actively using Artificial Intelligence is still well below 10%, and only an additional 30% are planning or discussing its use.
Figure 1 – AI usage in Germany (2021: n=603, 2020: n=603, 2019: n=606, 2018: n=604) 
There are several reasons for this discrepancy. On the one hand, it is evident that it is primarily large companies that are investing in AI. Some of these companies have their own teams working on the technology. The smaller the companies, the less happens in terms of in-house development. Nevertheless, there’s leverage of individual potentials through standard products. Broken down more concretely, there is usually a lack of time capacity, closely linked to budgets, or a lack of data.
What must strike you when reading this evaluation is the one-sided view. Arguments such as “too little time” or “no corresponding budget” only consider the cost side. What is completely left out is the benefit with the corresponding added value that is created. From this we deduce that such a potential analysis has never been carried out in many places.
As a solution to efficiently perform this analysis, we have developed the Cloudflight AI Patterns. These are a structured consulting approach to get from the pure cost view via the identification of the possible application areas of Artificial Intelligence in the company to an ROI consideration. As an obvious added value, companies can take individual measures according to economic calculations. In addition, possible synergies become visible, which facilitates longer-term decisions on a strategic level.
In addition to digital transformation, we now also frequently speak of AI transformation. Similarly, the term “AI Native” has already been coined for companies that build their business models and processes on the possibilities of this technology throughout.
But what distinguishes AI from other technologies or innovations to an extent that such an extensive consideration is justified or even necessary? It is important to understand that Artificial Intelligence is not a finished application, but a basic technology. As such, it has already been compared to electricity (Andrew Ng: “Just as electricity transformed almost everything 100 years ago, today I actually have a hard time thinking of an industry that I don’t think AI will transform in the next several years.”) . By analogy, AI can also be found in a wide variety of applications, in which it then delivers its added value.
Of course, there are AI products that companies can simply buy in, and which also immediately deliver benefits. Let’s take autonomous vehicles as an example, which we will have available in the foreseeable future. Business trips will then no longer require attention to driving per se. As a result, the people involved can use the driving time for other activities, which immediately creates a positive effect. However, this possibility is generally available. So, there will be some efficiency gain, but no sustainable possibility for differentiation.
The situation is different when it comes to a company’s core processes. Here, the potential to make a dent in the market is correspondingly higher. However, it requires a different kind of investment. The decisive factor here is to re-evaluate one’s own business against the background of the changed technological possibilities. Today, labor-intensive services are already being automated to some extent. This can pose a substantial threat to companies, but conversely also corresponding opportunities. These must be evaluated – for example, using familiar methods such as a SWOT analysis – and considered in the corporate strategy.
The challenge here is not only to take into account the current state-of-the-art, but also to anticipate further developments. Here, it is worthwhile to consult experts who are not only familiar with the current research topics but also their respective pace of innovation.
One should not be deceived by reports of the high number of AI projects that never make it into productive operation. The entire industry has gained experience from such failures – especially those companies that invested early on. Companies have learned that ethical issues need to be considered in AI transformation, that the introduction of larger AI systems is also an organizational development topic, or that realistic expectations and acceptance need to be secured as early as possible.
The technology has also evolved very quickly. By using increasingly powerful, publicly available frameworks, models or services, companies can already achieve impressive results with moderate effort.
AI Patterns Approach
We define a single AI pattern as a recurring use pattern of Artificial Intelligence. It is described by a unique name, a generic description of the challenge – detached from industries or business units – and several examples of use for a more concrete understanding.
Orienting oneself to such patterns has a variety of positive effects, starting with a common understanding of the characteristics of AI and the relevant terms, through empirical values on the actual achievable cost/benefit effects, to more efficient implementation due to reusable software components. The overall set of AI patterns also ensures that all relevant areas are considered throughout.
It is important to emphasize that orientation to recurring patterns does not stand in the way of individualization and the associated differentiation. There is sufficient scope for design in the actual application and implementation of the patterns, so that competitive advantages can very well be developed here.
The AI Areas
Artificial Intelligence is now a very broad field. Striking examples range from medical research to speech recognition and self-driving cars. Regarding current applications, we have delineated the following areas in order to also structure AI patterns:
Natural Language Processing
Natural Language Processing (NLP) is currently a very fast-growing field. On the one hand, this is supported by several technical breakthroughs in recent years, as well as the increasingly powerful language models available on which applications can build. On the other hand, it is an area of application for Artificial Intelligence that affects every single company regardless of industry or size.
Even if core processes have already been digitalized to a great extent, most companies still experience modality breaks. Time and again, people must transfer information from emails or documents into an additional system. This means that information that is basically already available in natural language has to be structured manually and entered again.
In addition to this pattern of information extraction, there are many similar applications, from a simple classification of documents or even just small text snippets to complex chatbot systems.
Another aspect of natural language is, of course, spoken language. In addition to the automation of services such as telephone surveys, speech recognition is also used where machines are to be operated while the hands are needed for other activities – such as, classically, driving a vehicle.
Image / Video Processing
In addition to speech, visual perception is an extremely important factor influencing our actions. We can see and react to dangers or avert them at an early stage by grasping a scene. Likewise, we can steer by purposefully grasping an object or following a road. This is now continuing in technology applications. At the end of 2020, for example, Tesla announced that it would rely purely on color cameras in its vehicles and would no longer require alternative technologies such as lasers .
Artificial Intelligence also helps here with descriptive approaches to overcome modality breaks between the real world and processes or decision paths cast in software. Obvious applications are machines and robots, which initially move in well-defined, delimited environments, but now also in public spaces. These are not only the much-cited autonomous vehicles, but also cleaning machines or general assistance robots, for example.
In addition to mobile robots, there is also a wide range of applications in stationary facilities. Access systems benefit from visual, biometric features. Co-bots, which work together with humans, must know their environment in order to work efficiently, but above all safely. Another very important topic for the industry is quality assurance.
Time Series Analysis
We speak of Time Series Analysis when not only a current value is relevant, but also its development. This can concern trends in economic data such as demand or prices. If one knows one’s own sales development, one can plan resources more efficiently, cushion selectively higher demand by producing in stock, order raw materials in time without having to keep more than necessary in stock, etc. The market brings in further dynamics here with fluctuating prices. Obviously, everyone wants to buy as cheaply as possible and sell as expensively as possible.
In the industrial environment, there is also data from production. Here, a large number of sensors are used to monitor temperatures, pressures, speeds, etc. In simple cases, a single sensor value can be used to draw conclusions about whether the underlying process is running as intended. In simple cases, a single sensor value can be used to draw conclusions as to whether the underlying process is running as intended. Often, however, it is the only the combination of different variables that provides this information. An important use case here is predictive maintenance. Maintenance intervals are no longer carried out preventively in defensively short intervals, but rather the actual closure or emerging defects are observed and reacted to as required.
The fields of application and techniques of Artificial Intelligence presented so far are on the one hand descriptive (meaning that they help to describe the real world through structured data) or predictive (meaning that they provide predictions about the future development of certain target variables). What both variants leave open are concrete actions to be derived from them. This is either left to subsequent software systems with formalized decision-making processes, or to humans, who are supported in their decision-making.
One field of Artificial Intelligence deals with so-called prescriptive methods, which directly recommend actions or develop longer-term strategies. These are based on behavioral psychology and reinforcement learning. The idea behind this is simple: an actor interferes within his environment and receives positive or negative feedback for individual actions. Over time, we learn to maximize positive feedback and also that short-term gains are not always the decisive factor, but rather long-term success.
There are countless applications in business. They range from trading strategies in financial markets, to asset management in changing environments, to optimizing consumption recommendations in ongoing customer relationships.
In a final collection of AI techniques, we summarize all those that are generally valid and cannot be clearly assigned to one of the subcategories. On the other hand, there are also specialized areas of application that are very relevant for individual industries, but do not have sufficient breadth to justify their own category.
Generic methods include, for example, anomaly detection. This can be performed on text, images, as well as time series of financial or sensor data. It should be mentioned here again explicitly, in order to deal sufficiently with overlapping applications in the analysis.
Another example is generative AI methods. Here, Artificial Intelligence imitates creativity and generates images, text, soundtracks, etc. that never existed. This ranges from artificially generated TV announcers to thousands of designs in industrial design generated based on definable requirements, from which the experts can then select.
Of course, one could go into more detail here. In addition to speech and images, audio processing would be a predestined field if one assumes human perception. However, practice has shown that the various use cases are highly relevant in individual cases but are used less widely across industries. The same is true for the generative AI approaches readily mentioned above. We cover this situation with the corresponding broader AI patterns here.
The business units
In every company there are different areas – from product development and production or service as often core topics to human resources, finance, marketing, sales, etc. The larger the company, the more or finer the units. It is obvious to start a transformation in the core areas in order to differentiate oneself on the market or in those areas in which optimizations achieve the greatest leverage due to high volumes.
Nevertheless, when we speak of AI transformation, all areas of the company must be included. However, the different relevance and the way in which the divisions contribute to the joint success of the company has the effect of differentiated objectives. One variable is the consideration of the necessary differentiation already mentioned above. Here, one can fall back on the general corporate strategy and derive from its core areas in which investments in individual solutions are profitable or perhaps even necessary. In other areas, standard solutions can be used. The AI market has matured to such an extent that there is now a very wide range of AI products.
In addition to differentiation, it is also important to weigh up the extent to which AI support, automation or autonomization makes economic sense from a cost/benefit perspective. Other aspects also play a role, of course, such as limited scaling options due to a lack of personnel or the situation on the labor market, possibly higher throughput speeds, usable synergy effects or simply marketing purposes. However, the goal of a transformation project is not to run after every possible application, but to know and be able to assess the potential of AI. AI-native also refers to the mindset of having to justify why AI should not be used in one area of a company instead of arguing in the other direction.
Production looks very diverse in different industries. In the sense of a uniform approach, we also consider the provision of a service as production. In the classic, industrial sense, however, this area is often linked to increasing the degree of automation of machines and systems, as well as the early detection and handling of errors, quality problems or time problems. The smaller the batch sizes, the more support in product design and in defining and optimizing the actual production process and its parameters also becomes a relevant factor.
More and more products are becoming intelligent themselves. This primarily concerns natively digital products – consumer apps and platforms as well as enterprise software. But traditionally analog products are also increasingly being enriched with software and intelligence. Depending on the type of product, all AI topics must be considered here.
When evaluating AI potential, it is also important to explore the possibilities of new products. For example, particularly efficiently implemented solutions from other areas within the company can become a product in their own right.
Marketing, Sales, After-Sales
Today, marketing is a field of application for Artificial Intelligence that should not be underestimated. On the one hand, this involves topics such as customer segmentation, targeted product placement, and addressing specific target groups. On the other hand, chatbots are playing an increasingly important role in customer care. It should be noted that the technology is not limited to websites but can also handle emails or other forms of customer communication and either respond itself or forward them to the right specialist department depending on the topic. The recognition of relevant information on an order can also be extracted and transferred to the downstream ERP systems.
Voice interfaces can also be used to automate telephone calls that follow a certain structure. Examples of this are telephone surveys or appointment reservations.
The applications in logistics range from route optimization and self-driving forklifts to autonomous control of warehouses and monitoring of stored goods to automation of “paper work” such as matching scanned delivery bills with orders and invoices.
If the supply chain itself is also included in the logistics area, support for supplier evaluation or optical quality control of delivered raw materials or components can also be added.
The same can be said for the other areas of the company. However, the above is intended to illustrate how versatile Artificial Intelligence can be used in a wide variety of processes. Which of these areas should be focused on is company-specific and depends, among other things, on the business purpose and model, the general degree of digitization maturity, significant personnel bottlenecks and existing cost structures.
The core aspect of the AI Patterns approach is the identification of opportunities to improve the various business areas through the construction kit of available technologies. In addition to hard factors such as general feasibility based on existing data and influencing variables, “soft” factors such as the acceptance of a system by subsequent users are key success factors. In far too many cases, promising approaches, prototypes or isolated solutions fall by the wayside because they are not taken up and facilitated by the necessary decision-makers and users.
Therefore, the AI Patterns approach starts with a workshop format in which stakeholders from the different business units and decision-making instances work together to identify the potentials. These workshops are accompanied by our AI and digitalization experts. Based on the results of the joint working sessions, they create an implementation roadmap with feasibility and effort indications at the level of granularity possible at this early stage of the study.
The detailing of the cost/benefit or ROI estimate as the basis for prioritizing individual measures is then again carried out in a mutual exchange.
According to the approach described here, it is possible to have a picture of possible and sensible measures for AI introduction in one’s own company after only one week of intensive collaboration, as well as to have prioritized these measures accordingly.
The documentation is composed of the identified applicable AI Patterns for each business area with an explanation of the AI Pattern itself. In addition, the pattern is concretized in the specific company context – the process in which it is used, current weaknesses that can be improved, the goal of a possible implementation, stakeholders and decision makers, etc… This is the basis for a rough ROI estimate and subsequently the prioritization and detailing of the most promising patterns.
This implementation roadmap serves as the basis for all further steps. This includes not only a software-technical implementation and introduction, but also a possible redesign of business processes, roles and task distributions.
We have presented here an insight into our structured approach to AI introduction or penetration of companies. The process is prepared and accompanied by AI experts specific to the company. However, the participation of the relevant stakeholders from the company is also essential, not only to bring in the technical view of the experts, but also to ensure the acceptance and maximum benefit of the developed systems and measures at an early stage.
Contact us to also work out the AI potentials in your company according to the Cloudflight AI Patterns approach and to leverage them for yourself! We look forward to working with you to ensure your success through digitalization and Artificial Intelligence.
 Bitkom Research (2021) – https://www.bitkom.org/sites/default/files/2021-04/bitkom-charts-kunstliche-intelligenz-21-04-2021_final.pdf
 Shana Lynch: Andrew Ng: Why AI Is the New Electricity – http://stanford.io/2mwODQU
 German Sharabok: – Why Tesla won’t use Lidar. And which technology is ideal for self-driving cars https://towardsdatascience.com/why-tesla-wont-use-lidar-57c325ae2ed5