Engine for higher productivity and innovation.
Provided that the mindset of the employees is right.
synopsis
Large and medium-sized companies are now more concerned with AI applications than ever before. Failure to do so can lead to a drastic deterioration in a company's competitive position in the longer term. Namely when, with the support of AI technologies, potential to reduce costs or increase sales remains unused. There are numerous examples of AI applications in the manufacturing industry and in multi-level sales that have led to measurable significant performance improvements. In most cases, AI technology is not used autonomously, but as a tool or aid for employees. AI tools are therefore only effective if employees trust the results of AI and therefore change their processes, working methods and decision-making behavior. In this chapter, we show practical examples and provide advice on what a company should pay attention to when choosing AI applications, which success factors to consider and how to change the mindset of employees in order to build trust in AI.
AI as a puzzle piece in a company's value chain
AI is an increasingly important part of the value chain as it increases efficiency and accelerates innovation and business processes. By integrating AI into various phases of value creation, data is made usable for optimization purposes. In production, AI improves quality control and increases productivity. In sales and marketing, AI makes it possible to identify sales potential, address customers in a personalized way and to place the right offer at the right time. Customer service uses AI-based tools to solve problems faster and more efficiently. The integration of AI into the value chain also supports the development of new products and services or in determining price positioning. By automating repetitive tasks, AI contributes to reducing costs and increasing the scalability of companies by achieving greater independence from professionals.
AI is revolutionizing the value chain in the manufacturing industry by enabling companies to respond faster to market demands and create sustainable added value. AI is a key enabler for sustainably strengthening a company's competitiveness.
Sales Recommender
Existing customer data is analyzed using machine learning algorithms. Algorithms create a purchase recommendation from customer characteristics such as industry affiliation or company size, previous buying behavior and the product reviews of the respective customer and similar customers as well as the characteristics of the various products. You know how it works from online shops: “You might also be interested in this product” or “Other customers also bought it.” These so-called recommender systems also work in B2B sales, provided that sufficient relevant data is available in the CRM and ERP systems. A classic use case for recommender tools in mechanical engineering is the after-sales area to generate recommendations for selling spare parts.
Offer Automation
With the help of AI, the offer process can be automated by algorithms analyzing data (e.g. historical offers, price lists, discount behavior, competitor prices), making price recommendations, creating personalized offers, and enabling automated communication with the customer. AI thus increases efficiency and speed in preparing offers.
Lead Crawler
On the basis of manually or automatically created customer profiles (e.g. based on existing customer data), algorithms search for new customers in internal and external data. Leads that could have business potential are passed on to sales. Sales qualifies whether there is actual sales potential or not. This feedback is fed back to the AI, which enables it to learn from the feedback and continuously optimize the search process.
Churn Prevention
Algorithms identify customers who are at increased risk of not buying or migrating to the competition in the foreseeable future. The AI determines this risk on the basis of customer data such as order behavior, payment history, offer values or communication data for sentiment analysis (sentiment). Salespeople get access to a dashboard that provides an overview of vulnerable customers. In this way, measures can be taken proactively.
Generative AI (e.g. voice assistants)
Voice assistants such as Chat-GPT from Open AI, Google Bard or open-source solutions can be used in a wide variety of areas of business. Basically, language assistants make sense in all knowledge-intensive organizational units, such as marketing, customer service, research and development, purchasing, HR or sales. The advantage of large language models (LLMs) is that they are already pre-trained and can be used productively for a company in a relatively short period of time. The greatest benefits are achieved with voice assistants when external as well as internal data sources are integrated. This data can be available in unstructured form (e.g. product descriptions as PDF, emails, websites) and processed by voice assistants.
Machine Parameter Optimization
In production, algorithms are used to automatically adjust machine parameters in order to generate optimal production output in terms of product quality and performance. Parameter optimizer tools are also used in product development, for example to develop recipes for chemical products more quickly. This involves replacing time-consuming physical experiments with computer simulations.
Predictive maintenance
For machine maintenance, algorithms can be used to monitor the status of machines in real time (condition monitoring). Continuous data analysis and the use of algorithms predict maintenance requirements for machines or individual components in order to prevent unplanned downtimes. This form of proactive maintenance increases the net uptime of machines and reduces maintenance costs by carrying out maintenance exactly when it is needed, rather than after a fixed maintenance interval.
Quality Control/Anomaly Detection (Computer Vision)
Computer vision can be used to monitor the quality of the products during production. AI-based quality control through image recognition uses so-called neural networks to recognize patterns in images. Through AI training with image-based sample data, the AI detects irregularities (anomalies) in the products, identifies defects, optical defects or faulty dimensions. This automated image analysis enables quick and accurate quality assessment and helps production employees to continuously optimize the production process and reduce waste.
Sales/Supply Forecasting
Algorithms can recognize patterns in historical sales data in order to generate a sales forecast. In addition to historical sales data, people usually use offers, sales opportunities, market statistics or economic data to achieve greater forecast accuracy. By integrating relevant data sources, very good forecasting quality can be achieved. In practice, AI often produces better forecasts than those created by employees. The use of AI helps to significantly reduce planning costs for sales, to increase the accuracy of forecasts and thus to make planning data more useful for downstream departments such as procurement, production or logistics. Especially in cases where a great deal of data is available (e.g. a very large number of different products or product segments as well as a variety of external and internal data sources) and where it is unknown a priori how this data depends on each other, AI-based forecasting models can also add value compared to conventional statistical approaches.
Industrial Embedded AI
Integrated AI technology in machines and in products in general enables the automated control and optimization of production machines and the support of production employees to make the right decisions. Embedded AI is used to increase productivity in the production process. Examples of applications for embedded AI can generally be found in robotics, OEE (Overall Equipment Effectiveness) optimization, predictive maintenance or in voice assistants for machine operators.
Success factors for AI deployments
Using AI to optimize business processes or develop new business models is on the agenda of many companies. Nevertheless, it is important to be aware of the challenges of AI projects. In this regard, AI projects are no different from other software implementation projects. AI and data science projects are often experimental for companies that are just starting out in order to find out to what extent the existing data can be used for AI at all and whether AI makes sense at all. The first AI project that a company implements is of decisive importance for the future role of AI in the company. If the first project is a “fail,” it is difficult to motivate management and employees to start another AI project. In this way, a company's AI journey could come to an abrupt end. To avoid this, a systematic analysis of AI ideas, potentials and use cases is required. The following factors are decisive for the success of AI projects.
Success factor #1
The company must have the right data
Without data, you don't have a data science project. Without the right data, there is no AI. It can also be difficult to collect, create, or procure the necessary data externally.
Even though you have access to the data, there are still a few challenges to overcome:
- How and where is the data saved?
- Is the underlying data available in sufficiently good quality?
- Can the data be cleaned?
- Can the data be used ethically and legally for the intended use case?
- Are there internal employees who understand the combination of data and business processes?
The topic of data management is one of the central challenges in AI projects. Here are a few recommendations:
- Establishment of internal guidelines on how to collect and use data to avoid the GIGO effect (garbage in — garbage out).
- Never assume that data is of high quality.
- Ensure that internal or external employees with the necessary skills in data science, advanced analytics and machine learning are deployed.
Success factor #2
Deploy the right people with the right skills
Finding, hiring, and retaining tech talent is never easy. And it's not just about finding qualified employees. You also have to find the right combination of employees. The days when a data scientist implements an entire AI project alone are over.
The implementation of AI solutions — from brainstorming to providing the solution to productive use — requires a wide range of skills and roles to be successful.
Success factor #3
Selecting the right AI use case with a crystal-clear objective
Have you ever started a project with a vague vision? Or a project with a clearly defined goal that is not realistic or does not provide any meaningful added value?
Some companies hire well-trained data scientists without having defined a clear direction. The new hires then emerge from a “research hole” lasting several weeks, only to discover that they had misunderstood the target variable, making the analyses irrelevant.
To minimize this risk, start your project the right way and ask the right questions before you start the project. You have to dive deep into a problem to really understand the underlying problem that needs to be solved.
Success factor #4
Think big — Start small — Deliver fast
The more uncertain the ROI of a software investment, the lower the chance that management will endorse the project. This also applies to AI projects. Decision-makers in companies that decide on AI investments shy away from large (risky) investments. In AI projects in particular, the stipulation “look for quick wins” is often given. In other words, think big — start small — deliver fast.
Think Big Determine with management which division of the company should be looking for AI potential (e.g. production, product development, SCM, sales/marketing). You will conduct Envision workshops with a few stakeholders from the selected department. Ideas are developed together and evaluated with internal/external experts. This is the basis for selecting the first use case and creating a road map to align the AI initiative.
Start Small Start with a use case that keeps resources, budget and project duration manageable. Look for “low hanging fruits” — especially at the start of the AI journey. The selected use case should be of low complexity and the business must recognize clear, significant added value.
Deliver Fast Choose a use case that provides initial results relatively promptly. In particular, the database is decisive here. When it becomes clear from the first data screenings that you have to invest a few months in data preparation, then the project is not exactly spurred on. Quick results motivate project participants to pursue the AI project with high priority.
Success factor #5
Implementing an AI solution doesn't mark the end of an AI project
According to the traditional definition, an IT project ends when the planned scope of services has been delivered. This also applies to AI projects. However, you must be aware that the ongoing operation of AI solutions requires special attention. AI models must be validated over a certain period of time and, if necessary, adapted and improved. This does not (yet) happen independently by AI. A team is needed for continuous development in order to maintain functionality and create sustainable added value for the specialist areas.
Success factor #6
The right project management method
Companies still like to use traditional project management methods (such as waterfall), which often lead to inefficient exchange of information and incorrect expectations among specialist departments. You often have a goal that you want to achieve — for example, higher machine productivity with predictive maintenance. However, it is not even known in advance whether the volume and quality of data are reasonable and sufficient to achieve this goal. Data science projects are therefore experimental at the outset. Work is carried out in the laboratory. When reasonable results are then achieved, it is possible to discuss how they can be used in the right place as far as possible and made usable. The expectations of the departments must be defined as measurably and realistically as possible. We should talk about technical and human interfaces again and again in order to use existing resources and adapt the technical specifications to the willingness and capabilities of the existing infrastructure and users. In project management, you should rely on agile project methods and transparent, open communication. This has a positive effect on change management.
Success factor #7
Include ethical and legal aspects
AI models optimize business processes without regard to basic social or ethical principles. In some cases, AI is a double-edged sword that can lead to ethical or legal problems.
Examples of data science that led to ethical and legal problems:
- Health risk scoring:
Healthcare providers in the U.S. used a health risk score to determine whether they should provide proactive health care treatment to every patient. Basically, it's a good idea. However, the model used healthcare costs to assess health risks. And because dark-skinned patients tended to have lower healthcare costs, this health risk score prioritized light-skinned patients for proactive treatment.
- Creating voter profiles on Facebook:
Cambridge Analytica used data from millions of Facebook profiles without user approval to create psychological profiles of voters. Facebook was then fined five billion dollars and Cambridge Analytica went bankrupt.
It is important to assess potential ethical and legal issues in advance and during the AI period of use.
Success factor #8
“Business transformation” only takes place in a corporate culture that is open to change.
Everyone knows the saying “Culture eats strategy for breakfast.” The best use cases, the best strategy and the most experienced team are of little use if the corporate culture does not allow innovation. We need an “open-minded” culture in which employees are curious to learn new things.
Executives regularly report that cultural challenges — not technological ones — are the biggest obstacle to the successful introduction of AI initiatives and thus represent an obstacle to optimizing business processes.
Remember that not all project participants are on board with full conviction. AI is something new, unknown, whose result is not 100% clear. AI may be able to replace work and even job profiles that people identify with. Trust in this topic develops over time. A high degree of transparency and open communication is therefore required. Curiosity and interest often increase during the project and thus also knowledge and understanding of data science and AI. These are important basic requirements if AI is to become part of the daily work of employees.
In particular, include employees who are not yet comfortable with the field of AI. Start with short, simple training and coaching sessions and emphasize the importance of data-driven decision making. In this way, you help everyone involved go through the change process.
Success factor #9
AI solutions need acceptance and trust
Data scientists often produce the best AI models, but employees don't want to use the models as a tool. Why
Scoring models, such as predicting the likelihood of leaving customers (churn prediction), are classic.
The problem with aggregate scoring models is that employees don't understand how the scores are formed. The score is a black box. As a result, users of AI models do not gain confidence and find it difficult to derive the right measures. What action should an employee take if the churn score is 87 out of 100? In other words, a highly vulnerable customer.
Acceptance among users can only be achieved by being able to look into this black box. An AI model should be explained to users and users must have the know-how to understand how AI basically works. AI recommendations and forecasts should be explained. It's called “explainable AI.” Understanding also means accepting means trusting.
Every company must intensively address its own AI potential. Analyze and evaluate these objectively and, if the potential is available, get started with the right use case and a qualified team. Think big — start small and deliver fast — but start.
Methodology for choosing the best AI use case
Choosing the right AI use case is critical as it determines the overall success of an implementation. A suitable use case must be closely linked to the company's goals and challenges. An incorrect selection can lead to significant waste of resources and destroy trust in AI technology. Acceptance within the company is another decisive factor. A well-chosen use case makes it easier to convince stakeholders because it shows clear benefits and concrete effects. Starting with the right AI use case promotes the willingness of employees to change processes and working methods and helps to reduce fears of change.
Choosing the right use case is particularly important at the beginning of the AI journey. You have to search for the “low hanging fruits.” Those use cases that generate high added value and can be implemented in a short period of time. How do you find “low hanging fruits”?
If you ask employees for ideas about what they could use AI in the company for, they get lots of ideas. The selection should absolutely follow a clear, comprehensible system so that decisions are comprehensible. Each use case idea should be evaluated based on the following three criteria:
- Feasibility
Is the implementation technically feasible? - Complexity
Is the implementation resource-intensive and requires the involvement of several business units or external partners? - Business Value
What measurable added value does the company have?
Assessing the business value of an AI project is sometimes a challenge. Even though an evaluation may be difficult in some cases and is sometimes associated with vague assumptions, one should not refrain from a numerical evaluation. If you optimize in the production sector, an evaluation based on cost savings or an increase in productivity is usually easy. Improvement in quality control can also be measured using the reject rate. It becomes more difficult on the sales side when you try to quantify sales growth potential. The aim is for the evaluation to be carried out with internal experts and thus to initiate a constructive dialogue. By critically examining the benefit assessment, you develop a uniform view and a clear opinion about an AI idea. If internal experts are not 100% convinced and motivated by an AI idea, you shouldn't implement it. Selecting a suitable use case is crucial for success in the AI journey. The systematic selection and involvement of relevant stakeholders ensure that AI is not only technically successful, but above all meets the company's strategic and business requirements. This helps to effectively utilize investments in AI and pave the way for sustainable integration into business processes.
The Symbiosis of Humans and AI
In critical areas of the company, such as production, the implementation of automated AI-based solutions is a major technical, economic and legal challenge, which is why people usually make the final decisions. In order to examine a claim, it must be documented how the decisions that led to it were made. This is very difficult with AI systems that behave like a black box. If you look at the strengths of people — such as flexibility or creativity — and those of AI — processing large amounts of data, consistent performance and accuracy in repetitive tasks — it quickly becomes clear that the maximum benefits of AI can be achieved by working with people. People with their experience and strengths will therefore continue to play an important role in the future, whether as decision makers or as employees who monitor, train, develop and optimize an AI system. The human-AI interface or user interface plays an important role in enabling the described collaboration and supporting decisions through well-explained forecasts or recommendations from the AI. The right choice and design of the user interface is important for acceptance. In addition to classic visualizations in the form of text or graphics, voice assistants such as chat GPT offer significant potential for the future.
By detecting abnormalities such as vibration, temperature rise, pressure drop or loss of power consumption, failures of machines or individual components or tools are predicted. Maintenance employees are informed via a web-based application at the control station. The decision as to when maintenance is actually carried out is still up to the employee. With the help of this application, maintenance work can be identified and scheduled at an early stage in order to avoid an unplanned downtime, which can cause considerable economic damage.
In terms of good explanability, the AI tool shows which anomalies occur in the process and which (correlating) factors are responsible for the maintenance forecast. The system indicates a probability that an outage could occur within the next 6, 12, or 24 hours. The system also shows which similar events took place in the past and whether the system was right or wrong in these cases. An AI system is only as good as the accuracy of the predictions are and whether users have confidence in the AI application or not. The way AI recommendations are visualized in the user interface plays an important role in building trust.
The AI system helps employees learn which anomalies lead to machine failures and how long maintenance cycles last and which factors influence them. On the other hand, employees can also use input mechanisms to contribute their experience and discontinue maintenance if, for example, they expect an imminent failure due to strange noises generated by a machine. You can give the system feedback as to whether a failure actually occurred within a certain period of time or whether maintenance carried out has led to the resolution of an anomaly. The adaptability of the AI system is also important. In other words, adapting to changing production conditions and integration with other IT systems (e.g. ERP). In order to develop AI systems with a high level of acceptance, the AI model, the business process and the user interface must be coordinated. This requires an interdisciplinary team that contributes expert knowledge in all of the above areas. Usability experts play an important role. They help to integrate the AI application and the right user interface into the workflow.
Conclusion
The successful introduction of AI in manufacturing companies requires not only technological implementations, but also a strategic approach that takes into account the mindset of employees. The integration of AI into the value chain offers potential to increase the efficiency, innovative strength and competitiveness of a company.
To this day, AI is generally used as a tool for employees, and its effectiveness depends heavily on employee confidence in the results. It also requires employees to be willing to adapt their working methods and follow AI recommendations. Implementing fully automated, autonomously functioning AI solutions in critical areas of the company is not the goal in most cases. Human decisions often remain essential due to technical, economic and legal complexity.
Instead, AI solutions should be meaningfully integrated into employees' work and decision-making processes. An AI acts as an assistant to enable quick and correct decisions. The user interface plays a decisive role here, especially in the context of “explainable AI.” An AI system gains trust when it provides clear and precise recommendations or forecasts that are understandable to employees. An AI system can only be used if employees trust it.
.jpeg)
