Successfully implementing AI in Business Processes
AI tools can make business processes significantly more efficient. However, their value does not stem from the tool alone, but rather from the interplay of clearly defined processes, reliable data, and people who can evaluate the results from a technical perspective.
AI in Business Processes:
Three Factors for sustainable Success
Artificial intelligence has made its way into businesses. It is already being used to assist with research, translations, text analysis, minute-taking, and processing large volumes of information. The automation of business processes—for example, in quality management (QM)—ranging from the creation of quality-related documents to complaint management—is now technically feasible.
Nevertheless, getting started quickly with an AI tool does not automatically lead to a reliable result. The key management question is therefore not, “Which tool can we use?” but rather, “For which process is AI suitable, what data is available, and who bears professional responsibility for the result?”
AI in Business:
Between Hype and lasting Benefits
The public debate often gives the impression that AI systems can solve complex tasks almost entirely on their own.
In practical business settings, the picture is more nuanced: AI can structure content, recognize patterns, generate suggestions, and speed up repetitive tasks. Whether these results are useful, however, depends largely on the context in which they are applied.
Especially in quality-critical or regulated processes, a result that seems plausible at first glance is not sufficient. It must be technically correct, traceable, reproducible, and suitable for its intended purpose. This is particularly true when further decisions, documents, or process steps are derived from the AI result.
Our practical experience: Professional-looking AI-generated texts are not automatically factually accurate
Our own experience shows just how quickly the capabilities of AI are overestimated. In an internal DHC project, a team of seven investigated how AI tools can assist in the creation of standard operating procedures for change management.
The participants had varying levels of qualifications and experience—ranging from back-office staff to a principal consultant with many years of experience in creating audit-ready GxP documents. The choice of prompts was left up to the employees. The initial result was impressive: the structure, cover page, outline, and chapter titles appeared logical and professional regardless of the employees’ educational backgrounds. However, a technical review revealed significant weaknesses. The further a person was removed from the specific task and the underlying process, the greater the need for revision. Even the version created with the support of an expert still required significant revision. The key difference, however, was that technical gaps could be identified more quickly and addressed more effectively.
A rough internal estimate indicated—depending on the task and source material—a potential savings of about 10 to 30 percent. However, this potential can only be realized if review, revision, and accountability are factored into the plan from the very beginning.
The key insight
“AI does not automatically produce quality. Rather, it amplifies the impact of the framework conditions in which it is used: a clear process, suitable data, and a robust expert evaluation.”
Why “just getting started” Isn’t an AI Strategy
A small pilot project can be useful. However, uncontrolled trial and error is no substitute for a well-thought-out plan. Anyone who wants to integrate AI into a business process must first determine which task is to be automated or supported, what the desired outcome looks like, and what consequences errors might have.
The introduction of innovative technologies thus follows the well-known principles of successful projects. Agility means testing assumptions early on and learning from the results. It does not mean foregoing an understanding of processes, quality criteria, governance, or responsibilities. A poorly defined process does not automatically improve with AI—if anything, it is merely executed faster and with less transparency.
Three Factors for Successful AI Projects
1. Process Knowledge: First Understand the Workflow, Then Automate It
The most important prerequisite is a precise understanding of the process. This includes not only the nominal steps, but also variations, exceptions, roles, decisions, and the required input data. Before implementing AI support, the following should therefore be clearly defined:- What process steps exist—and in what order?
- What inputs are required, and what quality must they meet?
- What outcome can be expected at each step?
- What variations and deviations might occur in practice?
- Who (as a person) and with what qualifications decides on the quality of AI-generated content, approves it, or escalates it as appropriate?
2. Data Quality: The Foundation for Reliable Results
AI applications rely on data. This applies equally to generative AI and machine learning applications—even if the specific technical use of the data differs. What matters is not only the volume of data, but also its suitability for the respective use case. The following criteria are particularly relevant for assessing data quality:- Accuracy: Is the content factually correct and free of relevant errors?
- Completeness and consistency: Does the data sufficiently cover the process, and are there no contradictions within it?
- Traceability: Are the origin, selection, processing, versioning, and use documented?
- Robustness: Does the system deliver usable results even in the face of realistic deviations and changed inputs?
3. Subject Matter Expertise: People Remain Responsible for Quality
Successful AI solutions require experienced subject matter experts. They do not need to understand every technical detail of the model. However, they must be able to assess whether a result is correct, complete, and appropriate for the specific process. To do this, they need an understanding of the causal relationships between the information provided and the generated result. In critical processes, it must also be clearly defined when an AI result is merely a suggestion, when human review is required, and who makes the final decision. AI can support responsibility, but it cannot assume it. The use of AI therefore requires both general and specialized knowledge, but above all, a wealth of practical experience.The right team is key to a project's success
An AI project should not consist solely of technical experts. Nor is a purely technical concept sufficient without expertise in data and technology. A deliberately balanced mix of at least three perspectives has proven effective:
- Process specialists who are familiar with the workflows, roles, and variations.
- Data analysts or data engineers who evaluate and process raw data and transform it into usable information.
- Subject matter experts who assess the quality and usability of AI results.
Depending on the risk and area of application, other roles may be added—such as IT architecture, information security, data protection, quality management, regulatory affairs, or validation. It is important that these perspectives be incorporated into the design not sequentially, but collectively. Their methods and priorities may differ. That is precisely why the project needs clear objectives, decision-making criteria, and responsibilities.
From Idea to viable AI Application: A pragmatic Approach
Für Entscheider und Fachverantwortliche empfiehlt sich ein schrittweises Vorgehen, das Nutzen und Risiko gemeinsam betrachtet:
- Use Case auswählen: Identifizieren Sie einen konkreten Prozess mit erkennbarem Nutzen, klarer Abgrenzung und messbarem Ergebnis.
- Prozess und Risiken beschreiben: Dokumentieren Sie Ablauf, Varianten, Inputs, Outputs, Rollen und mögliche Fehlerfolgen.
- Daten bewerten: Prüfen Sie Verfügbarkeit, Qualität, Herkunft, Zugriff, Schutzbedarf und Repräsentativität der Daten.
- Qualitätskriterien festlegen: Definieren Sie vor dem Pilot, wann ein Ergebnis als richtig, vollständig, nachvollziehbar und verwendbar gilt.
- Human Oversight gestalten: Legen Sie Prüf-, Freigabe- und Eskalationspunkte sowie Verantwortlichkeiten fest.
- Pilot mit realistischen Fällen durchführen: Testen Sie nicht nur Idealbeispiele, sondern auch typische Varianten und Grenzfälle.
- Nutzen und Aufwand messen: Bewerten Sie Qualität, Durchlaufzeit, Fehlerquote, Akzeptanz und erforderlichen Nachbearbeitungsaufwand.
- Put into operation: Regeln Sie Schulung, Monitoring, Änderungsmanagement, Dokumentation und kontinuierliche Verbesserung.
What AI can do —and what It can't
KI-Werkzeuge eignen sich besonders, wenn grosse Mengen strukturierter oder unstrukturierter Informationen verarbeitet, Inhalte klassifiziert, Entwürfe erstellt oder wiederkehrende Arbeitsschritte unterstützt werden sollen. Beispiele sind die Vorstrukturierung von Reklamationen, die Suche in Wissensbeständen, die Zusammenfassung von Dokumenten oder die Vorbereitung von Protokollen.
Nicht geeignet ist die Vorstellung, KI könne ohne fachliche Kontrolle zuverlässig über die Prozessqualität entscheiden. Plausible Formulierungen sind kein Beleg für sachliche Richtigkeit. In GxP-relevanten Anwendungen sind deshalb zusätzlich eine risikobasierte Bewertung, geeignete Governance und – abhängig vom bestimmungsgemässen Einsatz – die erforderliche Validierung beziehungsweise Qualifizierung zu berücksichtigen.
Conclusion: AI is valuable when it fits into the process
Making the most of AI tools in business processes is less a matter of having the most spectacular tool and more a matter of the quality of the overall system. Companies achieve sustainable benefits when they understand the process, have a firm grasp of the data, and secure the necessary subject matter expertise.
For management, this means: AI projects require clear goals, a realistic assessment of their benefits, and a governance framework that makes quality and accountability transparent.
For subject matter experts, this means: AI will not render their experience obsolete—it will become the decisive factor in determining whether a plausible draft can be turned into a robust final product.
Evaluate the right AI Use Case now
Are you exploring how AI can be applied to your operational or quality-related processes? DHC helps companies identify suitable use cases, analyze processes and data, and conduct risk-based assessments and validation of AI-supported systems.
Frequently Asked Questions about AI in Business Processes
What are the requirements for AI in business processes?
At least one clearly defined process, appropriate and traceable data, defined quality criteria, and experts who can evaluate AI results. For critical applications, governance, security, data protection, and, where applicable, validation requirements must also be met.
Why is process knowledge so important for AI projects?
Prozesswissen zeigt, welche Eingaben benötigt werden, welche Varianten auftreten und welches Ergebnis tatsächlich erwartet wird. Ohne dieses Wissen lässt sich ein formal überzeugendes, aber fachlich falsches Ergebnis nur schwer erkennen.
Does a business user need to understand the AI model from a technical standpoint?
Not every technical detail. However, he or she must understand the limitations of the application, be able to review the results from a technical perspective, and understand the relationships between inputs and outputs.
How can AI be used in regulated processes?
Using a risk-based approach: define the intended use, assess risks, establish responsibilities, implement data and access controls, monitor results, and maintain the necessary documentation for the specific use.
How should companies get started on an AI project?
With a clearly defined use case whose benefits and quality of results are measurable. A joint workshop bringing together process, data, and domain experts quickly establishes a solid basis for decision-making.