# Artificial intelligence in your business Source: https://www.florianhiess.com/en/expertise/artificial-intelligence/ Language: en AI Brain, leadership, change management, workflows and agents: Florian Hieß connects AI applications with the people and processes in your business. Author: Florian Hieß ## AI becomes effective when people move with it. Implementing AI in a business needs a task and people who can work with the results. An initial tool experiment can happen quickly. For ongoing work, context and responsibility need clarification. My experience includes building an AI Brain for businesses, and leadership and change management when introducing AI. It also includes workflows and agents in marketing. I want to clarify with you which work should be supported and what makes a useful result. Your team needs to know which information the AI may use and where someone reviews its work. This framework turns an idea into an experiment whose benefit can be assessed. ## Which task should AI support? AI should support a clearly described task. In marketing, this might include preparing approved information or preparing research. The suitable task depends on the actual process. I recommend describing the existing process first. What goes in, who handles it and what result is needed? Where do questions or repetitive work arise? Where is specialist judgement required? A limited form of support can then be designed. The experiment gets a purpose and a quality benchmark. That benchmark also includes how much review and reworking remain necessary. This avoids planning solely from a tool's features. We discuss suitable support for your business based on the task and existing responsibilities. ## What I mean by an AI Brain I use AI Brain here as a working term for maintained business context that AI tasks can draw on. It means information about the business and its working practices whose origin and responsibility remain verifiable. My experience includes setting up an AI Brain. AI workflows and agents can use this maintained business context. The suitable technical solution is decided separately for each business. As a planning approach, I would first organise existing knowledge: Which information is approved? Which details change? Who maintains them? Who can access which content? A large collection alone is not yet useful context. For marketing, checked offer descriptions and writing rules might be relevant. These are examples of possible knowledge areas. Whether your initiative needs them depends on the task. ## Shaping leadership and change Leadership and change management belong to AI introduction because people adapt their work while remaining responsible for results. My background includes these responsibilities during AI implementation. I recommend making the reason for implementation understandable. Which work should change, and how does the team recognise a useful contribution? Employees should be able to raise questions and identify limitations in the existing process. Expectations also need clarity. A draft can provide support while still needing careful specialist review. The effort for that review belongs to the process. Who undertakes it should be agreed before the experiment. Leaders must decide which freedom is appropriate and which actions need approval. With a limited starting point, these questions can be discussed through specific work. A shared way of working can emerge that is maintained beyond the initial experiment. ## Introducing workflows and agents sensibly Workflows and agents should follow from a task and its scope for action. A workflow describes a process with defined steps. An agent may also make decisions about further steps within a brief. I work with both in marketing and search. For implementation, I would examine which information is needed and which errors must become visible later. A recurring task may need a clear process. Open-ended research also requires source and stopping rules. Access and approvals are part of the design. May the system only read information? May it save drafts? May it trigger an action? These permissions should fit the task and be explicitly limited. [Agentic Search](https://www.florianhiess.com/en/expertise/agentic-search/) provides more about connecting agents and search tasks. Their use remains a decision about a specific process whose result people must assess. A limited process could consist of reviewing approved information and producing an internal draft. The responsible person then checks whether the details are reproduced correctly. This example illustrates a possible planning approach, not an additionally confirmed client workflow. For such an experiment, I would describe the input alongside the expected result. Which source may be used? Which details must remain? Which change should the system explicitly avoid? A useful brief answers these questions clearly for everyone involved. The output is then compared with the source. If information is missing or the context has changed, document the observation specifically. Revision can address the brief or available context. Specialist approval remains with a person capable of assessing the needed result. This gives the experiment a clear framework for later decisions about continuation and use. ## From a pilot to ongoing work Moving from a pilot to ongoing work requires a deliberate decision about benefit and responsibility. I recommend describing the quality needed and the observation that will determine continuation before the experiment. After the pilot, results and necessary reworking are assessed together. Knowledge maintenance and responsibilities must also fit further use. The guide [Introducing AI into marketing](https://www.florianhiess.com/en/ai/introducing-ai-in-marketing/) explains this approach in more detail. If you'd like to assess a possible starting point, use [contact and an initial consultation](https://www.florianhiess.com/en/contact/). You can first describe today's process. What information arrives, and what result must a responsible person subsequently be able to review? Further reading: [Online Marketing](https://www.florianhiess.com/en/expertise/online-marketing/). ## Frequently asked questions about artificial intelligence ### What is an AI Brain? AI Brain is my working term for maintained business context consisting of approved knowledge, offers and working rules. AI workflows and agents can use it. Information should have a verifiable origin and a responsible person. This definition does not prescribe particular software or architecture; those are chosen separately. ### Where should a business start with AI? A business should start with a specific task whose result people can assess. I recommend describing the existing process and required knowledge. This allows a limited experiment to be planned. Tool selection follows the use case and requirements for access and result review. ### What role does change management play? Change management supports shared changes to work during AI introduction. Employees need an understandable reason and opportunities to raise questions. I recommend clarifying expectations and responsibilities early. My background includes leadership and change management during AI implementation. ### When is an agent more useful than a workflow? An agent can be useful when a brief requires decisions about further steps. A clearly defined process may be handled through a workflow. I recommend first describing the task and permitted actions. An agent's additional decision-making freedom should have a clear purpose and appropriate checkpoints. ### What business information does an AI Brain need? The information an AI Brain needs depends on the intended tasks. Checked offer descriptions and writing rules are possible marketing examples. I recommend recording origin and responsibility for maintenance. Confidential or personal information must not automatically enter the context; permitted access must be clarified beforehand. ### How are suitable AI use cases selected? Suitable AI use cases are selected according to the task, available foundations and a result that can be reviewed. I recommend comparing possible tasks with their current processes. How much effort would result review require? Which access would be needed? A clearly limited experiment can answer these questions before permanently changing a process. ### How are employees involved in AI introduction? Employees should be able to contribute the existing workflow and their quality requirements. I recommend collecting questions when selecting an experiment. Later, experience with reworking should also be documented. This creates a basis for deciding which support the team actually wants to continue using. ### What responsibility remains with leaders? Leaders remain responsible for goals, responsibilities and permitted actions within the business. AI use does not replace these decisions. I recommend defining who reviews results and when approval is required. Time for training and ongoing context maintenance must also be included in planning. ### How can an initial AI pilot be limited? An initial AI pilot can be framed around one task and clearly limited access. Define which information may be used and the expected result. I recommend recording quality criteria and a stopping criterion in advance. This keeps later decisions about continuation or change understandable. ### How is the quality of an AI workflow checked? An AI workflow's quality is checked against the needed result and information used. This includes specialist accuracy and verifiable sources where the task requires them. I recommend also documenting reworking and errors. A convincingly written text alone does not prove a useful workflow. ### What happens after a successful AI pilot? After a successful pilot, decide how knowledge and the workflow will continue to be maintained. Responsibilities and approvals should fit ongoing use. I recommend documenting process changes and checking quality again. This work turns a limited experiment into a responsibly operated process. ### Which privacy and access questions need clarification first? Before an AI experiment, clarify permitted data, recipients and access rights. Handling confidential information also needs an explicit decision. I recommend discussing these questions with the responsible people within the business. General marketing advice does not replace assessment of specific privacy and security requirements.