The AI Agent –
a Surefire Success?

Dr. Dana Ruiter works as a Senior Consultant in an interdisciplinary team of AI specialists at Porsche Consulting. She holds a Ph.D. in Computational Linguistics and specializes in Artificial Intelligence, Data Analytics, and Natural Language Processing. In an interview with Porsche Consulting Magazine, she discusses which use cases are best suited for AI agents in business, how to ensure successful implementation, why excellent data quality is critical – and how the management consultancy Porsche Consulting supports its clients throughout this journey.

Dr. Dana Ruiter: Many companies rely on AI agents. But they should be aware of this: Not every use case is well served by an AI agent. It is important to clarify what is really needed before implementation. One example: We had a client who wanted to equip their sales team with data-driven recommendations for upselling customers. The client had envisioned an AI agent as a chatbot that sales employees could ask about upselling potential. But isn’t it better to calculate the potential in a prediction model and then send it automatically to the relevant sales employee? This avoids the question-and-answer game and the recommendations are also of a higher quality thanks to the prediction model.

AI agents are always useful when relevant information needs to be identified from a large number of documents and processed according to certain rules. The following also applies: The quality of an AI agent is heavily dependent on the quality of the data. So when deciding whether to use an agent, it must be clear what type of data is required. The data must then be stored centrally for the AI agent. Ideally, the data should be cleaned up beforehand so that only up-to-date data with well-maintained metadata are available.

Both! At Porsche Consulting, we pursue a “broad and deep” approach. This means: In order to use AI agents effectively in companies, it makes sense to implement smaller, selective AI agents at the process level as well as AI agents for broad use in the company as a whole.

The “deep” agents require a clear analysis of the process. This is the only way to understand where the AI agent can provide the greatest gain in efficiency or for which decisions it can provide useful support. It must then be clearly understood which data pools are needed to implement the agent – and how the information extracted from them is processed and made available to the end user. This can be done quickly. We implemented several deep agents for one client within a few weeks. Among other things, they help customer success managers to prepare for and follow up on customer meetings.

Things look trickier for the “broad” agent. Its aim is to give the wider workforce quick access to relevant internal data. In one project, we implemented a broad agent that not only provides internal information, but also helps employees send emails, identify free appointment slots, set appointments, and perform simple analyses via company dashboards. All of this requires a great deal of effort in terms of data preparation, data governance, and training employees in the use of the tool. Such broad agents are complex to implement and involve a high investment. However, they are an important pillar if a company wants to develop in the direction of “AI first” and establish AI as the basis for all processes and decisions.

The most serious errors lie in incomplete, poor quality data, and the introduction of AI agents where a different method would be more effective. Another problem is that many people want an “all-rounder”. Due to the hype surrounding AI, expectations are high. And the required investment in time and money is often underestimated. It makes sense to start with small pilot projects and then draw up an implementation plan for the further use of AI agents. This leads to the first efficiency levers and success stories and companies can gradually move closer to their vision. In this way, organizations remain dynamic and can adapt quickly if new developments in this rapidly growing field shake things up again in the future.

The entire staff is very important because they have to work with the AI agents in the end. A change management plan with a focus on training and empowering the workforce is crucial here. And in the best-case scenario, the employees themselves become developers. As a first step, the workforce should learn how to use artificial intelligence sensibly: What do you need to look out for to keep company data secure? What kind of questions can AI help with? How do I formulate the best prompts? The next step is to train motivated employees to become developers of AI agents themselves. They learn to recognize the potential for AI agents in their own processes and working methods and then take the initiative to implement meaningful agents on their own and share them with other colleagues.

In the white-collar sector, we are at around 30 percent, depending on the industry and process. But I’m not a fan of such rough figures, as the increase in efficiency always depends on the process in question. For example, we have already implemented AI agents where the respective department estimated the increase in efficiency to be 100 percent – in this case, it was about validating requests for quotations in the engineering area.

Expandable is a good keyword: To avoid long specifications, broad agents should always be implemented in phases. You start with what is termed a “minimum viable product” and then gradually add new functionalities as required. To achieve this, it is of course important that the broad agent is understood as a modular construct from the outset. Not only to enable future expansions. But also so that the agent itself can evolve with the rapidly developing AI environment. For example, by making it easy to replace the language model at the core of the agent when a newer, better language model comes onto the market.

They are relevant for all industries, but especially for the white-collar sectors. In other words, where people sit at computers a lot and work with data and documents: Legal departments, IT development, marketing, HR, and sales can benefit greatly from this. But special solutions in the engineering area or on the shop floor can also have an impact. For one client, for example, we used a language model to cluster frequently occurring problems on the machines in production, prioritize them and then automatically make them available to the shift supervisors. This creates transparency and helps them to train employees on the most common types of errors using best practices.

In short: the transfer from the process to the agent. We look at the processes and identify potential. And use this as the basis for evaluating which AI solution would be useful. The solution is then closely scrutinized with the client’s respective experts: How should the tool be used? What data is required? Which criteria are particularly important during processing? What should the results look like so that employees experience the greatest added value? As soon as the user story and the technical and non-technical requirements are clear, you can get started. The AI agent is brought to life in short sub-projects: either first as a proof of concept, or – in the case of smaller AI agents – directly as a minimum viable product. It is important for the AI experts at Porsche Consulting to involve the workforce. As part of dedicated training courses, employees learn how to use the new tools – or are empowered to implement their own small AI agents in the future.

In the industrial goods, consumer goods, software as a service, aerospace, and automotive sectors – also in an international context.

As with all disruptive developments, such as the internet, social media, and now artificial intelligence: If you don’t move with the times, the times will pass you by. And with a technology as rapidly developing as AI, it is important to understand it and use it effectively in the right cases. This has to be done proactively, because AI agents are not a surefire success. This keeps you competitive. It is no longer just about the profitability or competitiveness of individual companies – it is about the future viability of industry and the economy.