By Dr. Alexander Deicke
A customer wants to know whether the new machine produces less CO₂ than the previous model. The AI assistant generates a sentence for the quote in seconds: “20 percent fewer CO₂ emissions.” The figure is actually listed in a document. However, that document refers to electricity consumption under specific test conditions.
This example is fictional. The question behind it , however, is not: Who verifies whether a plausible phrasing has become a statement that the company can substantiate and stand behind?
This question was on my mind when I contributed to the anthology *AI and Sustainability in Sales *. After all, the value of an AI system isn’t determined solely by how quickly it generates a quote. It’s also determined by whether the sales team knows what data it’s allowed to use, what the statements are based on, and who decides when to send them out.
For an initial draft quote, the desired product variant, the quantity, and a few technical requirements may be sufficient. Therefore, the complete meeting minutes, contact history, and internal notes on price negotiations do not necessarily need to be included in the AI system.
For me , the evaluation begins with a simple question: What information is the tool actually missing if I omit it? A system connected to the CRM should not be able to access all customer data for a single transaction simply because it is technically possible.
In B2B sales as well, contact information and meeting notes may contain personal data. In such cases, the processing requires a legal basis and must be limited to the data necessary for the purpose. Separately, it must be clarified what confidential information—such as purchasing terms or unpublished product data—the company is permitted to disclose to the vendor.
An environmental benefit takes concrete form in sales. A figure in the product data sheet becomes an answer to a customer’s question—and may even serve as a deciding factor in the purchasing decision. This makes it all the more important that the statement does not lose its original context as it makes its way into the quote.
In the case of the machine example, I would first consult the test report: Which models were compared? Under what conditions were the measurements taken? Does the value refer to power consumption during operation or to actual calculated emissions? If only power consumption is documented, the statement is limited to power consumption. A separate basis is needed for a more detailed statement on CO₂ emissions.
This is also legally relevant. Misleading statements about a product’s environmental characteristics may be deemed unfair under Section 5 of the Unfair Competition Act (UWG) if they are likely to influence a business decision. This applies not only to statements made to consumers.
AI can help you find the right source and phrase its content in a way that’s easy to understand. However, it can’t turn a test report into evidence that doesn’t exist. If reliable information is missing, the team should be able to say, “We’ll look into that and get back to you.” That, too, is a useful sales response.
The sales department is familiar with customer requirements. The product team can interpret technical specifications. The procurement department is familiar with the agreements with the AI provider, and the IT department knows how to integrate it into the systems. The data protection and legal departments can assess the requirements associated with the specific use case.
We need these perspectives. But we also need someone to take responsibility for the entire process: Who determines which sources the AI draws from? Who changes the rules when a product version is phased out? Who can stop automated dispatch if faulty offers are detected?
That doesn’t necessarily mean that every standard letter has to go through the legal department. A draft appointment confirmation can be handled differently than an offer that includes a special discount, a delivery commitment, and an emissions comparison. The key is that the person approving the offer knows which details they can verify themselves and when they need to seek expert advice.
Instead of starting the next round of approvals by asking which tool is best, I would choose a specific proposal and walk through the process based on that:
If a question remains unanswered, that’s no reason to scrap the entire project. It simply highlights what still needs to be clarified before broader implementation. A quick draft is a good interim result. A quick draft that has been thoroughly reviewed is a result ready for sales.
The anthology *AI and Sustainability in Sales: Taking Responsibility, Recognizing Opportunities, Shaping Transformation* brings together diverse perspectives on the use of AI—from lead generation and quote preparation to customer retention. It is edited by Almaz Andezion, Achim Hohorst, Christoph Jacob, and Michael Pankow. The print edition is scheduled to be published in February 2027 by Springer Gabler. View the book on Springer’s website.
I contributed an article on the legal and organizational foundations of AI implementation. In it, I provide an overview of European and international regulatory frameworks. My primary focus is on how companies can integrate data protection, risk management, and responsibilities in a way that ensures their rules function effectively in day-to-day sales operations.
You can tell whether this has been successful by a customer’s question: “Where does this number come from?” The best answer to that isn’t “from AI,” but rather: “Here’s the proof.”