How wasting tokens becomes an ESG risk

The hidden AI cost: Why today's token waste destroys tomorrow's business value

Scientific studies on the energy efficiency of LLMs show that energy consumption per token increases linearly to exponentially with the length of the prompt. This is particularly noticeable with redundant sentences that your system processes. They cost twice as much both in money and electricity. Are companies aware of this issue? Not at all! Even though they usually insist, “Of course we train our people on how to write prompts!” Today, this isn’t such a big deal because the costs of using AI are still relatively low.

But there’s one thing every user needs to be aware of: Without a clean database in the background, the introduction of AI-based tools won’t do any good in the long run rather the opposite. Imagine a child has an unlimited amount of allowance and can do whatever they want with it. Even though there are certainly exceptions, it’s safe to assume that most of it will be spent on candy and the child will develop health problems after a short time.

Why Master Data Management is an essential concept to pursue

The IT cost structure of many companies currently looks similarly unsustainable because large LLMs are being fed unstructured, redundant data without restraint. At first glance, introducing a governance layer may seem like nothing more than an unnecessary cost, but it actually carries far greater implications. This is because, under the EU AI Act and the CSRD, companies will be required in the future to disclose the environmental footprint of their IT. The problem: This affects not only AI providers, but anyone who uses AI.

Poor “token sustainability” thus has a direct negative impact on the ESG rating and, consequently, on credit terms and investor confidence. The solution isn’t a new algorithm, but something well-known: Master Data Management. A clean, consolidated “golden record” instead of a data quagmire often reduces token consumption by 60 to 90 percent. As a side benefit, it breaks down silos. Sales, support, and logistics all work from the same reliable foundation at once, resulting in fewer errors in day-to-day operations all without any AI involved. And when you streamline your data in this way, you no longer need billion-dollar cloud models for many tasks. Smaller, locally hosted open-source models are entirely sufficient. The positive side effects:

• 100% Data Sovereignty
• Independence from the prices set by major providers
• Minimal energy consumption.

Stop treating AI as a playground for experimenting with prompts. Focus on the real business value. Because if you want to operate AI profitably and sustainably, you first need to do your homework on data management. Anything else is just expensive digital bloat.