Enterprise LLM Valuation: How to Monetise Proprietary AI in Switzerland

The question CFOs are starting to ask

"How do we value our AI investments?" The question was theoretical in 2023. In 2025, with companies having invested CHF 2–5 million in AI projects, it's very concrete. And traditional valuation methods — ROI, payback period, NPV — are ill-suited to AI assets.

Why traditional methods are insufficient

A proprietary LLM is not a machine tool. Its characteristics break the assumptions of classic financial models:

• Near-zero marginal cost: the model can handle 1 or 1 million queries without proportional variable cost
• Continuous improvement: value increases with use, without additional investment
• Internal network externalities: every department that uses it improves quality for all others
• Strategic optionality: a trained model can be redirected to new use cases

Three valuation approaches adapted to LLMs

Approach 1 — Substitution valuation

What would the same capability cost if delivered by humans? If your LLM answers 50,000 support questions per month with a 75% resolution rate, calculate the cost of an equivalent human support centre. That's your value baseline.

Approach 2 — Performance differential valuation

Measure the performance gap between your proprietary LLM and a public generalist LLM on your specific use cases. This differential, multiplied by the volume processed, represents the added value of fine-tuning.

Approach 3 — Competitive advantage valuation

Use intangible asset valuation methods (comparable to brand or patent valuation methods). This approach is more complex but more relevant for AI assets with strong strategic potential.

Direct monetisation: three models for Swiss companies

"In 5 years, companies that haven't started building their proprietary AI models will have the same disadvantage as those without a website in 2000."