How do you value a company's proprietary algorithms and AI models?
Short Answer
Proprietary algorithms and AI models are valued using the Relief from Royalty method (rate depending on application) or cost approach (development costs plus opportunity cost), considering the model's accuracy advantage and data moat.
Full Explanation
Valuing proprietary algorithms and AI models requires understanding both their technical characteristics and their commercial impact. The Relief from Royalty method is commonly applied, using royalty rates derived from software and technology licensing benchmarks. Rates vary significantly by application, and the following are illustrative ranges only (the actual rate depends on the licensing evidence for the model): general business process automation algorithms (5-10%), specialised analytical models for financial services or healthcare (10-15%), and cutting-edge AI models with significant competitive advantage (15-25%). The cost approach is also relevant, particularly for models where licensing comparables are scarce. The calculation includes: data scientist and engineer time invested in development, cost of training data (acquisition, cleaning, labelling), computational costs (GPU/cloud infrastructure for training), and an entrepreneurial profit margin reflecting the risk-adjusted return required for the development effort. Key valuation considerations specific to AI models include: accuracy advantage (how much better does the model perform compared to available alternatives — measured by metrics like AUC, F1 score, or custom performance benchmarks), data dependency (a model trained on proprietary data has higher value because competitors cannot easily replicate the training set), decay rate (AI models degrade as underlying data patterns shift, requiring retraining — this affects useful life), regulatory considerations (models used in regulated industries like financial services or healthcare face compliance requirements that affect both value and risk), and explainability (more interpretable models may be more valuable in regulated contexts where black-box approaches face restrictions). The useful life of AI models is typically short (2-5 years) due to rapid technological advancement and data drift, resulting in accelerated amortisation schedules.
Try It Yourself
Related Questions
AI-powered customer relationship assets are valued using the Multi-Period Excess Earnings Method (MPEEM), with adjustments for AI-driven improvements ...
AI amplifies Total Factor Productivity by enabling firms to extract more output from the same inputs, with the productivity gains increasingly attribu...
AI models typically have useful lives of 2-5 years for amortisation, reflecting rapid technological advancement, though the underlying training data a...
Further Reading
AI Cost Reduction: How to Measure Real Operational Savings
AI vendors promise dramatic cost reductions, but actual savings depend on total cost of ownership, implementation quality, and measurement rigour. This article provides a practical methodology for measuring real AI-driven operational savings — net of all hidden costs.
Read more →Want to see these concepts in action?
Take the free intangible asset assessment to see how these concepts apply to your business.