Every investment thesis for AI-enabled companies focuses on the upside: automation gains, scale advantages, competitive moats, and valuation premiums. Far fewer investors systematically assess the downside. This asymmetry is dangerous because AI introduces risk categories that traditional investment frameworks were not designed to capture.
This article provides a quantitative framework for assessing AI-specific risks, translating qualitative concerns into probability-weighted financial impact estimates that can be incorporated into investment models.
AI models in production degrade over time as the data distribution shifts away from the training data. A fraud detection model trained on 2024 transaction patterns becomes less effective as fraud techniques evolve. A demand forecasting model trained on pre-pandemic data failed dramatically during COVID-19.
Quantification approach: Estimate the revenue or cost savings attributable to the AI model. Apply a degradation curve based on the model's domain (fast-changing domains: 15-25% annual degradation; stable domains: 5-10%). Calculate the expected value loss over the investment horizon if retraining is delayed or fails.
AI models are only as good as their training data. If key data sources become unavailable — through regulatory changes, partner contract terminations, or technical failures — the AI system may become unusable.
Quantification approach: Map all critical data sources. For each, estimate the probability of disruption (contract expiry, regulatory change, technical failure) and the financial impact if the data source is lost. The expected loss is the sum of (probability x impact) across all data sources.
Data dependency is often the most material AI risk and the least assessed. A company whose AI capability depends on a single third-party data source has concentration risk that can destroy the entire AI valuation premium if that source is lost. Map data dependencies as carefully as you map revenue dependencies.
The regulatory environment for AI is evolving rapidly. The EU AI Act (August 2026), SEC enforcement on AI washing, and industry-specific regulations create compliance costs and operational constraints that affect AI value.
| Regulatory risk | Probability | Impact range | Jurisdiction |
|---|---|---|---|
| EU AI Act non-compliance | Medium-High | Up to 7% of global revenue | EU/EEA |
| SEC AI disclosure enforcement | Medium | Legal costs + reputational damage | US |
| GDPR data training violations | Medium | Up to 4% of global revenue | EU/EEA |
| Industry-specific AI restrictions | Variable | Operational limits + compliance costs | Sector-dependent |
| AI bias/discrimination claims | Low-Medium | Legal costs + reputational damage | All jurisdictions |
AI capability is often concentrated in a small number of key individuals. The departure of a lead ML engineer or the head of data science can cripple an organisation's AI function, particularly in smaller companies.
Quantification approach: Identify key AI personnel. Estimate the probability of departure (based on tenure, compensation benchmarking, market demand). Estimate the cost of replacement (recruitment, ramp-up, productivity loss during transition). For key-person dependencies, estimate the revenue impact if the individual departs and the AI function is disrupted.
A portfolio company's recommendation engine — responsible for 30% of online revenue — was built and maintained by a single senior ML engineer. When the engineer departed for a competitor, the model could not be updated or retrained for 8 months while a replacement was recruited and onboarded. Online revenue declined by 12% during this period, costing £1.4 million. The risk was identifiable but had not been quantified or mitigated.
AI technology evolves faster than any previous technology category. A model that represents cutting-edge capability today may be commodity in 18 months. Companies that cannot keep pace with the technology frontier risk losing competitive advantage.
AI systems that produce biased outputs, make discriminatory decisions, or behave in unexpected ways create reputational and legal exposure. The damage from an AI ethics failure can far exceed the direct financial cost — customer trust, brand value, and regulatory goodwill are all intangible assets at risk.
Companies that depend on third-party AI providers (OpenAI, Anthropic, Google) face platform dependency risk. Price increases, API changes, service disruptions, or policy changes by the provider can materially affect the dependent company's operations and economics.
Platform dependency risk is often invisible in due diligence because companies describe third-party AI integrations as "proprietary AI" capability. A company that routes 80% of its product functionality through a single AI API has a concentration risk comparable to a company that sources 80% of revenue from a single customer. Assess AI platform dependencies with the same rigour.
Map every AI system to the seven risk categories. For each system-risk combination, assess whether the risk is material. Focus detailed quantification on the material risks.
For each material risk, estimate the probability of occurrence (annually) and the financial impact if it occurs. Use ranges rather than point estimates: best case, expected case, worst case.
Expected annual loss = Probability x Expected impact. Sum across all material risks to get the total expected AI risk cost. This figure should be incorporated into the investment model as a risk-adjusted discount.
Model the worst-case scenario for each risk category. What happens to the investment thesis if the primary AI model degrades 40%? If the key data source is lost? If the AI team departs? If the investment case survives multiple stress scenarios, AI risk is manageable.
The simplest approach is to adjust the discount rate used in DCF or income-based valuations. AI-specific risks typically warrant an additional 3-8% discount rate premium above the company's standard cost of capital, depending on the severity and concentration of AI risks.
A more sophisticated approach discounts specific cash flow projections based on the probability of AI risk scenarios. If 20% of projected revenue depends on an AI system with a 15% annual probability of significant degradation, the expected value of that revenue stream is reduced by the probability-weighted loss.
The Opagio Growth Platform systematically assesses AI risk across all seven categories, providing investors with a structured risk profile for portfolio companies and acquisition targets. The questionnaire captures technology risk indicators as part of its comprehensive intangible asset evaluation.
AI investments carry risks that traditional frameworks miss. The seven-category risk framework — model degradation, data dependency, regulatory exposure, talent concentration, technology obsolescence, ethical/reputational risk, and platform dependency — provides a systematic approach to identifying and quantifying AI-specific downside. Incorporating these risks into valuation models through adjusted discount rates or probability-weighted cash flows produces more realistic investment assessments. In a market that overwhelmingly focuses on AI upside, rigorous downside analysis is a genuine competitive advantage for investors.
Ivan Gowan is Founder and CEO of Opagio. His experience managing technology risk at IG Group (LSE: IGG) across 15 years — including during the 2008 financial crisis and multiple regulatory transitions — informs Opagio's approach to technology risk assessment. Learn more about the Opagio team.
I was at IG Group when Facebook's IPO was mispriced. The market overvalued the platform and undervalued the intangible assets — the data, the network effects, the advertiser relationships. Today's AI market is making the same mistake in reverse. Here is how to tell the difference between genuine AI value and bubble pricing.
Read more →Get the latest insights on intangible asset growth and productivity delivered to your inbox.
Book a free consultation to see how the Opagio Growth Platform can help your business.