AI Valuation Methods: How to Value AI Assets in M&A

The AI Valuation Crisis: Why Deals Fail

For AI acquisitions specifically, the failure rate is even higher—many PE and strategic acquirers lack frameworks to properly value AI assets, leading to overpayment or unrealistic synergy assumptions.

The problem is fundamental: AI is predominantly an intangible asset, and most acquirers use financial models designed for tangible asset valuation. They apply standard software multiples (5–8x revenue) without understanding whether the AI asset has genuine competitive defensibility, sustainable competitive advantage, or material risk of model degradation.

---

Overview: Four Valuation Methods Compared

Method	What It Values	Best Used For	Key Inputs	Risk Level
Relief from Royalty (RFR)	Economic benefit of ownership vs. licensing	Mature AI models, software platforms, technology with clear competitive benchmarks	Revenue influenced, licensing rate, discount rate	Medium (depends on accurate licensing rate)
Multi-Period Excess Earnings (MPEEM)	Earnings attributable to the AI asset after deducting other contributory assets	Customer-facing AI, revenue-generating models, systems with clear customer relationship value	Revenue, margin, customer attrition, CACs	High (requires detailed forecasting)
Cost Approach	Cost to recreate equivalent asset from scratch, adjusted for obsolescence	Foundational technology, training data, proprietary infrastructure, systems where no licensing comparables exist	Development cost, rebuild time, obsolescence rates	Medium (sensitive to rebuild time assumptions)
Market Approach	Price paid for comparable AI assets in recent transactions	Situations where comparable transaction data exists (AI startups, SaaS platforms)	Comparable transaction multiples, revenue/EBITDA scaling	High (limited comparable data in AI space)

---

Method 1: Relief from Royalty (RFR)

What It Is

Relief from Royalty (RFR) values an AI asset by estimating what the acquirer would need to pay to license equivalent capability from a competitor if they did not own the asset. The difference between ownership and licensing is the value captured by the asset.

Logic: If you own a proprietary AI system, you do not have to pay licensing fees. The savings on those licensing fees over time represents the economic value of asset ownership.

When to Use It

RFR works best when:

• The AI system has clear competitive alternatives that charge identifiable licensing rates
• The asset generates revenue or cost savings that are straightforward to measure
• The model is mature and demonstrably performs at or above industry standards
• You are valuing technology, software, or data assets with established licensing markets

Example: Acquiring a company with a proprietary credit risk model used in lending. Comparable models (e.g. Equifax, FICO alternatives) charge 2–4% of credit portfolio value as licensing fees. RFR values the acquisition target's model by calculating the licensing savings.

The Calculation

Worked Example

Valuation Calculation:

• Annual licensing rate: £45m × 3% = £1.35m
• Projected licensing savings (5 years): £1.35m per year with 4% revenue growth
• Year 1: £1.35m | Year 2: £1.40m | Year 3: £1.46m | Year 4: £1.52m | Year 5: £1.58m
• Total undiscounted: £7.31m
• Discount rate: 18% WACC (technology asset)
• Present value: ~£4.2m

RFR-derived valuation: £4.2 million

Critical assumption: This assumes the model maintains parity with competitor offerings. If competitors improve their models, or if this model degrades, the value declines.

---

Method 2: Multi-Period Excess Earnings (MPEEM)

What It Is

MPEEM isolates the earnings attributable to a specific AI asset by deducting the economic contribution of all other assets (brand, workforce, capital, other technology). The remaining earnings—the "excess earnings"—belong to the AI asset being valued.

Logic: A customer relationship AI (chatbot, recommendation system) generates revenue. But that revenue also depends on brand, product quality, customer service, and other factors. MPEEM isolates how much revenue is specifically attributable to the AI system.

When to Use It

MPEEM works best when:

• The AI asset directly influences customer-facing revenue or retention
• You can reasonably quantify the contribution of other assets
• The AI system has a clear competitive advantage you can model
• You have historical data on customer behaviour with and without the AI

Example: Acquiring a company with a recommendation engine that personalises customer shopping experience. MPEEM isolates the incremental revenue from the recommendation engine by modelling customer behaviour with the old non-AI experience vs. the new AI experience.

The Calculation

MPEEM requires detailed financial modelling and an understanding of "Contributory Asset Charges" (CACs)—what other assets contribute to the revenue. The simplified framework:

Worked Example

MPEEM Calculation (Year 1):

• Revenue influenced by recommendation engine: £120m × 35% = £42m
• Gross margin on that revenue: £42m × 45% = £18.9m
• Contributory Asset Charges:
• — Brand/Marketing CAC: 8% of revenue influenced = £3.36m
• — Working Capital CAC: 5% = £2.1m
• — Workforce CAC: 4% = £1.68m
• — Other technology CAC: 3% = £1.26m
• Total CACs: £8.4m
• Excess Earnings (AI asset attribution): £18.9m – £8.4m = £10.5m
• Discount rate: 20% (customer-centric AI, medium risk)
• 5-year projection (assuming 3% annual growth): Year 1 £10.5m, Year 2 £10.8m, Year 3 £11.1m, Year 4 £11.4m, Year 5 £11.8m
• Total undiscounted: £55.6m
• Present value at 20% discount: ~£22.1m

MPEEM-derived valuation: £22.1 million

Critical assumptions: (1) 35% revenue attribution to the recommendation engine—this should be validated with A/B testing data; (2) CAC percentages—these should be benchmarked against industry standards; (3) stability of competitive advantage.

---

Method 3: Cost Approach

What It Is

The Cost Approach estimates what it would cost to recreate the AI asset from scratch, then adjusts for functional and economic obsolescence. If it would cost £50m to rebuild the model you are acquiring, but the model is already built and deployed, the cost approach captures that economic value.

Logic: If the acquirer had to rebuild this AI system themselves, what would it cost? That cost represents a floor on the asset's value (you will not pay more than the cost to rebuild).

When to Use It

Cost Approach works best when:

• The AI asset has no clear licensing comparables (RFR not available)
• The asset is foundational technology or training data without direct revenue attribution (MPEEM not suitable)
• You have good visibility into rebuild time and cost
• The asset is relatively mature and stable (not rapidly evolving)

Example: Acquiring a company with proprietary training data (10 years of accumulated customer interactions) and a foundation model trained on that data. There is no licensing comparable. The value derives from the cost to rebuild the dataset and retrain an equivalent model.

The Calculation

Worked Example

Cost Approach Calculation:

• Data Acquisition & Labelling:
• — Cost per record: £0.15 (labelling, validation, deduplication)
• — Quantity: 50 million records
• — Total data cost: £7.5m
• Engineering Effort:
• — Model development: 12 months of senior engineering (£150k/year) = £1.5m
• — Data pipeline, ETL, infrastructure: 8 months (£120k/year rate) = £0.8m
• — Quality assurance, testing: 4 months (£100k/year rate) = £0.33m
• — Total engineering: £2.63m
• Infrastructure & Tools:
• — Cloud compute, storage, development tools: £0.5m
• Overhead (40%): (£7.5m + £2.63m + £0.5m) × 40% = £4.13m
• Gross Replacement Cost: £15.26m
• Obsolescence Adjustment:
• — Functional obsolescence (model architecture newer today): 10% = £0.53m
• — Economic obsolescence (market value declined): 5% = £0.26m
• — Total obsolescence: 15% = £0.79m
• Net Replacement Cost (Fair Value): £15.26m – £0.79m = £14.47m

Cost Approach-derived valuation: £14.5 million (rounded)

Critical insight: This establishes a ceiling on value, but not necessarily the market value. If the data and model generate significant revenue (tested via MPEEM), value could be higher. If the asset has become obsolete, value could be lower.

---

Method 4: Market Approach

What It Is

The Market Approach values an AI asset by reference to prices paid for comparable assets in recent transactions. If a comparable AI company sold for 12x revenue, and your acquisition target has £10m revenue, estimated value is £120m.

Logic: Market transactions reveal what willing buyers are paying for assets. Observed prices provide empirical validation of value.

When to Use It

Market Approach works best when:

• Multiple comparable transactions exist (AI acquisitions are still relatively rare, so this is often limited)
• You are valuing a standardised AI product (SaaS, platforms) rather than proprietary custom systems
• Transactions are recent and in similar markets/geographies

Challenge: Comparable AI transaction data is sparse. The AI market is still immature. Many high-value acquisitions are confidential or part of larger acquisition bundles, making it difficult to isolate the AI asset value.

Worked Example

Scenario: Valuing an AI-powered customer service chatbot platform. Recent comparable transactions:

• Platform A (2024): £400m for £28m revenue (14.3x revenue multiple)
• Platform B (2024): £280m for £31m revenue (9.0x revenue multiple)
• Platform C (2023): £550m for £35m revenue (15.7x revenue multiple)

Average multiple: 13.0x

If your target has £20m revenue:

• Valuation range: £20m × 9.0x to £20m × 14.3x = £180m to £286m
• Midpoint (using average 13.0x): £260m

Market Approach-derived valuation: £180m–£286m (using range), £260m (midpoint)

Critical caveat: These multiples reflect market froth and perceived growth potential at the time of transaction. They are not anchored in economic fundamentals and can be misleading if the market is overheated (as it was for AI in 2023–2024).

---

Triangulation: Using All Four Methods Together

Professional valuers do not rely on a single method. They use all four, then triangulate to a defensible range. Here is how:

Interpretation: The four methods give wildly different results, which is typical. The valuator's job is to understand which methods are most relevant to the specific asset being acquired, and weight them accordingly.

For a customer-facing AI (recommendation engine, chatbot), MPEEM and Market Approach are most relevant. Weight them 60%, and RFR/Cost Approach 40%. For a foundational technology or data asset, Cost Approach and RFR are more relevant. For a SaaS AI platform in a liquid market, Market Approach carries more weight.

Final valuation range typically spans 30–50% of the high end to the low end of the four method results. Final negotiation price falls within that range, depending on strategic value, competitive positioning, and talent retention risk.

---

AI Valuation Red Flags: What Due Diligence Should Uncover

---

Valuation Framework Summary Table

Scenario	Primary Methods	Typical Valuation Range	Key Risk
Proprietary predictive model (credit, demand, churn)	RFR (40%), MPEEM (40%), Cost (20%)	3–8x annual economic value	Model degradation, data drift
Customer-facing AI (chatbot, recommendations, personalisation)	MPEEM (50%), Market (30%), RFR (20%)	8–15x revenue (if proven ROI)	Revenue attribution, market saturation
Proprietary training data + foundational model	Cost (40%), RFR (40%), MPEEM (20%)	2–5x rebuild cost	Data obsolescence, IP ownership
AI SaaS platform (liquid market, many comparables)	Market (50%), MPEEM (30%), RFR (20%)	8–20x revenue (depending on growth)	Market multiples compression, talent retention

---

What This Means for Buyers and Sellers

For Sellers: If you have built a defensible AI asset with clear revenue attribution, MPEEM and Market Approach will yield your highest valuations. Invest in demonstrating these metrics (A/B testing, customer segmentation analysis, retention curves) before selling. Companies that can prove ROI command 20–40% valuation premiums.

For Buyers: Use all four methods. If they diverge significantly (RFR says £4m, MPEEM says £22m, Cost says £14m), the spread reflects genuine uncertainty. Negotiate within the range, but weight your due diligence toward the highest-uncertainty areas. The most common acquirer mistake is using the highest valuation method without questioning its assumptions.

---

Next Steps

For detailed valuation frameworks and calculators for each method, see the Opagio Valuator Tool. For understanding the intangible assets your AI creates, see Intangible Asset Categories. For risk assessment guidance on AI acquisitions, see AI-Washing Detection.