Case Study: IP Valuation Case Study: Pricing a Generative-AI Patent Portfolio

This IP valuation case study follows a representative scenario in the fastest-moving corner of the patent world: generative AI. A venture-backed company preparing a Series C round and fielding acquisition interest needed a defensible value for its patent portfolio, in a field where filings have tripled in two years and clean comparables are scarce. It illustrates how PerspireIP selects and combines valuation approaches when the technology is new, the market is loud, and the number has to survive a diligence team.

The Challenge

The company had built a portfolio of roughly 120 patent families around generative-AI model architecture and deployment. It was raising a Series C and had received informal acquisition interest, and both processes demanded the same thing: a credible dollar value for the IP that investors and an acquirer’s advisors would accept. Management’s own estimate spanned an order of magnitude, which is a red flag in any data room.

The valuation environment made this genuinely hard. WIPO’s Patent Landscape Report on generative AI records that newly published patent families in the field grew from about 14,000 in 2013 to 37,800 in 2025, with the 2024-2025 window alone outpacing the entire preceding decade. A field expanding that fast produces few stable, arm’s-length comparables, rapid obsolescence risk, and a temptation to anchor on hype rather than evidence. The company needed a number built on method, not momentum.

Our Approach

We began where every engagement described on our IP valuation service begins: matching the approach to the asset, the purpose and the available evidence. For a revenue-adjacent, actively licensed-into technology, the income approach was primary, corroborated by the market approach and sanity-checked against cost.

• Income approach (relief-from-royalty). We modelled the royalty the company would have to pay to license equivalent technology, then discounted that avoided stream over a deliberately conservative useful life to reflect the obsolescence risk a fast-moving field carries.
• Market approach (comparable benchmarking). We benchmarked against arm’s-length software and AI licences, where published royalty surveys place typical rates in the 8-12% range for software, then adjusted each comparable for claim scope, remaining term and deal structure rather than copying a headline rate.
• Cost approach (floor check). We estimated replacement cost as a floor and a cross-check on the income conclusion, useful because a slice of the portfolio covered internal tooling with no isolable income stream.

Throughout, we separated the portfolio into a small set of core families that carried most of the value and a longer tail of defensive filings, because valuing them identically would have overstated the whole.

What This IP Valuation Case Study Found

Three findings shaped the conclusion. First, value was highly concentrated: a minority of the 120 families accounted for the large majority of the defensible value, consistent with how patent portfolios generally distribute worth. Second, the market comparables clustered tightly once adjusted for scope and term, which gave the relief-from-royalty rate real evidentiary support rather than a rule-of-thumb starting point.

Third, useful life, not the royalty rate, was the input the value was most sensitive to. In a field where WIPO data shows filing volumes doubling in a couple of years, assuming a long economic life would have inflated the number and invited exactly the challenge a diligence team is trained to make. Documenting a conservative, defensible life was what let the conclusion hold up.

The Outcome

The engagement delivered a valuation range rather than a single false-precision figure, with the core families valued on corroborated income evidence and the tail valued conservatively. Every input, the royalty benchmark, the discount rate, the useful life and the comparables, traced to a cited source, and the memo laid out the sensitivities that would move the number.

In the representative scenario, that let management enter the Series C and the acquisition conversation with a number their own advisors could defend line by line, replacing an order-of-magnitude guess with a range grounded in method and market evidence. The same memo was structured to be reusable if the matter ever moved toward a licensing negotiation or a damages claim.

Crucially, the deliverable did not stop at a figure. It set out the standard of value used, the approach chosen for each slice of the portfolio and why, and the specific assumptions a counterparty would be most likely to probe, so the company walked into every conversation already knowing where the pressure would come from and how the analysis answered it.

What This Means for Similar Matters

Two lessons generalise from this IP valuation case study. First, in a hype-driven field the discipline is subtraction, not addition: a conservative useful life and adjusted comparables produce a number that survives scrutiny, while an optimistic one invites it. Second, a valuation is a document, not a datapoint. The range mattered less than the fact that a reader could follow every assumption to its source and arrive at the same conclusion.

For any company whose value lives in fast-moving IP, whether AI, semiconductors or biotech, the takeaway is the same: build the number on the recognised approaches, tie every input to real evidence, and document it to the standard the toughest audience will apply.

Reading the Number: Why a Range Beats a Point

Prospects often ask for a single figure, but a single figure is the least defensible thing a valuation can offer. A point estimate hides its own uncertainty, and the first question any sophisticated counterparty asks is how sensitive that number is to its assumptions. A range makes the uncertainty explicit and, paradoxically, makes the valuation more persuasive: it shows the analyst understood the drivers well enough to bound them.

In this scenario the range was narrow where the evidence was strong, the core families with clustered market comparables, and wider where it was thin, the defensive tail and the internal tooling valued on cost. That shape is itself information. It told management exactly which assets to foreground in the raise and which questions to expect from a diligence team, and it gave the acquirer’s advisors a structure they could interrogate rather than a headline they would reflexively discount.

The broader point applies well beyond AI. Intangible assets now represent roughly 90% of the market value of the S&P 500, so for most technology companies the valuation of the IP is close to the valuation of the business. A number built and documented this way does more than close a financing round; it becomes a reusable asset the company can lean on in the next licence negotiation, the next audit, or the next dispute.

Data Sources

This representative scenario is built from recognised valuation standards and primary market data:

• WIPO — Patent Landscape Report: Generative AI — Generative-AI patent families grew from ~14,000 (2013) to 37,800 (2025)
• IVSC — IVS 210 Intangible Assets — The standard defining the income, market and cost approaches
• Ocean Tomo — Intangible Asset Market Value Study — Intangibles at ~90% of S&P 500 market value

Discuss a Similar Matter

Have a portfolio that needs a defensible number for a raise, a sale or a dispute? We build valuations to the standard your audience will apply.

Frequently Asked Questions