Executive Summary

The sell-off across Software-as-a-Service (SaaS) equities since our launch (in November 2025) has been material and indiscriminate. While portfolio performance has been negatively impacted in the short term, we believe the market is mispricing durability by conflating feature-level AI disruption with true business-model displacement. Artificial intelligence is reshaping where value accrues in software stacks, but it is not uniformly destroying incumbents.

Our mandate is to anticipate how emerging technologies, particularly AI, will affect portfolio companies before disruption becomes visible in financial results. This requires grounding our medium-term assumptions in near-term signals from today’s market while remaining alert to non-linear shifts that could invalidate historical precedents. It is a difficult task in a rapidly evolving environment, but one in which our team has deep experience.

AI-native startups excel at interface-first products, short and bounded workflows, and use cases built on public or transient data. By contrast, systems-of-record software, which is embedded in regulated workflows, supported by proprietary data, and protected by high switching costs remain structurally resilient. In many cases, AI could potentially enhance the value of these platforms rather than undermine them.

Markets are overreacting to narrative risk while underweighting observable evidence including enterprise buyer behavior, implementation realities, and capital cycle dynamics. Public SaaS valuations have de-rated to levels last seen during periods of systemic stress, despite limited changes to sales or earnings forecasts. This dislocation creates an attractive opportunity for patient capital with multi-year horizons.

1. Introduction: Markets, AI, and Time Horizons

The ongoing SaaS sell-off has clearly weighed on returns. However, short-term price action does not invalidate the long-term thesis. Rather, it reflects the structural incentives of public markets, where many participants are judged on monthly or quarterly performance and respond to uncertainty by reducing exposure wholesale.

Ahead of launch, we believed AI would be a net positive for active stock selection. In theory, AI should increase dispersion by amplifying competitive advantages for companies with strong moats, data assets, and pricing power. What we have observed instead is broad-based selling across SaaS, where perceived AI “winners” rapidly become “losers” with little underlying change in fundamentals.

This is not unusual in periods of technological transition. What is unusual is the speed at which narratives now outrun evidence, maybe a broader societal dynamic driven by news consumption shifting to social platforms, where information is increasingly produced and amplified outside traditional journalistic standards.

2. Anticipating Disruption: A Framework for Evaluating AI Risk

Our mandate is to anticipate how emerging technologies will impact portfolio companies before disruption becomes visible in financial results. This requires balancing two perspectives: understanding where AI capabilities are today as a guide to where they're heading, while remaining alert to non-linear shifts that could invalidate historical patterns.

AI-native product launches reveal what new entrants are building and where they are gaining early traction versus struggling to scale, while the rapid evolution of foundation models raises questions about how quickly capabilities are advancing and which new use cases become viable at each capability threshold. At the same time, hyperscalers such as OpenAI, DeepSeek, and Anthropic are pouring billions into AI infrastructure, creating pressure for rapid commercialization as someone ultimately has to fund these investments. Incumbent players are responding aggressively—since January 2025, more than 85% of software M&A activity has been AI-related—showing how established companies are internalizing AI capabilities. This contrasts with enterprise procurement behavior, where CIO and CTO surveys consistently indicate a preference for sourcing AI from existing vendors, even as equity markets price in the opposite outcome. Ultimately, the key question is actual customer adoption: beyond announcements and demos, which AI features customers are truly using and paying for.

The coding tools offered by Claude and others have clearly surprised the market with their capabilities. But we anticipated this trajectory before launch and constructed the portfolio accordingly, focusing on software with entrenched moats, proprietary data, high switching costs, and system-of-record status.

The critical insight from our analysis: AI is not disrupting software uniformly. Instead, it is shifting value toward certain layers of the stack while reinforcing others. The question is not whether AI will impact software economics, it will, but where specifically that impact creates displacement risk versus where it strengthens incumbent positions.

Understanding this distinction is essential to separating genuine threats from market narratives.

3. Where AI-Native Startups Win

AI-native competitors are genuinely disruptive when several conditions are met simultaneously:

• Interface-first products where value accrues at the interaction layer (prompt, chat, agent)

• Short, bounded workflows with limited state persistence

• Public or transient data with low governance friction

• Bypassable distribution, allowing products to spread through usage rather than procurement

In these environments, the underlying AI models standardise much of the core capability, so differentiation shifts to how well a product understands user intent, assembles the right context, and delivers useful outputs. The traditional SaaS model, where complex data and business logic sit behind a simple interface, begins to flip. The interface itself becomes the primary source of competitive advantage.

This creates real risk for aggregation businesses and platforms that sit between users and services. When AI agents can complete tasks end-to-end on a user’s behalf, the need for intermediaries can diminish, putting pressure on companies whose value is concentrated at the interface layer.

4. Where AI-Native Startups Lose

By contrast, AI-native entrants struggle in environments defined by:

• Systems of record, where data integrity, auditability, and determinism are non-negotiable

• Proprietary and regulated data, subject to residency, access control, and explainability requirements

• High switching costs, including data migration, retraining, process redesign, and regulatory re-approval

• Long implementation cycles, which give incumbents time to respond

  
  

In these contexts, superior user experience rarely overcomes structural lock-in. Adoption decisions are governed by a simple equation:

Adopt new AI product if UX gain + productivity delta > switching cost + risk premium

For mission-critical enterprise software, switching costs typically dominate. AI therefore augments incumbent platforms rather than displacing them. 

5. Incumbents Are More Adaptable Than the Market Assumes

A common market narrative is that AI-native startups will outpace incumbents due to speed and innovation. This view is anchored in the last major software transition, when cloud-native SaaS disrupted slow-moving on-premise vendors.

Today’s incumbents are different. They are already cloud-native, ship software continuously, and have access to the same AI tooling as startups. The historical two-year innovation lead enjoyed by venture-backed challengers has compressed to weeks. In many cases, incumbents can replicate or integrate new AI features faster than startups can scale distribution.  See Appendix B for further thoughts on how the economics compare.

Crucially, enterprise buyers consistently express a preference to source AI capabilities from existing vendors rather than unproven startups, particularly in regulated or mission-critical contexts. Survey after survey of CIOs and CTOs supports this view, even as equity markets price the opposite outcome.

6. Non Linear Shifts 

Several non‑linear forces may break historical software patterns in an AI-driven environment. Step‑function advances in foundation models, such as reliable reasoning or autonomous tool use (agent to agent) can rapidly collapse entire categories, replacing many point solutions with generalised systems. At the same time, intelligence costs are falling sharply, enabling previously uneconomic use cases like continuous agents and automated back‑office workflows, which in turn challenge traditional SaaS pricing and margin models. Distribution may also shift, as AI agents embedded at the platform or OS level bypass app‑level adoption paths. Data moats could erode quickly as synthetic data and shared model improvements reduce the defensibility of proprietary datasets. And if small, AI‑augmented teams can match the output of much larger organisations, conventional links between headcount, revenue, and market power begin to break down.

Yet these shifts should not be overstated. Many software categories remain resilient due to deep workflow integration, regulatory demands, enterprise procurement cycles, and domain‑specific trust. History shows that incumbents often adapt effectively when they control distribution and customer relationships, and generalised AI systems may still struggle with the reliability and compliance required in specialised domains. In the near term, a more plausible outcome is hybrid evolution: incumbents layering AI into existing systems, maintaining core moats even as functionality expands. 

Non-linearity is possible, and if these shifts materialise, we will adapt the portfolio accordingly, drawing on our experience navigating disruption. We have already identified the companies most likely to benefit under such a scenario.

7. Portfolio Construction and Sector Implications

Approximately ~40% of the portfolio is invested in SaaS businesses across technology and healthcare. These companies share common characteristics:

• Entrenched competitive moats

• Proprietary datasets accumulated over many years

• High switching costs and workflow embedding

• System-of-record or mission-critical status

  
  

Many have already partnered with leading foundation model providers and are embedding AI natively into their platforms. Monetisation strategies increasingly rely on usage-based or outcome-based pricing, aligning vendor revenue with customer value creation.

Importantly, incremental AI-driven revenue often expands total addressable markets by displacing customer labour, rather than competing directly for existing software budgets.

8. Valuations and Private Equity¹ 

SaaS valuations have de-rated materially. In several cases, multiples are at or near historical troughs (see Appendix A1), despite limited changes to sales and earnings expectations. Unless terminal values are permanently impaired, which we do not believe is supported by our base case, internal rates of return increase as entry prices fall.

Private equity behaviour reinforces this view. With longer time horizons and no mark‑to‑market pressure, private capital can absorb short‑term margin investment and wait for AI‑driven productivity gains to materialise. Recent take‑private activity in the portfolio highlights the gap between public and private market expectations. While sceptics may argue that private equity is merely supporting its own valuations, we believe these transactions are underpinned by the ability to generate meaningful cash extraction even under modest growth assumptions.

We have made no material portfolio changes since launch, aside from replacing holdings that became take-out targets. Depressed valuations and strategic optionality make further approaches plausible.  

9. Conclusion: Patience as a Structural Advantage

Markets are overestimating AI-native threats to workflow-embedded software and underestimating threats to interface-driven aggregation businesses. The result is indiscriminate selling that ignores fundamental differences in durability.

We believe the current environment rewards patience, not prediction. Companies that maintain market share, pricing power, and data advantages can absorb near-term AI investment and compound value over time. Those without these attributes may struggle, but capital discipline will ultimately reassert itself.

At a portfolio level, our IRR hurdle has increased materially and now approaches 25% (See A5), well above our original target². We remain highly optimistic about the next five years. 

Goodhart Future Leaders - Focus Areas of Investment

Healthcare Software

• Regulated workflows, proprietary data, and platform primacy create low AI-native disruption risk.

• AI primarily enhances productivity and platform stickiness.

Consumer Marketplaces and Classifieds

• Strong data liquidity and network effects limit disintermediation.

• AI improves matching, pricing, and user experience rather than replacing platforms.

Developer Tools and Infrastructure

• AI increases code volume and system complexity, driving demand for governance, observability, and orchestration.

• Monetisation shifts from seats toward hybrid usage models.

ERP

• AI-native UX competition exists, but compliance, trust, and switching costs favour incumbents.

• AI expands TAM by automating labour-intensive processes.

(Full company-level analyses available on request.)

Appendix A: Valuation Floors³, Durability, and the Cost of Being Wrong

A1. Approaching Historic Support Levels

Recent market action shows several large-cap software names trading at, or below, historic valuation support levels measured on EV / Sales multiples. Many have already breached prior support bands. These levels have historically coincided with periods where the market was pricing in either severe structural disruption or terminal decline.

Source: Bloomberg. US Software refers to “S&P 500 Software Industry GICS Level 3 Index”

The significance is not that these multiples represent “fair value” under normal growth assumptions, but that they have previously acted as floors even during periods of heightened uncertainty. The current breach suggests investors are no longer debating the slope of future growth, but are implicitly questioning whether growth exists at all. For further downside we argue that churn needs to pick up significantly which is unproven in recent results.

EV/Sales masks the substantial margin expansion achieved across software businesses since the shift to SaaS economics. When viewed through earnings rather than revenue multiples, the sector has effectively relinquished its long‑standing premium to the wider market—reverting to valuation levels last observed when companies were first migrating to SaaS.

Source: Bloomberg. US Software refers to “S&P 500 Software Industry GICS Level 3 Index”

A2. Software Economics Under Run-Off Assumptions

A key error in the prevailing AI-driven bear narrative is the implicit assumption that software economics are fragile once growth slows. Empirically, the opposite is true. Mature software assets, when run for cash rather than growth, can sustain exceptionally high operating margins due to:

• Minimal incremental cost to serve existing customers

• Sharp reductions in sales & marketing spend

• Lower ongoing R&D requirements focused on maintenance rather than expansion

  
  

Evidence shows that legacy software assets can be operated at ~70% EBIT margins when optimised for cash extraction rather than growth.  

A3. AI, Coding Costs, and the Misframing of Risk

While AI materially reduces the cost of coding, software has never been primarily about the cost of code production. If it were, free and open-source solutions would have achieved far greater enterprise penetration decades ago. Instead, customers routinely pay for software layered on top of open-source foundations to solve the last-mile problems that matter most:

• Reliability and uptime

• Scalability

• Integration with complex legacy systems

• Ongoing maintenance, compliance, and support

  
  

AI changes the toolset, but it does not eliminate these requirements. As such, falling coding costs should be viewed as margin-enhancing rather than structurally value-destructive for incumbents with entrenched customer relationships.

A4. Stickiness and the Myth of Rapid Displacement

Concerns around terminal growth often underestimate the extreme inertia embedded in mission-critical software. The continued growth in lines of COBOL code—despite the language being over 60 years old—illustrates the durability of systems that “work well enough.”

Customers exhibit a strong preference to avoid disruption to core systems, reinforcing long asset lives and long-duration cash flows even in the face of technological change.

A5. Interpreting Current Valuations

The question is not whether AI will impact software economics, it will, but whether current valuations already discount an overly pessimistic future. Our analysis suggests that, for several large incumbents, equity prices are approaching levels consistent with terminal decline scenarios rather than modestly impaired growth.

It seems unlikely that a high-quality software asset would be placed into run-off, but framing valuation through a liquidation lens provides a useful worst case. It highlights how asymmetric the setup has become: limited downside if durability persists, and meaningful upside if growth merely proves to be lower rather than nonexistent.

In our view, sentiment, not fundamentals is currently driving the market. Over time, we expect the durability of growth and cash flows to reassert itself in valuation multiples.  

As of 11/02, the weighted average portfolio expected IRR is 25.1% and is trading at an aggregate FY6 PE of 7.5x. With forecasted sales growth of 15.3% and 9.1% margin expansion, resulting in 28.6% average margin by FY6, we therefore see plenty of upside within the portfolio⁴. 

Our internal Perfect Foresight analysis supports this outlook⁵. Since December 1989, companies acquired at an ex-post entry valuation of 6–8x FY6 PE have delivered 16.7% annualised returns over the subsequent six years. When combined with 10–20% sales growth and positive margin expansion resulting in between 20 to 30% margin by FY6, median returns rise to 22.7% annualised.

Median Historic Stock Returns by Entry Valuation (Ex-Post FY6 PE)

Median Historic Stock Returns by Entry Valuation (Ex-Post FY6 PE)

Subset of Companies 10–20% Sales Growth, Margin Expansion and 20–30% FY6 Margins⁶ 

Source: Goodhart analysis (31/12/1989 to 31/12/2025). Entry valuation refers to ex-post FY6 PE (price at ‘purchase’ relative to earnings six years forward). “Conditional” subset includes companies that over the equivalent  6-year period achieved 10–20% annualised sales growth, experienced operating margin expansion over the period, and delivered between 20–30% FY6 margins. Returns shown are median annualised six-year total returns. Past performance is not a reliable indicator of future results.

Appendix B: Software Economics — Incumbent vs AI-Native P&L Structures

A useful way to cut through the AI narrative is to compare business models, not technologies. Below we outline what an “average” mature software company looks like from a P&L perspective, and contrast it with an AI-native startup. Crucially, both benefit from AI-driven reductions in development cost, but the distribution of value differs materially.

B1. Mature Software Company: Steady-State P&L

A scaled, incumbent software business with a large installed base typically exhibits the following economic structure:

• Revenue: Predominantly recurring (subscriptions, maintenance)

• Gross Margin: 85–95%+

  • Hosting, support, and delivery costs are low relative to revenue

• R&D: ~10–15% of revenue

  • Focused on maintenance, incremental features, compliance, and roadmap evolution

• Sales & Marketing: ~15–25% of revenue (and materially lower if run for cash)

  • High renewal rates reduce the need for aggressive customer acquisition

• G&A: ~5–10% of revenue

  
  

In steady state, this structure supports: - EBIT margins: 30–40% in growth mode - EBIT margins: 50–70% when optimized for cash extraction

Importantly, AI-driven productivity gains (e.g. faster coding, automated testing, support tooling) largely accrue to margins. The incumbent does not need to reprice aggressively, because customers are paying for reliability, integration, and continuity—not lines of code.

B2. AI-Native Startup: P&L in Practice

An AI-native software company typically presents a superficially similar but economically distinct P&L:

• Revenue: Early-stage, usage- or outcome-based, often volatile

• Gross Margin: 60–80% (often overstated)

  • Inference costs, model hosting, third-party APIs, and compute scale with usage

• R&D: 30–60% of revenue

  • Continuous model training, fine-tuning, and product iteration

• Sales & Marketing: 30–50% of revenue

  • Low switching costs necessitate aggressive customer acquisition

• G&A: 10%+ of revenue

  
  

While AI lowers the unit cost of development, it does not eliminate: 

• Ongoing model costs 

• Rapid feature imitation by competitors 

• Customer churn driven by weak lock-in

  
  

As a result, many AI-native businesses struggle to demonstrate operating leverage, even as revenue grows.

B3. The Key Misconception: Cost of Coding vs Cost of Ownership

AI reduces the cost of building software for everyone. It does not eliminate the cost of owning software as a mission-critical system. Incumbents already operate near the economic ceiling of software margins, whereas AI-native firms often face structurally higher variable costs and weaker durability.

Put differently: for incumbents, AI is margin-accretive, while for AI-native startups it is often survival-critical but not margin-expansive.

B4. Implications for Valuation

If both incumbents and AI-native firms can reduce product development costs using the same tools, then AI is not, in itself, a sufficient explanation for large valuation dispersion. The differentiator remains:

• Installed base longevity

• Switching costs

• Ability to convert revenue into long-duration free cash flow

  
  

From this perspective, fears that AI structurally destroys incumbent software economics appear overstated. The technology compresses development costs across the ecosystem, but it does not compress the value of durable customer relationships, nonlinear shifts could potentially change this, but are hard to predict. 

Footnotes

1 References to private equity transactions are illustrative of market activity and do not imply validation of valuation levels or future performance of portfolio holdings.

2 Expected IRR represents an internal, model-derived estimate based on current portfolio holdings, management forecasts, and Goodhart assumptions regarding revenue growth, margin expansion, exit multiples and capital structure. It is not a target, projection or guarantee of future performance. Actual outcomes may differ materially. Expected IRR is highly sensitive to changes in assumptions, including but not limited to growth rates, profitability, discount rates and terminal values.

3 References to “limited downside” or valuation “floors” reflect historical valuation observations and do not imply capital protection. Equity investments are subject to loss of capital, including the risk of permanent impairment. Valuation levels may not represent support in future market environments.

4 Expected IRR represents an internal, model-derived estimate based on current portfolio holdings, management forecasts, and Goodhart assumptions regarding revenue growth, margin expansion, exit multiples and capital structure. It is not a target, projection or guarantee of future performance. Actual outcomes may differ materially. Expected IRR is highly sensitive to changes in assumptions, including but not limited to growth rates, profitability, discount rates and terminal values.

5 The “Perfect Foresight” analysis is a hypothetical, ex-post exercise that assumes knowledge of future earnings six years forward. Entry valuations are calculated using ex-post FY6 earnings and therefore incorporate look-ahead bias. This analysis is provided for illustrative purposes only to demonstrate valuation sensitivity and does not represent actual portfolio investments, realised returns, or a strategy that could have been implemented in practice. Past performance is not indicative of future results.

6 The conditional subset reflects companies that, over the subsequent six-year period, achieved specified growth and margin characteristics. These characteristics are identified ex-post and therefore incorporate selection and survivorship bias. Results shown are median historical outcomes and do not represent actual portfolio performance or expected future returns.