A rigorous assessment of which business models will compound in the AI economy, which will be disrupted, and why the SaaS-pocalypse narrative – while partially right – misses the more important story

There is a widely circulated thesis in technology investing that AI will devastate the enterprise software industry by making many SaaS applications redundant. The logic is straightforward: if a general-purpose AI agent can perform the workflows that a specialized software application handles, then the application layer collapses. Investors have started calling this the ‘SaaS-pocalypse.’ It is a provocative thesis. It is also only about a quarter right.

The SaaS-pocalypse narrative captures something real: narrow, workflow-automation software that sits between a company and its own data – doing nothing more than helping users perform repetitive tasks within a single function – is genuinely at risk. An AI agent that can replicate those workflows at near-zero marginal cost is a direct substitute. If that is your entire value proposition, you have a problem.

But the thesis fails as a general prediction because it conflates two very different kinds of software business, and it systematically underestimates the durability of businesses whose value comes from places that AI cannot reach: proprietary data accumulated over years of transactions, end-customer relationships with deep trust and legal accountability, network effects that compound as participants grow, and the irreplaceable human judgment required to navigate genuinely complex decisions with high stakes.

Previous articles of this series built the mathematical tools for analyzing work, value delivery, and AI optimization at the firm level. We then applied those tools at the portfolio level – asking the investor’s question: which businesses will compound in an AI economy, and which will be pressured or displaced? The answer is more nuanced and more actionable than either the bulls or the bears on AI disruption have suggested.

“AI does not destroy businesses that own irreplaceable data, hold genuine end-customer relationships, or require coordinated human judgment at the core of their value delivery. It destroys businesses whose only value was mediating access to a workflow that AI can now perform directly.”

I. The Correct Analytical Frame: Value Durability, Not Technology Disruption

The mistake most investors make when evaluating AI’s impact on a business is to ask a technology question: ‘Can AI do what this company does?’ That is the wrong question. The right question is an economic one: ‘Where does this company’s value come from, and is that source durable in an AI economy?’

The Mathematics of Value provides a framework to for answering this precisely. A business’s total value to its customers is the sum of eight components: functional output, outcomes, trust and relationships, speed, risk reduction, information and knowledge, network effects, and experience. AI changes the production cost and delivery speed of each component differently. The businesses that survive and compound are those whose dominant value components AI reinforces rather than substitutes.

The moat durability test can be formalized as a ratio:

The Moat Durability Ratio

Where

M_durable

Moat Durability Ratio

A ratio greater than 1 signals AI-resilient value: the numerator captures value components that AI strengthens or cannot replicate (data, trust, relationships, networks), while the denominator captures components that AI can deliver directly at scale (functional outputs, speed). Businesses with M > 1 are compounders in an AI economy. Businesses with M < 1 face structural pressure.

V_data + V_rel + V_net

Durable Value Components

The three value components most resistant to AI substitution: proprietary data and information value (V_data), relational and trust value (V_rel), and network value (V_net). These are the numerator because they compound over time rather than being competed away.

V_func + V_time

AI-Replicable Value Components

Functional output quality and speed of delivery – the value components where AI can match or exceed human performance at dramatically lower cost. Businesses whose value is primarily functional and speed-based face the most direct competitive pressure from AI-enabled alternatives.

Key Insight

The SaaS-pocalypse narrative is correct specifically about businesses where M_durable < 1: their value is predominantly functional and speed-based, which is exactly where AI is most disruptive. But it is wrong as a general statement about software or service businesses, because many of the most durable business models are software-enabled – they just derive their value from sources that are genuinely AI-resilient.

II. The SaaS Displacement Risk Index

Not all SaaS businesses are equally at risk. The degree of displacement risk depends on a precise combination of factors: how automatable the core workflow is, how easy it is for customers to switch, and how much of the product’s value is derived from data and network effects versus pure workflow execution. The Displacement Risk Index formalizes this assessment:

Definition

Where

DRI_k

Displacement Risk Index for Company k

A normalized score representing how exposed company k is to AI-driven substitution. DRI > θ (a sector-specific threshold, typically 0.6–0.7) signals high displacement risk. DRI < 0.4 signals structural resilience. Most narrow workflow SaaS products score DRI > 0.7.

Σ a_i·w_i

Weighted Automation Coverage

The sum of automation indicators across all tasks in the product’s workflow, weighted by importance. This is the core of the DRI: how much of what the product does can an AI agent replicate? Drawn directly from weighted automation model, high Σa_i·w_i means most of the product’s functional value is automatable.

Φ_switch

Switching Cost and Data Lock-In

The total friction of replacing the product: data migration costs, retraining, integration rebuilding, and the irreplaceability of proprietary data held within the platform. High Φ_switch protects against displacement even when automation coverage is high. This is why data ownership is the most important moat variable.

The DRI framework reveals a critical insight that is frequently missed: a high degree of workflow automation (Σa_i·w_i) is only fatal when switching cost is low. A product that handles a highly automatable workflow but holds years of proprietary, irreplaceable customer data is far more durable than its automation profile suggests. The data moat suppresses the denominator, keeping DRI below the displacement threshold even as AI improves.

Key Insight

Data lock-in is not the same as switching cost in the traditional sense. Traditional switching cost is friction. Data lock-in is the loss of something irreplaceable – a history of transactions, behavioral signals, institutional knowledge, or benchmarking context that a new vendor cannot replicate. The distinction matters enormously for displacement risk modeling.

III. The Businesses That Will Compound

The following analysis identifies the business model attributes that create durable competitive advantage in an AI economy. These are not simply ‘good businesses’ in the traditional sense – they are businesses whose specific structural characteristics make them more valuable, not less, as AI capabilities expand.

Category A: Proprietary Data Owners

The single most durable source of competitive advantage in an AI economy is owning a proprietary dataset that competitors cannot replicate. AI models are only as good as the data they are trained and grounded on. A business that has accumulated years of unique transactional, behavioral, or domain-specific data sits at the top of the value hierarchy, because its AI outputs will be structurally superior to any competitor’s – regardless of how capable the underlying model is.

The data moat compounds over time according to the following dynamic:

The Data Moat Value Equation

Where

V_data

Data Moat Value

The economic value of the proprietary data asset. It grows exponentially at rate λ as the dataset accumulates more transactions, more behavioral signals, and more training examples — creating a compounding advantage that cannot be purchased or replicated from scratch.

Knowledge Sensitivity

How much the firm’s customers value the firm’s data-derived insight relative to generic alternatives. High in investment management, clinical healthcare, legal, financial services, and domain-specific industrial applications.

I_signal − I_noise

Signal-to-Noise Ratio

The net signal content of the proprietary dataset. Competitors using generic data or public information have lower (I_signal − I_noise). A firm with a proprietary dataset can train models with structurally higher signal, producing outputs that are simply better than what any AI built on public data can match.

e^λt

Compounding Factor

The exponential growth in data moat value as t (time operating and collecting data) increases. This is the most important term for investors: the data advantage compounds with each passing year of operation, creating an increasing gap between early data accumulation leaders and late entrants.

Examples of durable data moat businesses: Financial data aggregators with proprietary trading or behavioral data; clinical data platforms with longitudinal patient records; real estate databases with decades of transaction history; specialized insurance platforms with actuarial loss history; logistics networks with years of route and demand data.
What to look for in due diligence: How many years of proprietary data does the company hold? Is it structured and annotated, or raw? Can it be replicated by a well-funded new entrant within five years? Is the data growing in volume and quality as a direct by-product of the company’s core operations?

Category B: Businesses That Generate Valuable Data as a By-Product of Their Core Service

A related but distinct category: businesses in which delivering the service continuously generates new proprietary data, and that data has value beyond the original transaction. These businesses are doubly advantaged in an AI economy: they use AI to improve service delivery (reducing cost and improving quality), and that improved service generates more data, which further improves the AI, in a virtuous loop that competitors cannot easily enter.

The defining characteristic: The data is a natural exhaust of the transaction, not a separately harvested asset. It grows without friction as the business scales.
Business model examples: Wealth management platforms where every portfolio decision and client interaction generates behavioral and outcome data; legal research platforms where every query and precedent search enriches the training corpus; healthcare diagnostics businesses where each scan or test result improves the model; supply chain platforms where every shipment generates route optimization and vendor reliability data.
Investment implication: When evaluating these businesses, the most important question is not ‘what is the current revenue?’ but ‘how valuable is the data being generated, and what does it enable that competitors cannot replicate?’ The data asset may be worth more than the current business model captures.

Category C: End-Relationship Owners

There is a meaningful structural difference between businesses that own the customer relationship directly and businesses that serve other businesses that ultimately serve end customers. End-relationship owners – those with direct accountability to the end consumer or client, with trusted access to their most sensitive information and highest-stakes decisions – occupy a position that AI cannot displace. They are the locus of trust, accountability, and judgment.

The relational value model is directly applicable here. End-relationship value is the composite of trust, alignment, and integrity – and these are multiplicative. A business that loses trust or alignment collapses its relational value to near-zero regardless of its functional capability. This makes end-relationship ownership one of the most durable competitive positions in any economy.

End-Relationship Value

V_end = v^rel + v^out + v^net + ξ(S_perceived − S_expected)

What end-relationship ownership looks like: Investment advisors and wealth managers with fiduciary accountability; primary care physicians and specialist networks with longitudinal patient relationships; tax and legal advisors with years of institutional knowledge about a specific client; commercial real estate advisors who know a client’s space needs, culture, and growth plans intimately.
Why AI strengthens rather than displaces: AI augments the end-user by making them faster, better informed, and more proactive. It does not replace them because the customer’s willingness to act on a recommendation is grounded in trust in the specific human advisor, not trust in the AI model. The liability and the relationship both remain human.
The vulnerability to watch: End-relationship owners who fail to adopt AI will be disadvantaged relative to those who do – not because AI replaces the relationship, but because AI-augmented advisors can serve more clients with better outcomes, thereby compressing the market share available to non-adopters.

Category D: Businesses Requiring Coordinated Judgment Across Complex, High-Stakes Decisions

The Mathematics of Work identifies σ_i – the judgment requirement – as the primary determinant of automation resistance at the task level. At the business model level, the equivalent concept is: businesses whose core value requires the simultaneous coordination of many high-σ_i tasks, where the interaction effects between decisions are complex, the stakes are high, and the accountability is legally or ethically non-delegable. These businesses are structurally insulated from AI substitution.

Characteristics: Multiple interacting high-judgment decisions must be made in coordination, not in sequence. The error cost is high (legal liability, regulatory sanction, client harm). The context is specific to a particular client, case, or situation and resists generalization.
Business model examples: Investment banking M&A advisory (legal, financial, regulatory, cultural, and strategic judgment must all be coordinated simultaneously); complex litigation (strategy, evidence assessment, client management, and negotiation all interact); institutional investment management where the client relationship, risk tolerance, tax situation, and market dynamics must be continuously balanced; crisis communications where reputation, legal exposure, and narrative must be managed in real time.
Why AI strengthens rather than displaces: AI handles the J_A component – research synthesis, document drafting, scenario modeling, data analysis – freeing the human judgment center to focus entirely on J_H: the coordinated, high-stakes decisions that create the actual value. The result is a better outcome for clients at lower cost for the firm.

IV. Comprehensive Attribute Table: Businesses Positioned to Thrive

The following table consolidates the full set of business model attributes that create durable competitive advantage in an AI economy. These are the characteristics we look for in portfolio companies and acquisition targets.

BUSINESS ATTRIBUTE	WHY IT MATTERS IN THE AI ECONOMY
Owns irreplaceable proprietary data	Transactional, behavioral, or domain-specific data accumulated over years cannot be replicated by new entrants or general-purpose AI. The data moat compounds annually, widening the performance gap between the data owner and any competitor.
Generates proprietary data as a natural by-product of operations	Every transaction, client interaction, or service delivery enriches the firm’s training corpus without incremental cost. AI model quality improves automatically as the business scales, creating a self-reinforcing advantage loop.
Owns the direct end-customer relationship with fiduciary accountability or high dependency	When the customer’s trust is legally and ethically grounded in a specific firm or individual – not just in a tool or platform – AI cannot sever the relationship. It can only augment the advisor’s capability within it.
Operates in a regulated or high-liability environment	Regulatory requirements and legal accountability preserve the human-in-the-loop requirement for high-stakes decisions regardless of AI capability. Compliance, fiduciary duty, and professional liability standards are structural demand signals for human judgment.
Derives value from network effects across a curated participant base	Network value (V^net = η·n(n-1)) compounds as participants grow, creating barriers that AI cannot substitute. Curated networks – where quality of participants matters, not just quantity – are especially durable because curation requires judgment.
Provides multi-dimensional, coordinated judgment across complex decisions	When business value requires the simultaneous coordination of many high-σ tasks – each interacting with the others – no AI agent can replicate the outcome because the interaction effects themselves are the value. Investment advisory, M&A, complex litigation, and enterprise architecture are examples.
Embeds deeply into customer workflows with high switching cost	High Φ_switch — switching cost — protects incumbents even in businesses with moderate automation risk. Deep API integrations, proprietary data formats, and years of institutional memory embedded in the product create barriers that AI cannot reduce.
Serves mission-critical functions where error cost is catastrophic	Businesses in the critical path of a customer’s most important operations command switching costs that are driven by risk aversion, not just friction. The customer cannot afford to be wrong about a replacement – so they do not switch.
Owns or aggregates scarce, non-substitutable inputs	Physical assets, exclusive licenses, proprietary methodologies, or access rights that competitors cannot replicate create permanent differentiation regardless of AI capabilities. AI can optimize the use of scarce inputs but cannot create new ones.
Has built a brand associated with trust, quality, and accountability	In high-stakes contexts, brand trust is a form of relational value (V^rel) that commands a premium independent of functional output quality. AI can produce equivalent functional outputs; it cannot replicate thirty years of reputation built on accountable delivery.
Uses AI to expand the value proposition, not just reduce cost	Businesses that deploy AI to offer services that were previously economically infeasible – expanding n in the value chain rather than shrinking headcount – achieve compounding revenue growth alongside margin improvement.
Serves human life-stage or high-emotion decision contexts	Retirement planning, estate management, medical decisions, educational transitions – decisions that are emotionally weighted, infrequent, and irreversible — require human presence regardless of AI capability. The customer’s ξ (experience sensitivity) in these contexts is exceptionally high.

V. The SaaS-Pocalypse: What It Gets Right

The displacement thesis is correct about a specific, large, and economically significant category of software businesses: those whose entire value proposition is helping users perform repetitive, bounded, function-specific tasks within a single business area – without owning meaningful data, without network effects, and without a direct relationship with the end customer they ultimately serve.

These businesses are not failing because AI is uniquely hostile to them. They are failing because they were always essentially workflow wrappers – products that abstracted the difficulty of a task without creating lasting economic value in the form of data, relationships, or network effects. The margin they captured was a tax on friction. AI eliminates the friction. The tax disappears.

Key Insight

A workflow-automation SaaS product whose core value is ‘we make this repetitive task easier’ has always been competing against the clock on its own obsolescence. Every improvement in AI capability reduces its value proposition. The SaaS-pocalypse is not a new threat – it is the culmination of a race the product was always losing.

VI. Comprehensive Attribute Table: Businesses That Will Struggle

The following attributes identify business models that face structural pressure as AI capabilities expand. This is not a prediction that every company in these categories will fail immediately — it is a prediction that their competitive moats are eroding and their valuation multiples will compress.

BUSINESS ATTRIBUTE / MODEL	WHY IT MATTERS IN THE AI ECONOMY
Narrow workflow automation within a single business function, especially when there is a dominant , established System of Record in market	Products that automate a specific, bounded task — expense reporting, contract redlining, basic scheduling, email drafting, data entry — without owning the underlying data or the customer relationship are the most direct AI substitution targets. AI agents perform these tasks natively.
Sits between a company and its own data prohibiting the company from using AI to augment its value proposition	Products that serve primarily as an interface layer — helping users query, visualize, or manipulate data that the customer already owns — are structurally vulnerable. AI provides this interface natively. The product’s only value was the wrapper; AI removes the need for the wrapper.
Depends on feature complexity or a large catalogue of integrations rather than data , hard to replicate differentiation, or network as its moat	Products that hold customers through the complexity of their feature set — rather than through irreplaceable data or network access — face commoditization risk as AI makes feature replication dramatically cheaper and faster. Feature complexity is not a moat; it is a temporary migration cost.
Provides a point solution for a workflow that AI handles holistically	Businesses that built point solutions for one step in a multi-step process are vulnerable to platforms that AI enables to handle the entire process end-to-end. The economic logic favors the holistic solution: it eliminates handoff costs, captures all the data, and delivers a superior customer experience.
No proprietary data generated by or stored within the product or dependency on another product for valuable data	If the product’s core functionality does not generate or accumulate proprietary data, there is no compounding advantage and no data lock-in. Any AI-enabled alternative that offers equivalent functionality has no data gap to overcome, making competitive replacement purely a pricing and UX decision.
Primarily serves back-office or support functions with no customer-facing component	Back-office workflow tools — HR administration, accounts payable processing, basic legal document management, IT ticketing — are among the highest-priority targets in every enterprise AI deployment program. The AOV (Automation Opportunity Value) for these tasks is large and the judgment requirement (σ_i) is low.
Sells to departments that are themselves under automation pressure	Vendors that primarily sell to functions facing significant AI-driven workforce reduction – large back-office teams, manual data processing groups, basic customer service organizations – face both a shrinking budget pool and a customer base that is under pressure to replace them as part of its own transformation.
High price relative to AI-substitutable functional value	Products priced at a premium relative to the functional output they deliver are especially vulnerable when AI provides equivalent functional output at dramatically lower cost. Price compression will be rapid once enterprise buyers demonstrate that AI alternatives produce acceptable results.
Built on an integration moat that AI orchestration threatens	Some products derive their value primarily from managing integrations between other systems. AI-native orchestration layers are reducing the need for dedicated integration middleware, as AI agents can now navigate APIs and data sources directly without purpose-built connectors.
Lacks a direct end-customer relationship — is two or more steps removed	Businesses that serve the enterprise customers of other enterprises – especially in generic functional categories – have no relational anchor when the buyer decides to automate the function. There is no V^rel to preserve. The buying decision becomes purely economic, and purely economic decisions favor lower-cost alternatives.
Category-defining product that has become a commodity workflow	Many early SaaS category winners – the first product to digitize expense reports, contract management, or basic CRM – are now category definitions rather than differentiated products. When a category is fully defined and understood, an AI agent can implement its core workflow from scratch in days. Category definition is not defensibility.
Revenue model depends on seat-based pricing tied to headcount	Seat-based pricing is directly exposed to the headcount reduction effect of AI automation. As AI reduces the number of humans who need to use the product, the total addressable seat count contracts. The product may retain value per seat while simultaneously losing seats – creating a unit economics problem that cannot be solved by price increases.

VII. The Investment Implication: Valuation Multiple Divergence

The most significant investment implication of the AI economy is not that some businesses will fail. It is that the valuation premium commanded by durable businesses will expand while the multiple assigned to at-risk businesses compresses – and that this divergence will happen faster and with more permanence than most investors currently anticipate.

The mathematical structure of this divergence is straightforward:

Where

ΔEV

Enterprise Value Divergence

The gap in enterprise value between AI-resilient businesses (winners) and AI-exposed businesses (losers). As μ_winner·ΔΠ_winner for winners grows (improving profitability at expanding multiples) and μ_loser·ΔΠ_loser for losers contracts (declining profitability at compressing multiples), the EV gap widens non-linearly.

μ_winner·ΔΠ_winner

Winner: Multiple Times Profit Growth

AI-resilient businesses benefit from both margin expansion (automation lowers their own cost structure) and multiple expansion (investors pay higher multiples for businesses with data moats, network effects, and end-customer relationships that compound in value as AI expands). Both effects run simultaneously.

μ_loser·ΔΠ_loser

Loser: Multiple Times Profit Decline

AI-exposed businesses face simultaneous margin compression (AI-enabled competitors undercut price) and multiple compression (investors reduce the multiple they will pay for businesses with eroding moats). Both effects are self-reinforcing and accelerate as displacement becomes visible.

For investors, the practical implication is a barbell portfolio strategy: overweight businesses with strong M_durable scores (proprietary data, end-customer relationships, network effects, coordinated judgment requirements) and underweight businesses with high DRI scores (automatable workflows, no data lock-in, no end-customer relationship). The middle — businesses that are neither clearly durable nor clearly at risk – requires the most careful individual assessment.

One additional calibration: the timing of competitive pressure is not uniform. Businesses serving large enterprise customers with long procurement cycles and high switching costs will experience displacement pressure more slowly than businesses serving SMB customers who can switch in 30 days. The DRI score tells you the magnitude of risk; the customer profile and switching cost tell you the timing.

Key Insight

The most dangerous investment thesis in the current environment is ‘this business is too important to displace.’ Importance is not the same as irreplaceability. A narrow workflow SaaS used in thousands of enterprises is important — and replaceable. The correct test is not importance but the M_durable score: where does the value come from, and is that source durable?

VIII. The BIP Capital Portfolio Thesis

At BIP Capital, the analytical framework described in this article directly informs how we evaluate businesses. We are looking for companies that score well on the durable attributes table and poorly on the displacement risk table. We are specifically seeking businesses that exhibit three or more of the following compounding characteristics simultaneously:

A proprietary data asset that grows naturally as a by-product of operations, combined with a demonstrated ability to use that data to deliver customer outcomes that competitors cannot replicate.
A direct, trusted, accountable relationship with the end customer, ideally in a regulated or high-liability context where the legal and ethical structure of the relationship preserves the human layer.
Network effects that compound with participant growth and are curated — meaning the quality of participants matters, creating a judgment-dependent selection process that AI cannot automate away.
AI adoption that is already underway and being used to expand the value proposition rather than simply reduce costs – businesses already executing the Model 3 (Value Chain Expansion) dynamic.
A management team that understands the distinction between automation for margin and redeployment for growth, and that is explicitly managing the L_r / L_c allocation.

Conversely, we are cautious about businesses that have historically competed on workflow simplification, feature complexity, or integration management without building the underlying data and relationship assets that create lasting value. These businesses may be excellent operators with strong management teams – but they are fighting the AI tide rather than riding it.

“The great sorting of the AI economy is not between technology companies and traditional companies. It is between businesses that own irreplaceable data, relationships, networks, judgment — and businesses that only owned a friction point. AI eliminates friction. It cannot eliminate irreplaceability.”

Conclusion: A Measured Assessment

The SaaS-pocalypse is real, but it is not the whole story. It is the story of one specific category of business — narrow workflow automation without data ownership, network effects, or end-customer relationships — facing a structural substitution threat that compounds with each improvement in AI capability. For that category, the bears are right.

But for the broader universe of businesses that build genuine value through proprietary data, trusted customer relationships, curated networks, and the coordinated human judgment that high-stakes decisions require, AI is an amplifier, not a threat. These businesses will use AI to expand their value propositions, reduce their costs, and widen the gap between themselves and competitors who either lack the data assets to benefit fully or face displacement of their core workflows.

The investment opportunity in this transition is significant and time-sensitive. Valuation multiples have not yet fully diverged between these two categories. The businesses with the highest M_durable scores — proprietary data owners, end-relationship holders, network-effect businesses, and coordinated judgment providers — are still being evaluated by many investors using frameworks built for a pre-AI economy. That mispricing will not persist. The analytical tools in this article series give investors the framework to identify and act on it before the market fully adjusts.

As always, we are interested in connecting with business leaders and investors who are rigorously thinking through these questions. The analysis gets better in conversation.

‍

Next IN PATH