Is AI going to kill your favourite SaaS company?

Our AI-resilience checklist

Mar 20, 2026

AI is advancing at lightning speed.

In just a few months, the narrative that AI is making software companies even stronger has been called into question by the release of Claude’s latest capabilities, which have reshaped expectations across the Software landscape.

We’re seeing market disruption happening in real time.

And I’ve been using Claude Code for a couple of months now, also for stock analysis, and I do believe the disruption is real.

It’s no longer: “Can I build it?” It’s should I build it?

But the breakdown of software companies won’t come overnight. Multiples will have a hard time rerating. Most of them will die slowly.

While this industry drawdown is playing out in real time, every investor should be asking the same two questions:

How do I know if a company is likely to be disrupted by AI?
Which names are being wrongfully sold off in this industry drawdown?

To answer both questions, we developed the following checklist.

Checklist

The Checklist consists of nine lenses across four tiers, with three exposure types (Exposed, Neutral, Antifragile).

We assigned a weight to each of them

Tier weights:

The scoring system is as follows:

Exposed = -1 point
Neutral = 0 points
Antifragile = +1 point

Score range: -13.5 to +13.5 (weighted)

Let’s dive into the framework…

Tier 1: Survival - Do you still have a business in 5 years?

1. Pricing Model (The “Cash Flow” Pillar)

Question: Does the company charge for an outcome or per user seat?

This is the most immediate threat to cash flow. If you charge per human seat and AI makes humans 10x more productive, your revenue will mathematically collapse by 90%. The shift from paying for work to paying for outcomes is the difference between a thriving business and a dying one.

Exposed: The company charges its customers solely per seat. Revenue is directly tied to headcount, which AI will compress.
Neutral: The company blends seat-based and usage-based charges. Some revenue still scales with headcount, but AI-driven efficiency can be offset by fees tied to data volume, API calls, or feature usage.
Antifragile: The company charges based on consumption, outcomes, or value delivered (compute, transactions, storage, data volume). AI adoption amplifies revenue as usage grows.

2. Failure Sensitivity (The “Liability” Pillar)

Question: How accurate does AI need to be to replace this software?

This defines the “Trust Gap”. AI is probabilistic; if a guess results in a typo in a blog post, it’s fine. If it results in a plane crash or a cybersecurity breach, it’s a disaster.

Exposed: 90% accuracy is good enough for the company’s core use case (e.g., marketing copy, basic graphic design). The software is likely to be replaced by AI.
Neutral: AI needs to be consistently reliable (97-99% accurate), but occasional errors are tolerable with human review. The software is partially defensible.
Antifragile: Failure or hallucination is catastrophic (e.g., medical diagnostics, cybersecurity, power grid management, financial compliance). The software is harder to replace with a probabilistic AI model.

Tier 2: Defence - Preventing competitors from eating your lunch

3. Workflow Integration (The “Stickiness” Pillar)

Question: How many system intersections does it control?

Even if an AI tool is better, a company won’t switch if that software is the central nervous system connected to 50 other apps. Replacing a core workflow is like performing heart surgery; most companies would rather just upgrade their current heart with AI.

Exposed: The software operates as a standalone tool with minimal integrations, few dependencies, and low workflow entanglement. It is easier to replace.
Neutral: The software connects to multiple systems but does not serve as the central coordination layer. Replacing it creates friction, but not systemic breakdown.
Antifragile: The software sits at the intersection of many systems, acting as a coordination hub or workflow backbone. Replacement becomes structurally difficult.

4. Switching Cost (The “Friction” Pillar)

Question: How much accumulated value would a customer lose by leaving?

This is distinct from workflow integration. A product can have few integrations but enormous switching friction due to years of accumulated data, configurations, trained employees, and organisational muscle memory. The question isn’t how connected it is? It’s how painful it is to walk away.

Exposed: Customer data is easily exportable, minimal customisation is built on top, and retraining cost is low. A competitor can replicate the experience with minimal migration effort.
Neutral: Customers have a moderate level of accumulated investment, such as historical data, custom configurations, or trained workflows, but migration is feasible within a quarter with dedicated effort.
Antifragile: Customers have years of accumulated data, configurations, custom automations, compliance audit trails, and deeply trained teams. Migration would require months of effort, retraining thousands of employees, and risking operational disruption. The cost of switching far exceeds the cost of staying.

5. Customer Base (The “Inertia” Pillar)

Question: Who does it primarily serve?

Enterprise clients aren’t a moat on their own. However, the cost of software is often negligible compared to retraining 10,000 employees and reconfiguring 100 internal processes. There is a reason why so many people still use Microsoft solutions for use cases where better alternatives exist.

Exposed: The software mainly serves individuals or small businesses. Switching costs are lower, and experimentation is higher.
Neutral: The company serves both SMB and enterprise clients. Defensibility depends on contract structure, integration depth, and switching friction.
Antifragile: The software primarily serves large enterprises. Displacement is significantly harder. Enterprise environments involve procurement cycles, compliance requirements, multi-year contracts, security audits, and cross-department integration.

6. Network Effects (The “Ecosystem” Pillar)

Question: Does the product become more valuable as more participants join the ecosystem?

A product with true network effects creates a self-reinforcing flywheel that is difficult to replicate. It’s not enough to have integrations (that’s Workflow Integration) or switching costs (that’s Friction). This lens measures whether the ecosystem itself compounds in value as scale increases. A marketplace with 10,000 apps is not just 10x better than one with 1,000. There’s a quality difference.

Exposed: The product is a standalone tool with no meaningful ecosystem dynamics. Value is linear; each customer gets the same experience regardless of how many others use the platform. No integration marketplace, no third-party developer community, no cross-customer data benefits.

Neutral: The product has some ecosystem elements, a modest integration marketplace, partner channel, or community, but these are additive rather than compounding. Competitors could replicate the ecosystem with sufficient investment. The product would still function without the ecosystem.

Antifragile: The product sits at the centre of a self-reinforcing ecosystem where value compounds with scale.

Tier 3: Offence - Can AI make this company stronger?

7. AI Adoption Capability (The “Offence” Pillar)

Question: How effectively is the company integrating AI to strengthen its product and expand its market?

Defence alone isn’t enough. The companies that will win the AI era are the ones that use AI to widen their moat by embedding it into their product to deliver more value, expanding into adjacent markets, or creating entirely new revenue streams. A company that is antifragile on defence but slow on AI adoption can still lose to an incumbent that moves faster.

Exposed: The company has no meaningful AI strategy, no AI features shipped, and no public roadmap for AI integration. It is passively waiting for disruption rather than responding to it.
Neutral: The company has announced or shipped AI features (e.g., copilots, assistants, AI-generated suggestions, AI-powered automation), but these are incremental enhancements to existing products. AI is a marketing message and a feature, not a proven revenue driver. Most software companies in 2025-2026 fall into this category.
Antifragile: The company has demonstrated measurable AI impact on its business by not just shipping features, but also proving results. Only a few companies to this day have this.

Tier 4: Alpha - Secondary moats that compound over time

8. Data Ownership (The “Intelligence” Pillar)

Question: How much proprietary data is involved?

Data is only useful if it’s proprietary and clean. If your data is just a mirror of what’s on the public web, it’s not a moat. It only matters if it’s “under-the-hood” data that public models can’t see.

Exposed: The software relies primarily on public, user-exportable, or easily replicable data.
Neutral: The company holds some proprietary data, but it is customer-specific and not meaningfully aggregated across the network.
Antifragile: The company controls proprietary, network-aggregated data that improves as more users join. It benefits from strong data gravity. The system becomes smarter at scale in ways public AI models cannot replicate.

9. Regulatory Exposure (The “Red Tape” Pillar)

Question: Do legal or compliance barriers protect the business?

Regulation is a speed bump moat. It keeps small startups out because they can’t afford the legal fees, but it doesn’t stop a determined incumbent. It’s a great secondary defence, but a weak primary one.

Exposed: The software operates in a lightly regulated environment with minimal legal constraints.
Neutral: The company operates in a regulated environment, but compliance is standardised and replicable. Regulation slows entrants but does not prevent disruption.
Antifragile: The software operates in a highly regulated domain where legal approval, certifications, audits, and liability structures are complex and time-consuming. Regulation becomes a structural moat.

Stress testing HubSpot, SYS-DAT, and Rubrik

Now that we have established the framework, we are going to test it on three companies:

HubSpot is a CRM platform that helps small to medium-sized businesses attract customers, automate marketing and sales, and manage relationships all in one integrated system.
SYS-DAT is an Italian IT services company that supports manufacturing businesses in Italy through ERP solutions, cybersecurity, cloud infrastructure, and data management.
Rubrik is a cybersecurity company that protects and manages enterprise data by providing backup, recovery, and ransomware resilience across cloud and on-premise environments.

Walk through the checklist

HubSpot is down 35% since the beginning of the year. Let’s see how well their moat holds up against AI:

1. Pricing Model

This is the Achilles’ Heel. HubSpot’s revenue is historically tied to Core Seats. If AI allows a small team to do the work of a department, the customer will naturally reduce their seat count, directly eroding HubSpot’s top-line revenue. They are attempting to shift toward platform-wide fees, but the seat-based legacy model remains high risk.

Verdict: Exposed

2. Failure Sensitivity

HubSpot’s primary use cases, like marketing emails, blog posts, and sales outreach, are probabilistic tasks. If an AI generates a draft that is 90% correct or the customer information has minor errors, it’s not the end of the world. However, there will still be human interaction needed, since customer information and blog posts vary and are customer-facing.

Verdict: Neutral

3. Workflow Integration

This is HubSpot’s greatest strength. It acts as the System of Record for the mid-market, connecting marketing, sales, and service. With thousands of App Marketplace integrations, pulling HubSpot out of a business often causes a systemic breakdown in operations.

Verdict: Antifragile

4. Switching Cost

HubSpot primarily serves the 20–500-employee segment, where years of CRM data, sequences, and workflows create meaningful switching friction. Switching is painful but not prohibitive, especially for the smaller end of HubSpot’s base, where customisation depth is lower and new tools could accelerate the process.

Verdict: Neutral

5. Customer Base

HubSpot is the king of the mid-market. While these businesses are stickier than individuals, they are also the first to experiment with lean, AI-only marketing teams to save on overhead during industry shifts. HubSpot primarily utilises annual contracts for its professional and enterprise tiers, while offering more flexible monthly options for starter plans.

Verdict: Neutral

6. Network Effects (The “Ecosystem” Pillar)

HubSpot’s 1,900+ connector marketplace and growing Breeze agent ecosystem are real, but they are additive rather than compounding. The integrations are overwhelmingly connectors to third-party tools like Slack, Google Workspace, Salesforce, and Zapier that push and pull data in and out of HubSpot, rather than ISVs building differentiated, standalone products on top of HubSpot that customers depend on in their own right.

Verdict: Neutral

7. AI Adoption Capability

Breeze Customer Agent resolves over 50% of support tickets during early HubSpot deployments. However, no AI-specific revenue has been disclosed, and the features (copilots, assistants, summarisation) are broadly similar to those offered by every major CRM competitor. Strong execution, but not yet a proven revenue driver or new-market unlock.

Verdict: Neutral

8. Data Ownership

HubSpot holds a massive amount of customer-specific data, but it is siloed. While they have launched “Breeze” to help aggregate insights, the data is ultimately user-owned and exportable. They lack a “network-aggregated” moat where one user’s data makes the tool significantly smarter for everyone else (unlike a cybersecurity firm).

Verdict: Neutral

9. Regulatory Exposure

HubSpot primarily operates in marketing, sales, and customer service - lightly regulated activities with no meaningful certification barriers. Any well-capitalised AI-native startup can compete in this space.

Verdict: Exposed

Weighted Score = (Tier 1 sum x 2) + (Tier 2 sum x 1.5) + (Tier 3 sum x 1.5) + (Tier 4 sum x 1)

HubSpot gets a final score of -1,5 (Score range: -13.5 to +13.5)

Now let’s walk through the checklist with SYS-DAT, which has been sold off by 20% ytd. You can read our deep dive here.