Invest · AI Disruption Due Diligence
What happens to this thesis when intelligence gets cheap?
A specialised layer on top of technology and AI due diligence, for cases where the investment case rests on an advantage that cheaper software and automation could erode.
You are not buying the target’s current product. You are buying its future ability to defend cash flow.
What this adds to conventional work
Conventional technical diligence describes the asset as it stands. This asks a different question: what is the asset worth once a competent team with frontier tools can approach the same customer-visible result on a fraction of the budget — and once every competitor gets the same productivity gain at the same time?
That is the gap. A company can pass every conventional test and still be a poor holding, because its differentiation depreciates faster than its cost base declines.
It applies to a machinery business as much as to a SaaS business. In software the exposure usually sits in reproducible functionality. In industry it sits in engineering throughput, knowledge work, service processes and the customer interface — where an AI-enabled competitor can change the cost of serving the same customer without touching the physical product.
What it is not
It is not a claim that software companies are finished, or that mission-critical systems lose their value. The opposite is frequently true: as generated code becomes abundant, the things that make a system dependable — installed base, process integration, operational reliability, domain knowledge, security, accountability — become the scarce part. Separating the two is most of the work.
01
Dimensions examined
Weighted to the target. A software business and a machinery business get the same method and a different centre of gravity.
Reproducibility
- The 80% rebuild test, applied concretely to this product or service
- AI substitutability of customer-visible functionality
- Build-cost compression, present and projected
- Platform absorption risk
What is actually owned
- Technology moat, separated from technology quality
- Proprietary data: irreproducibility rather than volume
- Distribution and control of the customer relationship
- Workflow embedment, physical integration and operational switching costs
Industrial and operational exposure
- Engineering and design compression, and what competitors do with it
- Knowledge-work and administrative automation
- Service, support and customer-interface automation
- Robotics and process automation potential, including the cases that do not pay
Economics
- Pricing power under falling production cost
- Who captures the compression: company, customer, competitor, employee or platform
- Margin and EBITDA implications, including the ones that reverse
- Workforce implications by function
Forward view
- Likely competitive response and the arrival of AI-enabled entrants
- Three-to-five year disruption scenarios
- Which parts of current enterprise value are exposed
- Where AI creates an opportunity this company is positioned to take
02
How the work runs
Scope, duration and depth follow the target and the stage of the transaction. The sequence does not change.
The thesis in the buyer’s own words
Before any technical work: what is being underwritten, at what price, on what assumptions. The assumptions are what gets tested — the diligence is only useful if it can contradict them.
Evidence
Product, architecture, data assets, telemetry, contracts, roadmap, engineering and operating economics. Management sessions, and customer conversations where access allows. Reading what the company does, not only what it says.
Adversarial reconstruction
What would it take, concretely, for a well-funded team with frontier tools to reproduce what customers pay for? The points at which that reconstruction fails are the moat. The points at which it succeeds are the exposure.
Scenarios
Three-to-five year cases under different assumptions about compression speed, platform behaviour and competitive response — including the case where the thesis is right and the case where it is early.
Read-out
Findings, red flags, the opportunities the thesis has not priced in, and the questions that remain open. Delivered as an argument that can be attacked, not as a rating.
03
What you receive
Written work, presented and defended in person to the deal team or investment committee.
- AI Disruption Map — where value is exposed and where it is anchored
- Technology moat assessment
- Platform absorption risk assessment
- AI compression view and its margin consequences
- Three-to-five year scenario analysis
- Direct challenges to the investment thesis
- Red flags
- Strategic opportunities the current plan does not capture
These are the shape of the work, not a fixed package. Which of them matter depends on the target, and saying so in advance is part of the job.
04
Where this is worth doing
A good fit
- Software or technology-enabled targets where the moat is asserted rather than demonstrated
- Categories a hyperscaler or frontier platform could plausibly absorb
- Industrial targets whose competitors could reach the same customer far more cheaply with AI
- Theses that rely on productivity gains reaching EBITDA
- Assets where the value sits in data, workflow or distribution and nobody has separated those from the product
A poor fit
- A code audit or a security assessment — necessary work, but not this work
- Confirmatory diligence intended to support a decision already made
- Situations where no independent judgment can change the outcome
Challenge an investment thesis
Describe the target, the stage of the transaction and the part of the thesis that would hurt most if it were wrong.
For private equity, growth equity, family offices, strategic acquirers and boards.