Key Takeaways
- Byron Ling evaluated roughly 30,000 founder meetings across a decade, concluding that pedigree matters far less than an earned right to market insight.
- First-level thinking reacts to headline valuation drops or broad AI commoditization fears, while second-level thinking asks what the consensus misses about value capture.
- Customer diligence requires testing ground-level awareness: asking buyers whether they even recognize foundational model names like Sonnet, Gemini, or Codex exposes whether platform threats are real or theoretical.
- Enduring moats require founders to present a ten-year economic roadmap showing how an initial entry wedge converts into long-term gross margin defense.
- Twelve Below applies Value Chain Profit Pool Mapping to pinpoint where operating margins pool across an industry before backing early-stage software companies.
The Value Chain Profit Pool Mapping
To separate temporary software features from defensible businesses, Ling adapts Howard Marks' second-level investing philosophy into a structured economic mapping process.
Component 1: X-Axis Value Chain Mapping
Plot all distinct stakeholders, intermediaries, suppliers, and buyers across an entire industry's value chain along the horizontal axis.
Component 2: Y-Axis Margin Distribution
Map the gross and operating margin profile of each stakeholder group along the vertical axis to visually identify where profit pools are concentrated.
Component 3: Startup Wedge and Defensibility Intersection
Determine where the startup's product wedge enters the value chain, how it shifts margin capture over 10 years, and whether its integration points protect against platform commoditization.
Ling explains the origin of this discipline: “I borrowed this from an essay Howard Marks wrote years ago, which is first level thinking is really if a stock price drops, everyone assumes you should sell the stock, where second order thinking forces you to say, what do you believe and what do you believe relative to the market?”
He pairs this analysis with founder insight: “What we find is the excellent entrepreneurs have become obsessed about the market. Sometimes that's through professional experience, in many cases it's just from hard work and research. And so the responses that we prefer to hear are 'I've earned the right to this insight.'”
When evaluating competitive threats from foundational models, Ling checks whether end buyers care about underlying tech layers: “One of the frameworks I use if you're thinking about that threat is go talk to a bunch of customers in that end market and ask them do they even know the names of the different models? Do they even know what Codex is? Do they even know what Gemini or Sonnet or Oak Pass [is].”
When This Works (and When It Doesn't)
Value Chain Profit Pool Mapping works best when evaluating startup business models to test whether an initial product wedge can expand into high-margin profit pools rather than getting trapped in low-margin layers. It excels in legacy vertical markets like logistics, construction, or healthcare administration, where value chains are fragmented and intermediate distributors capture excessive margin without software defense.
It struggles in fast-moving platform shifts where profit pools have not settled. If an infrastructure layer or API layer captures all the rents before application software can establish workflows, mapping historical margins gives a false sense of security. Historical profit pools show where money went yesterday, not where open source models or platform aggregators will redirect it tomorrow.
Why It Matters
Early-stage venture capital frequently mistakes rapid product adoption for long-term equity value. A startup can acquire initial users by automating a tedious workflow, yet capture zero economic surplus if the underlying value chain forces margins down to commodity levels. Ling's second-level lens shows that real defensibility is determined by value chain distribution, deep workflow integrations, and end-customer inertia, not by raw technical novelty or founder credentials.