Earlier this week I guest-lectured for Cornell’s SYSEN 5213, Designing Robust Ventures course where students had already been introduced to R1 — the first of the nine requirements for creating ventures with a viable Core Market Architecture
R1: Circumvent the Critical Cost Driver — Shape the venture’s product form factor to remove the one cost most likely to prevent profitability at scale — not optimize it, eliminate it entirely.
The concept is simple to state. It’s much harder to apply in practice, when a technically sophisticated mechanism gets confused for a structurally relevant one. This post walks through 5 real world cases and the diagnostic checklist we used in the lecture to pull the two outcomes apart.
The Scoring Rubric
As a preface, here’s the five-point scale we used across the cases, drawn from an evaluation framework we built for investors to score how ventures solve individual requirements and overcome the associated “market barrier” that prevents a profitable market from forming.
| Score | Label | What it means |
|---|---|---|
| +2 | Productized | The venture’s product directly eliminates or greatly reduces the market barrier, with minimal ongoing operational effort |
| +1 | Addressed | A clear, deliberate strategy materially reduces the barrier, but requires ongoing operational effort to sustain |
| 0 | Ignored | No strategy beyond the industry default — the venture inherits the barrier as-is |
| -1 | Amplified | The venture’s approach makes the barrier worse than the industry default |
| -2 | Structural | The venture’s approach makes the barrier permanent — profitability on this dimension is structurally blocked |
The distance between +2 and +1 matters more than it looks: a +1 fix scales its cost with volume (more customers, more headcount to sustain it), while a +2 fix doesn’t. That distinction runs through every case below.
Same Company, Opposite Outcomes
The clearest illustration didn’t come from a pitch deck — it came from two products I worked on while at Facebook: Marketplace (which grew out of C2C Groups) and Order Food on Facebook (which shut down, following a pattern for nearly every other B2C transact-on-Facebook product).
Same company. Same distribution reach. Same engineering talent. Same starting point — a product idea layered onto Facebook. Only one of them worked, which highlights a core variable that drove success or failure: did the product restructure the real cost barrier, or just add a new interface on top of an unchanged one?
Marketplace solved a genuine, decades-old cost driver in consumer-to-consumer commerce: building and maintaining localized liquidity, discovery, and trust between strangers, from zero — the reason no durable “better Craigslist” had ever emerged. Marketplace didn’t build cheaper versions of those three things. It repurposed assets that already existed for entirely different reasons — the social graph and identity system (built for social connection), Groups (built for community), Feed and Messenger (built for content and communication) — and that recombination solved liquidity, discovery, and trust simultaneously, at a fraction of the cost any new entrant or incumbent would have to spend building trust infrastructure from scratch.
And while Marketplace didn’t eliminate fraud, it addressed it structurally, through the product itself, not through a trust-and-safety team scaling headcount to compensate. R1 Score = +2, Productized.
Order Food looked like the same playbook — same platform, same reach — but the mechanism wasn’t there. Facebook’s actual contribution was flooding surfaces with Order Food prompts and inserting identity into a checkout flow (autofill, essentially): neither a structural fix. The real cost drivers — restaurant integration, delivery logistics, payment processing — were ones we wouldn’t/couldn’t operationally take on: so we outsourced them to partners. Because Facebook never owned or restructured any part of that operation, we locked ourselves into the industry’s existing architecture, leaving us little room to innovate and create real value. And without that value, Order Food couldn’t deliver the one thing it promised in return (transaction volume), and everything fell apart. R1 Score = 0, Ignored.
This wasn’t a one-off failure, either: the same pattern—not addressing the core cost driver—sank nearly every B2C transact-on-Facebook product that followed.
Circumvent vs. Optimize, Formalized
The Facebook cases show the distinction intuitively. The next pair makes it explicit: two ventures with technically sophisticated-sounding mechanisms, only one of which actually circumvented anything.
Brex launched in 2017 to solve a real problem: young, well-funded startups were routinely rejected for business credit, or forced into personal guarantees, because banks underwrite off credit history — something a two-month-old company doesn’t have. The conventional fix would be optimizing that underwriting cycle: faster reviews, lighter documentation.
Brex didn’t optimize it. It built its own card-issuing and processing stack, wired to open banking APIs, to monitor a company’s cash position daily instead of the industry’s monthly or annual cycle. That monitoring of daily cash positions let them access massive, lower-cost warehouse lines of credit from banks like Barclays and Credit Suisse and later even from public markets.
That’s not a faster version of the old underwriting cycle — it’s a different one, which makes the old cycle irrelevant rather than merely slow. A transformational reduction in underwriting risk/cost, low-cost credit to offer startups roughly 20x higher limits, no personal guarantee required. R1 Score = +2, Productized.
Deepgram, a speech-to-text infrastructure company founded in 2015, makes a good contrast precisely because its 2020 pitch is a good pitch with a real technical story. Its mechanism — an end-to-end deep learning model replacing the traditional multi-stage speech-recognition pipeline — is a genuine upgrade in accuracy.
But the actual cost driver in speech recognition is human-labeled audio data, collected and annotated for every new domain or vocabulary. The end-to-end architecture doesn’t touch that cost — arguably it needs more labeled data than a modular pipeline, which lets you cheaply swap just a vocabulary layer for a new domain.
Even years later, after real improvements (synthetic data generation, self-serve vocabulary tools), the highest-value enterprise customers still require weeks of manual “model improvement partnership” work. R1 Score = 0 at the 2020 pitch stage, evolving since toward a hybrid Product + Operational picture — but never a clean +2.
Note, the lesson isn’t that Deepgram is a weak company. It’s that “technically impressive” and “structurally relevant” are not the same claim, and the difference can be hard to spot.
Solving R1 Without Writing Software
Not every R1 fix is a technical architecture. eShares — later renamed Carta — launched in 2014 to fix how private companies track equity ownership, at a time when this was almost universally managed through spreadsheets and paper certificates, reconciled by hand across the company, its law firm, and its shareholders. The barrier wasn’t ignorance; it persisted partly because the party that provided the existing system any legal weight — law firms — billed hourly for the very reconciliation work a fix would eliminate. (eShares’ original law-firm-first sales strategy failed for exactly this reason.)
The eventual fix wasn’t better software — it was becoming the SEC-registered transfer agent, the actual legal record. One record, one legal source of truth, not several to keep in sync and fight over. This strategy not only circumvented the critical cost driver, it provided the foundation for building differentiated sofware and powered eShares evolution into what Carta is today. R1 Score = +2, Productized. Sometimes the barrier isn’t a technology gap.
Four Questions for Any Pitch
Distilled into something usable on any venture or product pitch, including your own:
- Obsolete, or just cheaper? Does this make the old cost structure irrelevant — or just less painful?
- Architectural, or operational? The cost of product and infrastructure fixes don’t scale with volume. People and service fixes do, often becoming massively more costly.
- Why hasn’t this been fixed already? Ignorance, a misaligned incentive, or a genuine technical or regulatory wall each point toward a different kind of solution, and indicate ventures with vastly different potential.
- Does solving this transform the solution space? Truly solving R1 should put the venture on a different path than the conventional structure it’s replacing — not just a better-run version of the same one.
Testing It Against a Live Pitch
The cases above are all resolved with hindsight — we know how each one turned out. To close the session, we ran the same four questions against a live, unresolved pre-seed pitch that had come through Cornell’s own entrepreneurship listserv: a “Directory for X” style venture. What surfaced quickly was that the critical cost driver wasn’t even addressed in the pitch: a classic and costly cold-start problem for a two-sided marketplace, further compounded by fragmented inventory with continuous decay. Aggregating and maintaining the inventory would be prohibitively expensive.
No hindsight, no answer key — just the diagnostic. It held up well as a way to separate a compelling problem statement (which is often genuinely well-argued) from evidence of an actual structural fix, and it’s a good reminder that this framework isn’t just for grading famous companies after the fact. It’s meant to be run in real time, on a deck or proposal someone just handed you.