Mailchimp sold for twelve billion dollars having raised nothing at all. Ninety-five per cent of venture funds fail to return adequate capital. The pattern underneath both is forty years old and has a name.
Eliyahu Goldratt's claim was that every system has exactly one binding constraint, and that everything you do which is not aimed at that constraint is theatre. Exploit before you elevate: wring the existing bottleneck dry before you spend money enlarging it.
Capital breaks this. Money does not remove a constraint — it buries it under payroll, tooling and narrative, and by the time it surfaces again there is nothing left to exploit. The evidence for that is on the balance sheets, in both directions.
Companies that reached serious scale while the constraint was still visible to them:
| Company | Outcome | Raised |
|---|---|---|
| Mailchimp — email marketing, Atlanta, founded 2001 | $12B acquisition by Intuit, 2021, on ~$800M revenue | $0 |
| Basecamp — project software | 20+ years profitable, still independent | $0 |
| GitHub | $7.5B acquisition by Microsoft | late, small |
| Atlassian | $50B at IPO | none for 8 years |
Mailchimp took twenty years to reach a twelve-billion-dollar exit and raised nothing on the way. That is the number worth sitting with. It is not that the money was unavailable. It is that the discipline of not having it kept the bottleneck in plain sight for two decades.
Premature scaling is the precise failure Goldratt predicts: elevating before exploiting. You hire ahead of the bottleneck, build ahead of demand, and spend the runway solving problems that were not the binding one. WeWork reached a $47B peak valuation before its constraint became visible. Theranos reached $9B with a constraint that had never been solved at all.
The point is not that venture capital is bad. It is that capital is an accelerant applied to whatever is actually there — and when nobody has identified the constraint, what it accelerates is the concealment of it.
Build-Measure-Learn is constraint exploitation with a feedback loop attached. Build the minimum experiment that tests the riskiest assumption — the minimum learning, not the minimum product, which is where most MVPs go wrong: founders optimise for features when the job is to answer a question. Measure with innovation accounting rather than vanity metrics. Learn, then decide whether the constraint moved.
Read that way, the Lean Startup is not a methodology that competes with Theory of Constraints. It is the same idea with a shorter cycle time, and it inherits the same failure mode: run the loop against a constraint you have not correctly identified and you will iterate rapidly in the wrong direction.
Most missions organisations ask how do we get more missionaries and more funding? before asking whether they are exploiting the capacity they already hold. Goldratt would call that elevating before exploiting, and it is the most expensive order to do it in.
This connects directly to the cost question in the missions review. A sector that cannot state what mobilising one person costs is a sector that cannot tell whether its constraint is money, people, or something else entirely — and so it defaults to raising more of both. Ramen-profitable, in that framing, is not a financial condition. It is a negotiating position: the ability to say no long enough to find out what is actually binding.
The Kauffman 95% figure and the Startup Genome 74% are both widely cited and both are older than they look; each rests on a specific sample and definition that does not travel perfectly to other contexts. Quote them with the source attached.
Survivorship bias is the live risk in section 1. Mailchimp and Basecamp are the bootstrapped companies you have heard of. The bootstrapped companies that stayed small or died quietly do not have case studies, and this review did not attempt to count them. The four cases prove that the path exists at scale. They do not establish a base rate, and nothing here should be read as one.
Theory of Constraints is a framework, not a finding. Its claim that every system has exactly one binding constraint is useful as a discipline and unfalsifiable as a proposition.