Six Claims, Each of Them Falsifiable, and What Happens If Any One of Them Is Wrong
cr4fts.com/n/zdj4y
Most investment theses are a list of preferences dressed as analysis. Sectors the firm likes, stages it writes at, geographies it is comfortable in. They are useful for sorting inbound mail and useless for deciding anything hard, because a list of preferences cannot be wrong. It can only be unfashionable.
Ours is a chain of claims, and each link is falsifiable. If any one of them turns out to be false, the companies we are building are the wrong companies and we would rather find that out from the argument than from the outcome. This is the whole of it, in order, with the case for each laid out in full elsewhere.
Cr4fts co-founds companies out of a workspace in Nairobi. We put in the cheque, the room and the people, take a minority stake at incorporation, and build alongside a founder who owns the company outright at the end of it. Intake is rolling, and we take on as many as we can do properly rather than as many as we could raise for.
That describes the mechanics and explains nothing. Any number of firms could run the same structure and be making entirely different bets. The structure follows from the six claims below, and the claims are the part worth arguing with.
The single figure underneath all of them is this one.
The cost of a general economic input has fallen by more than two orders of magnitude in under three years. Stanford's AI Index puts the cost of querying a model at GPT-3.5 capability at twenty dollars per million tokens in November 2022 and seven cents by October 2024. Inputs do not do that in a product cycle. They do it perhaps half a dozen times a century, and each time the arrangement of an entire economy gets renegotiated afterwards.
The historical case is electrification. Edison's lamp was patented in 1880, electric motors were under five percent of factory drive by 1900, and the productivity statistics did not move until the 1920s. The delay was not the technology. It was the forty years required to stop bolting motors onto steam-era line shafts and start designing plants around power being available anywhere.
Almost every large organisation is currently at the retrofit stage and reading its disappointing results as a verdict on the technology. It is not. It is a verdict on retrofitting.
The full argument is in A Shift, Not a Cycle, including why the loud period and the productive period are different periods, and why the firms waiting for proof are the ones with the shortest runway.
If the first claim holds, the second follows mechanically. Incumbency is not a property of a company. It is a relationship between a company and a set of conditions, and most of what gets called a moat is a coordination cost somebody paid once and nobody has had to pay since.
Average tenure in the S&P 500 has fallen from around thirty-three years in 1965 to a forecast fourteen by 2026. The interesting part is not the churn but the mechanism. A firm that has led an industry for twenty years has spent twenty years removing everything that did not serve its position, which produces extraordinary fit to one environment and none of the slack to fit another.
They do not fail by doing something stupid. They fail by doing, competently and at scale, the thing that used to work.
The audit that separates a real moat from an expired one is set out in Every Industry Is Back in Play, along with the industries where this argument does not apply and the test that tells them apart.
The number under every business plan is what it costs to get from nothing to a working version, and it has fallen by roughly two orders of magnitude in fifteen years. Almost nobody has updated the conclusions derived from it.
This matters most at the threshold. A workflow tool for forty thousand customers paying thirty dollars a month is uninvestable against a two-million-dollar build cost and straightforwardly good against fifty thousand. Nothing about the market changed. The number underneath it did, and the conclusion inverted.
The corollary is the part people skip. Distribution did not get cheaper. Trust did not get cheaper. The last mile from eighty percent to ninety-nine costs what it always did. A founder who takes the first half of this argument without the second builds something adequate that nobody buys.
The arithmetic, and the three beliefs it invalidates, are in Building Keeps Getting Cheaper.
Output grows at best linearly with headcount. Coordination grows as n(n-1)/2. Three people maintain three channels, thirty maintain four hundred and thirty-five, and the cost appears nowhere on a balance sheet.
None of that is new. Brooks published it in 1975 and Hackman spent forty years confirming the shape. What is new is that the minimum team required to build something serious has fallen below the point where the arithmetic starts punishing you. For most of the last fifty years the coordination tax was unavoidable, because shipping anything credible needed more people than could fit around a table. It no longer does.
The decisive advantage is not speed. It is that a small team holds the whole problem in a shared head, while a large one works from somebody else's compression of it, and the compression drops exactly the odd specific detail that usually contains the answer.
Where this fails, and the two weaknesses that a studio has to earn its place by addressing, are in Small Teams Win This Round.
The first four claims are global. This one is why we are in Nairobi rather than anywhere else.
In a mature market almost every process has a default, and displacing one means paying the switching cost for everything arranged around it. That bill is what protects most incumbents in developed economies. Where a process has never been formalised, the bill does not exist, because there is nothing to migrate off.
Mobile money is the proof and it is almost always cited wrongly. It was not a cheaper bank account. Value held against a phone number, cash moving through shopkeepers, economics built for tiny transfers: every primitive was designed against an actual constraint rather than degraded from a template. Sub-Saharan Africa now carries roughly two thirds of global mobile money transaction value.
The argument, including the three objections that are correct and the questions worth answering before writing any code, is in The Opening Is Widest Here.
This is the one we are most often asked to soften, and we will not.
Africa is behind at the technological frontier and the distance is growing. Under two percent of global data centre capacity. Around three percent of global research output. Most sub-Saharan countries spending between 0.1 and 0.4 percent of GDP on research against an African Union target of one percent, while frontier economies spend three to five percent of far larger economies. Those gaps compound, and nothing in the next decade closes them.
The mistake is treating that as one gap. Producing frontier capability requires exactly what is missing here. Applying it requires an interface, which costs the same in Nairobi as in California, and knowledge of a specific problem, which is cheaper here than anywhere. Three years ago that distinction made no practical difference because applying the technology still meant assembling it. It is now the only distinction that matters.
The measurements, and the failure mode where all of this is right and simply ten years early, are in Behind, and Closing the Distance.
Taken individually each claim is arguable and several are unremarkable. The reason we organise a firm around them is that they compose, and the composition is narrow.
A shift without cheap building produces a market only incumbents can afford to enter. Cheap building without unset defaults produces a hard fight over switching costs, which is the situation across every developed economy and why so much software there competes on marginal improvement. Unset defaults without cheap building produces a market nobody can afford to serve, which is what persisted here for twenty years. Small teams without a shift produces a lifestyle business, which is a perfectly good thing to want and not what we do.
All six hold at once, in one place, for a period measured in years rather than decades. That is the entire proposition. It is not that Africa is rising, or that artificial intelligence changes everything, or that small teams are virtuous. It is that a specific set of conditions overlaps right now and will not overlap indefinitely.
Each claim fails in a specific way, and naming the failure is the only thing that makes a thesis worth holding rather than repeating.
The shift fails if the cost curve reverses, which is possible: some of the decline is providers buying market share rather than underlying economics, and that portion can be withdrawn. It also fails, more subtly, if the redesign phase never arrives because the technology turns out to be genuinely useful only as assistance, in which case the incumbents currently retrofitting are right and we are the ones misreading the evidence.
The incumbency claim fails wherever the advantage was physical or regulatory rather than a coordination cost, and we get this wrong in both directions. We have passed on things where the moat was thinner than it looked and looked seriously at things where it was thicker.
The small teams claim fails in any domain where the minimum viable team is still large, and there are more of those than the current enthusiasm admits.
The emerging markets claim fails on purchasing power. It is the objection we take most seriously, and it is why almost everything we back sells to small businesses rather than to individuals: a business using your product to make or save money can justify a price, and a person usually cannot.
And the last claim fails if access to frontier capability becomes gated by geography, capital or regulation. Today it is available to anyone with a payment card. That is a commercial decision by a handful of firms, not a law of nature.
A thesis that does not constrain behaviour is decoration, so here is what these six force on us.
We back people who already carry the problem, because the only input in this transition whose price has not fallen is knowing exactly how a process fails. Enthusiasm about a market is not the same thing and we can usually tell within an hour.
We insist on a wedge narrow enough to build inside six months and still be useful, because six months is roughly how long the cost of building stays stable enough to plan against. Anything longer is a bet on where the curve goes, and we would rather bet on a customer.
We look for revenue in month four rather than year four. In a market where the default is a notebook rather than a competitor, a first payment is the only reliable evidence that the absence of a default is an opportunity rather than an absence.
And we track renewal harder than growth, because the failure mode we most fear is not being wrong. It is being right ten years early, and renewal is the only signal that separates those two while there is still time to act on it.
Six claims, each falsifiable, each argued in full. If we are wrong about any of them we would rather hear it from someone who has read the argument than watch it arrive in the numbers three years from now.