43% of failed ventures never found product-market fit. A working test in weeks costs less than being wrong for a year.

What actually kills new ventures

CB Insights published its latest post-mortem analysis in March 2026, covering 431 VC-backed companies that shut down since 2023, 385 of them with identifiable causes. The surface cause is unsurprising: 70 percent ran out of capital. But CB Insights is explicit that running out of money is almost always the final cause of death, not the root problem. The root problems sit underneath: 43 percent never achieved product-market fit, 29 percent were beaten by timing or macro conditions, and 19 percent had unit economics that could never sustain the business.

These are venture-backed companies, staffed and funded specifically to find a market, and still the single most common root cause was building something the market did not want enough to pay for. An SME extending into a new offer, a new segment or a new market faces the same distribution of risk with less cushion. The lesson is not that new ventures are doomed. It is that the fatal error is knowable early, and most companies choose not to know it until the money is gone.

The economics of a cheap real test

Consider the two ways to commit to a new offer. The first: three months of planning, a launch budget for the year, hiring against the forecast. If the market says no, you learn it in month eight or ten, after most of the budget, and the retreat costs almost as much as the advance. The second: a working minimum version of the offer built in three to six weeks for a small fraction of the annual budget, sold at a real price to real prospects. If the market says no, you have spent weeks and a rounding error, and you still own everything you learned.

The asymmetry is the whole argument. The cost of the test is small and fixed. The cost of being wrong at full commitment is large and compounding, because a funded plan generates its own momentum: staff hired, commitments made, sunk costs defending the decision each month. In the CB Insights data this is visible as the gap between the 43 percent who lacked product-market fit and the 70 percent who died of capital exhaustion; companies spend a long time funding a question they could have answered cheaply at the start. A plan can be wrong for a year. A test can only be wrong for six weeks.

A test also produces a different quality of information. Surveys and interviews measure politeness; a price measures demand. The plan's forecast rests on what people say. The test's result rests on what they do.

There is a second return that rarely makes it into the comparison. Even a test that passes changes the launch that follows it. The first ten paying customers tell you which objection actually blocks the sale, which feature carried the decision, what the offer should be called and what the second version must fix. A launch informed by that evidence starts months ahead of one informed by a workshop, and the budget it draws down is allocated against observed behaviour rather than projected behaviour. The test is not a delay before the real work. It is the first, cheapest iteration of the real work.

What a good validation build contains

The real offer, narrowed

Not a brochure or a clickable mock-up: the smallest version of the product or service a customer can actually use and pay for. Narrow the scope, never the reality. One segment, one use case, one path through the product, delivered properly.

A real price

Free pilots test nothing except goodwill. The build must be sold at a price consistent with the eventual offer, because willingness to pay is the variable the whole plan depends on and the only one a plan cannot forecast.

Real distribution

Sell it the way the full offer would be sold: the same channel, the same buyer, the same objections. A test sold through the founder's friends validates the friendship.

One number and a threshold, agreed in advance

Decide before launch what result means proceed: paying customers, conversion from qualified conversations, repeat usage. Writing the threshold down first is what keeps the result honest, because after launch every number can be argued into a success.

A deadline

Six weeks of build and a fixed selling window. The deadline is not a constraint on quality; it is what forces the scope down to the part of the offer that actually carries the risk.

Weeks, not months, is the point

The speed is not about impatience. A test that takes two quarters becomes a project, acquires defenders and a budget line, and ends up as hard to kill as the launch it was meant to de-risk. Kept to weeks, it stays an experiment: cheap enough to run before the annual budget is set, fast enough to run twice if the first result is ambiguous, and small enough that a negative answer is a finding rather than a failure. The 29 percent of shutdowns attributed to timing are a reminder that even the market's answer has a shelf life; a fast test reads the market as it is now, not as it was when the plan was drafted.

Not everything should be tested this way. An offer whose value only appears at scale, or a regulated product with a fixed compliance floor, has a minimum viable size that is genuinely large. But those are the exceptions, and most of the new offers we see inside mid-sized companies are not them. For the rest, the sequence is simple: build the smallest real version, sell it at the real price, count what happens, and let the year's budget follow the evidence instead of preceding it.