Minimum-Cost Hypothesis Test
Buy the smallest reliable answer before funding a full-scale bet
- Difficulty
- Moderate
- Time to result
- ~weeks to results
- Steps
- 6
- Confidence
- 99%
Jenkins argues for keeping a business simple until evidence requires more complexity. Before spending heavily on a marketing channel or rollout, identify the central hypothesis and design the cheapest test capable of producing a statistically useful answer. His accounting separates value from learning: if a trial's acquisition cost is twice the target, part of the spend still bought customers at the acceptable rate, while the remainder bought evidence not to repeat the channel. The output is a decision backed by measured unit economics, not a vague impression. The same evidence also improves fundraising because a founder can explain what was tested, what it cost, and what scaling is reasonably expected to do rather than asking for speculative marketing money.
Origin
Jenkins says Moonpig's limited funding forced small experiments. He returned to an MBA statistics textbook to calculate the least he could spend for a statistically significant answer about customer-acquisition cost.
Core principles
- 01Test the central hypothesis before scaling its implementation
- 02Spend only enough to produce a statistically useful answer
- 03Separate the cost of acquired customers from the cost of learning
- 04Use measured evidence to make the next funding request credible
How to run it
- 1
Name the decisive hypothesis
Express the assumption as a testable statement and specify what decision depends on the result.
Pro tip Test the assumption that can invalidate the plan before polishing secondary details.
Watch out Do not disguise a broad rollout as an experiment.
- 2
Define the target economics
Choose the success measure, such as willingness to pay or customer-acquisition cost, and set the acceptable threshold in advance.
Pro tip Use a metric connected to the business model rather than attention alone.
- 3
Design the smallest realistic test
Remove unnecessary scale and features while preserving the behaviour the hypothesis is meant to measure.
Pro tip A temporary setup can test demand before permanent infrastructure is built.
Watch out An unrealistic test can produce a cheap but misleading answer.
- 4
Set an evidence floor
Estimate the sample and spend needed for a useful answer rather than choosing an arbitrary budget.
Pro tip Use appropriate statistical help when the decision is material.
Watch out Statistical significance does not repair biased sampling or a poor measure.
- 5
Price the learning
Separate the portion of spend that created acceptable customer value from the excess spent discovering that the channel or offer missed the target.
Pro tip Record the failed route so the organization does not buy the same lesson twice.
- 6
Scale or stop
Increase investment only when the measured result supports it; otherwise revise the hypothesis or end that route.
Pro tip Carry the evidence and assumptions into any funding plan.
Watch out Do not extrapolate beyond what the test actually measured.
In the wild
Bartlett describes advising a gym owner not to fund ten locations before testing whether long-haul drivers would exercise at service stations. A temporary shipping-container gym could first measure usage, then introduce a price to test willingness to pay.
→ The founder could seek rollout capital only after observing both use and payment on a small scale.
Jenkins contrasts spending £2,500 to estimate a channel's customer-acquisition cost with spending £25,000 to learn the same negative lesson. If the observed cost was £20 against a £10 target, he would treat part of the spend as customer acquisition and part as the price of evidence.
→ The channel could be rejected before a much larger loss while preserving a measured basis for the decision.
Common mistakes
Scaling before testing
Building the full rollout makes the cost of discovering a false assumption unnecessarily large.
Testing attention instead of payment
Interest can support an early signal, but it does not answer a willingness-to-pay hypothesis until price is introduced.
Calling all failed spend waste
A bounded negative test can still acquire some customers and prevent a much larger repeat of the mistake.
Is it for you?
Best for
It is best for reversible product, pricing, demand, and acquisition questions that can be tested with a bounded sample.
Not ideal for
It is not ideal when a small test cannot represent the real conditions or when safety, legal, or ethical requirements demand a complete solution from the start.
From the transcript
“find the simplest way to test the hypothesis as cheaply as you can”
“why waste it when you can get the answer for two and a half grand”
“prove what you can prove on the least amount of money first”
From the episode
Moonpig Founder: How I Built A $150 Million Business WITHOUT Sacrifice: Nick Jenkins