Watch
Listen
“No one tests the depth of a river with both feet.”
Ashanti Proverb
I wrote code for 3 hours before work each day. It was 2012 and I was building Shommi, a tool that tracked Twitter to find potential customers for businesses. I told myself a working solution would sell itself.
After 3 months I ran out of things to improve. On a Saturday I drove to a strip of car dealerships on the north side of Atlanta and walked into 3 of them. I told them people in their area were tweeting about buying cars.
They were polite and said “interesting.”
The most interested person at those 3 dealerships wanted to hire me for Ruby on Rails contract work.
3 months of mornings bought 1 Saturday’s worth of information.
Long feedback loops make finding out expensive. Short loops make failure cheap.
Unilever’s mathematicians tried to design a detergent nozzle first. Fluid dynamics and phase transition were their specialty.
The nozzle clogged anyway.
Unilever’s biologists came at it without the theory. They made 10 variants, kept whichever sprayed 1 or 2 percent better, made 10 variants of that one and ran the loop again.
45 generations later, they had a nozzle shaped unlike anything the mathematicians designed.
449 rejects bought a nozzle that sprayed many times better.
13 years after Shommi, I changed the scale of the test. I paid hours instead of months to test the idea that became HeadsUp. I handcrafted Notion documents laying out what each founder’s competitors were doing and sent them to 8 founders.
One offered $300 a quarter before a line of code existed.
Google runs 12,000 split tests a year because each test is cheap to run. In 1 test on 41 shades of blue, a single color shift added $200 million in annual revenue.
Saras Sarasvathy sat 27 expert entrepreneurs down with the same startup problem and had them think out loud. They set what they can afford to lose before they start, then skip forecasting what they might gain. She named it affordable loss.
Size each test so a miss can’t finish you.
📄
What test could you shrink until failure is cheap?
