Discovery is cheaper than delivery so do more of it
💎CORE ARGUMENT
Finding out that an idea is wrong costs an afternoon in discovery and a quarter in delivery. The saving is so large that I should be doing far more discovery than feels comfortable.
The arithmetic is not subtle. Eight customer interviews are about eight hours of talking, the same again of preparation and notes, and one train ticket. A sprint is six people for two weeks. If the interviews kill one bad idea in five, they have paid for themselves several times over. They rarely kill an idea outright, though. More often they bend it, and that is where most of the value is.
Still I under-invest, every time. Discovery produces nothing I can demo. Delivery produces tickets that close. A week of interviews looks from outside like a week of chatting, and a week of building the wrong thing looks like progress until the day it ships.
I have paid full price once. In 2021 I spent nine months on a carrier onboarding wizard that eleven people used in its first quarter. Nobody had watched a carrier onboard. Against that, one morning on a dispatch floor in Rotterdam this February showed that the pilot's list was too long to use. I have never bought a cheaper finding.
So the rule is to buy information where it is cheapest, and it is cheapest before the code exists.
It breaks down at the edges. Discovery cannot find what only real load and real data reveal: the launch taught me that in one night, and 💎 Reality has more detail than any plan can hold is the note that contradicts this one. It can also turn into a refuge. A ninth interview is easier than a decision. 💎 A finished small thing teaches more than an unfinished large one is the corrective, since at some point the cheapest experiment left is a small thing actually shipped.
🧱BUILT FROM
🌍WHY IT MATTERS
- No idea of mine enters a sprint until someone has watched a customer do the job it is meant to help.
- I put a number on discovery in planning: hours and train fares, set against a sprint.
- I stop when the answers repeat. Feeling ready is not the signal.
Still open.
- How many interviews before the answers repeat? Eight is my working number and I cannot defend it.
- What is the discovery equivalent of a load test: something that can fail on a date?
- Who pays for discovery when the roadmap is already full?
☑️NEXT STEPS
- Sit on a dispatch floor before the next planning round
- Book the two design-partner sessions for scorecards
- Write down the kill criteria before the November prototype sessions
📖 Cagan on product discovery: the risks of value, usability, feasibility and viability should be met before the build. 📖 Torres on opportunity trees: discovery as a weekly habit, where I had treated it as a phase.
💎 A finished small thing teaches more than an unfinished large one: a shipped small thing is discovery by other means. 💎 An opportunity is a customer problem and not a feature in disguise: discovery is only cheap when it is aimed at a problem. Aimed at my own solution it is expensive.
The cost of being wrong rises with every stage an idea passes through. An interview tests an assumption for hours, a prototype for days, a release for months and for a slice of customer trust. Moving the test earlier buys the same knowledge for less.
Customers describe the past well and the future badly. A product that changes how the work is done cannot be found by asking, and some of the best decisions at Halden were made by people who stopped interviewing and built.
Being tried properly, for the first time, in 📂 Carrier Scorecards discovery. It is the main thing I practise under ♦️ Product discovery. The expensive lesson is 🕰️ The launch nobody used.