The short answer
Write the assumption down with a number and a range before looking at any data. Test it against twelve to twenty-four months of records the business already keeps. Size any gap in dollars and name the reason it survived. Confirm it in a period you did not search. Then test it live against a control, with a stop condition agreed before you start.
Every business runs on assumptions: that a customer segment is profitable, that a route is efficient, that a discount wins the order. Most have never been tested, not because nobody cared, but because testing one used to cost a study, a budget, and a year. The records needed to test them have been accumulating the whole time.
1 · Name it
Write the assumption down as a number with a range, and note how confident you are. Do it before anyone opens the data. Once people know an outcome, they overestimate how predictable it was and rarely notice that their view shifted9. Science handles this by registering the prediction first10. A business can do the same with a shared document and a timestamp.
2 · Find it
Let the records answer. Quotes, invoices, orders, routing, and contracts are years of the business recording its own behavior. Test each written assumption, and also look for gaps nobody raised. Those are often the largest, because nobody was looking. Label them as exploratory and keep count of how many things you tested, since searching many ways can find patterns by chance.
3 · Size it
Put a dollar range on the gap, from your own data, and name the reason it survived: an incentive that kept everyone playing the old game, missing information, or a habit that made sense for each person and costs the business. If you cannot name a reason, the gap is probably noise, or someone has already closed it.
4 · Prove it
Hold back a stretch of history before you start searching. When you think you have found something, check that it also appears there. Even then, treat it as a candidate. History makes a finding credible. Only a live test against a control makes it real.
5 · Build it, with a stop condition
Before building anything, write down the evidence that would end the work: for example, less than 1.5 points of margin against control after six weeks. The decision to stop is then made against a number agreed in advance, not an argument after the fact.
Why this order works
Each step answers a question a skeptic would ask. Did you just find what you were looking for? Did you miss what nobody asked? Why has nobody taken it? Is it a coincidence? What if it doesn’t work? A result that survives all five is worth building.
Sources
- 9Fischhoff, B. (1975). Hindsight ≠ foresight. Journal of Experimental Psychology: Human Perception and Performance, 1(3), 288–299.
- 10Nosek, B. et al. (2018). The preregistration revolution. PNAS, 115(11), 2600–2606.
Market figures describe the landscape and are not Modven results.