A strategic assumption is a claim a strategy depends upon that nobody has checked. A product market fit score reports on the users a product has already reached, so even a good score is evidence about the present. The choices a strategy makes about the next two years rest on claims of a different kind. Rita Gunther McGrath and Ian MacMillan set out the method for handling those claims in Harvard Business Review in July 1995, and they called it discovery driven planning. Their first instrument was a list of every assumption a new venture depended upon, with a date against each line for proving it.
The sections below cover what a strategy assumes and how to write it down. They then separate the load bearing assumptions from the incidental ones and rank the list by consequence and by cost. The page ends with the form an assumption takes, and then with the assumptions no test will settle.
Writing down what a strategy assumes
McGrath and MacMillan built discovery driven planning for ventures whose numbers no forecast could reach. What makes it work is putting the assumptions on paper, where somebody can argue with them. Their method starts from the result the investment has to produce and then works backwards. It asks what would have to be true about customers, price, volume, cost, channel and timing for that result to arrive. The output is a list with a date against each line, and funding arrives against the dates as each assumption survives its test.
An assumption nobody has written down still governs the plan. It reaches the plan as a number in a spreadsheet, a date on a roadmap or a sentence in a board deck, where it looks like a fact and carries no author. Writing it out as a sentence with a subject and a verb is what makes it arguable, and the sentence has to be exact enough that somebody could show it to be false.
Load bearing assumptions and incidental ones
A list of assumptions written that way runs long, and a few lines on it carry the whole strategy. A load bearing assumption is one the strategy cannot work without. A product led plan might claim that operations managers will adopt a tool with no sales conversation. If that claim turns out to be false, the distribution strategy has nothing left to stand on, and the pricing, the packaging and the onboarding all belong to a different plan. An incidental assumption is one the strategy can survive. A guess that the average account will invite four colleagues in the first month can be wrong by half and move only the growth rate.
The test that separates the two asks what stops working when the assumption turns out to be false. A load bearing assumption takes a strategic choice with it, and an incidental one moves a number.
Ranking assumptions by consequence and by the cost of a test
Consequence alone does not order the list. Two assumptions can both sink the strategy and still differ by a factor of a hundred in what it costs to settle them. The cost of finding out is the second axis, counted in time, in money and in the work given up while the team runs the test. Plotting the list against both axes produces four groups. Each group carries a different instruction.
The top left group is where testing starts, because an assumption that would sink the strategy and costs a fortnight to settle has no argument for delay. The top right group holds the assumptions that carry the same weight and admit no test at any price.
If a team has left the top left group untested, it is carrying a risk it chose to carry without saying so. The bottom half of the grid absorbs most of the enthusiasm in a discovery workshop, because the cheap incidental questions are the pleasant ones to answer. The answers move a forecast by a few points and change no decision at all.
Turning an assumption into a statement a test can settle
An assumption plotted on the grid still has to say something a test could contradict. The form that works names three things, which are the group, the behaviour and the threshold that would count as confirmation. Everybody then reads the result the same way, before anybody has run the test. A claim that customers want faster reporting cannot be tested, because no result could contradict it. A fortnight of trials settles a claim that at least a third of operations managers on a trial will build a report of their own in the first week with no help.
David Bland and Alexander Osterwalder made the same point in Testing Business Ideas in 2019. Every assumption there goes into one of desirability, viability and feasibility, and it is then plotted on a grid of importance against the evidence behind it. The quadrant they send teams to first holds the assumptions that matter most and have the least evidence. That is the instruction the consequence grid gives under different names.
Assumptions that no test will settle
Some assumptions survive the grid and defeat it. A claim about whether a regulator will license a category at all carries enough consequence to sit in the top row, and it has no test at any price, because the answer does not exist yet. Whether a technology will be cheap enough to matter in four years is the same kind of claim. A research project aimed at either one spends money and returns the same uncertainty it started with.
Those assumptions stay on the list and move to a different treatment. A claim that carries the whole bet and admits no test has a literature of its own, and that literature has its own vocabulary for how much remains unknown and its own set of postures a team can take.