You didn't write down the important part
Cazoo had a good vision and a clear strategy. The teams executed well. The company went bankrupt anyway.
Not because of the execution. Because of one sentence nobody ever wrote down: capital will be cheap for as long as we need it.
I was interim CPO there for the first six months, helping set up the product org. I wasn't there for the SPAC listing and I wasn't there for the collapse. But I was there for the stretch when that sentence was holding everything up, and I was one of the people standing on it. I didn't write it down either. Nobody argued about it, because nobody ever said it out loud. In the three years after I left, the business scaled into an $8 billion listing without the unit economics to stand on its own. Then the era of near-zero interest rates ended, the assumption collapsed almost overnight, and two years after listing it was bankrupt.
I put a line about this in The Decision Stack: no amount of good execution at the bottom of the stack can compensate for an unexamined assumption at the top.
But Cazoo wasn't unusual. Every organisation I've worked with has a sentence like that holding it up. Almost none of them can tell you what it is.
Could you?
Every answer in your stack is a bet
The Decision Stack asks five questions. Where are we going? How will we get there? What's important now, and how do we measure progress? What actions are we taking to move forward? And how do we make the right choices?
Every layer of your Decision Stack is a choice. But every choice was made on top of things you believed and hadn't proven. So the stack was never just five answers. It's five answers and the assumptions underneath them, and the assumptions are the half nobody writes down.
Your vision assumes the world you're describing will still be worth building when you get there. Your strategy assumes the mechanism you picked still works, and that the conditions it depends on still hold. Your objectives assume the number you're chasing still stands for the thing you actually care about. Your opportunities assume customers want what you think they want. Your principles assume the tradeoff you codified two years ago is still the one you'd make today.
And here's the part that should bother you. Most product people meet assumptions exactly once, right at the bottom, where an opportunity meets a solution. A good exec team might run the exercise at the strategy layer, usually at an offsite, usually once, usually never again. Almost nobody does it for their vision. Almost nobody does it for their objectives. And I have never seen a team do it for their principles.
A coherent stack tells you the choices fit together. It can't tell you the assumptions were right.
The assumptions, layer by layer
The assumptions are different at every level, which is why the tools you already have don't reach all the way up.
Vision assumptions are about the world. That the future you're describing is possible, still wanted, and still yours to build.
At Mind the Product, our vision was about connecting and supporting product people everywhere. Underneath it sat a belief nobody had ever stated: that those people would gather in a room. In-person events were 95% of revenue. When COVID hit, most conference businesses in our position closed.
We didn't. I'd love to tell you we saw it coming, but we got lucky about which layer broke. Product people still wanted to connect, they just couldn't do it in a room. So the assumption that failed was a strategy assumption, not a vision one.
But we'd made some of that luck. We'd spent real time as a team getting the vision right, and we'd written it about what we wanted for product people rather than the format we happened to deliver it in. If we'd put "the best product conferences in the world" on the wall, the vision would have gone down with the events. Because we'd written it about connection, there was somewhere to stand while we rebuilt everything underneath it. The vision held. The strategy moved into subscriptions and online events and new ways to bring people together.
So here's the useful bit: the way you keep fragile assumptions out of your vision is to write it about the outcome and not the format. And knowing which layer an assumption lives at tells you how much you have to rebuild. That's the difference between a pivot and a funeral.
Strategy assumptions are about advantage. The mechanism by which you win, and the external conditions that mechanism depends on. Cazoo's was cheap capital. When growth at all costs overextended the business just as that assumption fell apart, the business fell apart with it.
Objective assumptions are about cause. That moving this number moves the thing you actually care about.
Wells Fargo believed products per household measured relationship depth. In the 1990s it did. Then they made it the target, and the layers below optimised for it exactly as you'd expect. The SEC's investigation says the metric ended up "inflated by accounts and services that were unused, unneeded, or unauthorized." That's Goodhart's Law, written into a legal document. Nothing external had changed. Chasing the number destroyed the correlation it was built on, and the number kept going up the entire time.
Opportunity and solution assumptions are about customers. Will they want it, can we build it, does it work for the business?
We’re very good at this layer of assumptions. Assumption mapping, opportunity solution trees, the four product risks: if your teams run experiments before they commit, they're using tools that work. It's also the only layer that's solved, which I’ll get into in part two of this series.
Principle assumptions are about tradeoffs. That the choice you codified is still the choice you'd make.
Intel had a decision rule for chip deals: we only take the business if the price clears our forecast cost at the forecast volume. When Apple came asking about the original iPhone, the rule said no. Paul Otellini, years later: "the forecasted cost was wrong and the volume was 100x what anyone thought."
The rule was applied correctly. The rule survived. The inputs had gone stale, and nobody re-tested them, because the rule felt objective. When did you last re-test the inputs to a rule you trust?
None of this is a new observation, by the way. Richard Rumelt put it plainly: "A new strategy is, in the language of science, a hypothesis, and its implementation is an experiment." Kaplan and Norton got there in 1996, when they wrote that "a strategy is a set of hypotheses about cause and effect" and built the balanced scorecard around exactly that. The arrows between the boxes were the whole point. Then it got industrialised. Companies kept the boxes and the measures but dropped the arrows, and an assumption tracker turned into a dashboard.
A good stack makes this worse
Here's the uncomfortable part. The whole promise of a well-built Decision Stack is that decisions lower down stop being guesses. Your vision rules out directions, your strategy narrows the field, your objectives focus the effort, and by the time a team is choosing what to build the choice is closer to a conclusion than a guess.
That's true. It's also why a bad assumption at the top of a good stack is more dangerous than a bad assumption at the top of a broken stack.
In a disconnected organisation a rotten belief stays where it is. Nobody's listening anyway. In a coherent one it propagates. Every layer beneath it inherits it, acts on it, and executes against it with conviction and speed.
Cazoo executed brilliantly all the way to bankruptcy.
And if you're now handing parts of the stack to agents, they'll run a rotten assumption faster and more consistently than any team you've ever led.
So are you sure you'd spot it? Karl Weick has bad news. "People make sense of things by seeing a world on which they already imposed what they believe. In other words, people discover their own inventions." An unexamined assumption decides what counts as evidence in the first place. You wouldn't notice you were wrong, because the assumption is the thing doing the noticing.
Three moves
None of these need a workshop, a consultant, or a new tool. And you run them at whichever layer you're standing on, not just the one at the bottom.
Name them. Take one layer and ask Roger Martin's question: what would have to be true for this to be right? Write the answers down as sentences. Not analysis, not a deck. Sentences.
This is nearly free and it does most of the work, because the expensive failures aren't assumptions that failed a test. They're assumptions nobody ever wrote down. WeWork's S-1 contained $4 billion of committed revenue and $47 billion of lease obligations, both numbers calculated by WeWork, both in the same document. Nobody needed to run an experiment. Somebody needed to put the two numbers next to each other.
Sort them. RAND has been doing this since the nineties and their vocabulary is better than ours (a certain popular LLM has done its bit to bring the phrase back into fashion, but they got there first). An assumption is load-bearing if its failure would force real changes to the plan. It's vulnerable if it could plausibly fail within the lifetime of the plan.
Notice how much better "vulnerable" is than "how confident are we". Given infinite time every assumption fails. The useful question is whether yours outlives this plan. Only the ones that are both load-bearing and vulnerable earn any further work. (And a test isn't the only kind of further work. At the top of the stack it usually isn't even the likely one. That's part four too.)
Set a signpost. For each assumption that survives the sort, write down the specific thing you'd see if it were failing, then put a name and a date on it.
This is the part everybody skips and it's where the value is. Annie Duke calls it states and dates: “if I am, or am not, in a particular state by a particular date, then I act”. A signpost with nobody's name on it is just an interesting chart.
And for an objective, the signpost is almost never the metric itself. Wells Fargo fired 5,300 people over five years for gaming the number while reporting that same number to investors as evidence of success. A number that's still going up is not evidence that the belief underneath it still holds.
Or borrow Andy Grove's question
If three moves is two too many, there's a shortcut, and my favourite version of it is from Intel in 1985.
Andy Grove asked Gordon Moore what a new CEO would do if the board fired them both. Moore answered immediately: he'd get us out of memory chips. Grove's response was "Why shouldn't you and I walk out the door, come back in, and do it ourselves?"
The detail I love is what they found when they looked. Intel's DRAM share had fallen from 82% to 1%, and only one of their eight fabs was still making memory chips. The organisation had already re-based itself on reality. Only the top of the stack was still running the old assumption, which Grove called “the self-evident truth that Intel was a memory company."
So ask it. If the board brought in a new CEO tomorrow, what would they change? One sentence, works at any layer, and it strips out sunk cost, identity, and politics in a single move. It's also easier to hear than "your strategy is wrong", because everyone in the room gets to answer as somebody else.
So what is your stack standing on?
Cazoo's assumption was eleven words long and nobody ever wrote it down.
It wasn't hidden, and nobody was hiding it. It wasn't complicated, it wasn't contested, and it wasn't even wrong when it was made. It was simply never written down, so it was never owned, so nobody was watching it.
Your stack is five choices. It's also every assumption underneath them. You only wrote down the choices.
So pick a layer. Any layer, not just the one at the bottom. Write down the one sentence that has to be true for that choice to be right. Then write down what you'd see if it stopped being true.
This is part one of four on the assumptions under your Decision Stack - I'll update these links as I publish. Part two is We got rigorous about the cheap decisions, on why our tools only test the bottom layer. Part three is Somebody already knows you're wrong, on why the evidence that an assumption is failing never climbs. Part four is The board doesn't want your roadmap, on taking the assumptions to the room that owns them.
