Guess First, Then Try to Prove Yourself Wrong

Part 5 of a series on structured problem-solving and senior-level communication.

Reading time: about 9 minutes.

---

Three weeks into looking at why a client's costs are running over, the folder has forty-one files in it. Extracts from two systems, a survey, a competitor benchmark someone found, notes from eleven interviews. The analysis is genuinely thorough and nobody can say what it means, because none of it was gathered to settle anything in particular. It was gathered because it was available.

The meeting where this becomes visible always goes the same way. Someone asks what the answer looks like so far. The response is that it is still early and a few more things need pulling. It is not early. Three weeks is most of the time there was. What is missing is not data, it is a claim that any of the data could have contradicted.

What This Article Answers

• What does hypothesis-driven actually mean in practice? (Committing to a probable answer early, then spending your time trying to break it.)
• Isn't guessing before you have the data just bias with extra steps?
• How do I form a hypothesis when I genuinely know nothing about the domain?
• What do I do when my hypothesis turns out to be wrong in week two?
• How do I avoid quietly collecting only the evidence that agrees with me?
• Does this work when the stakes are high and being wrong is expensive?
• How is this different from just having an opinion early?

TL;DR

Analysis without a hypothesis expands to fill the time available and produces a folder instead of an answer. The fix is to write down your best guess before you feel qualified to have one, then organise every piece of work around trying to disprove it. The hypothesis is not a prediction you get scored on. It is a tool for deciding what not to look at.

• A hypothesis has to be plausible, not correct. Its job is to make the next piece of work obvious.
• Being wrong early is the good outcome. A hypothesis killed in week one cost you a week and bought you a much better second guess.
• The real risk is not guessing wrong, it's falling in love. Premature closure and cherry-picking are documented failure modes, and they get worse when you have committed publicly.
• Write the disproof condition when you write the hypothesis. "I would abandon this if X" is what separates a hypothesis from an opinion.
• Data without a hypothesis is not neutral, it is just unfinished. Someone still has to decide what it means, usually under more time pressure than you had.

Two three-week analyses. One gathers broadly and narrows at the end, under deadline. The other commits to a guess on day one and spends three weeks narrowing. Same time, different shape, and only one of them has an answer on the last day.

The Day-One Answer

The last part was about getting the question right before you start answering it. This one is about what happens immediately after: you have a well-framed question, no answer, and a deadline. The instinct is to go and find out. That instinct is what produces the forty-one-file folder.

The alternative is to write down what you think the answer probably is, before you are entitled to think so. Not a hunch you keep to yourself, a sentence you commit to on paper. The day-one answer. Costs are running over because a change in the delivery model added work nobody re-priced. Renewals are slipping because accounts are not being contacted before their term ends. Something specific enough that a reasonable person could look at it and say, no, that's not it.

The point of writing it down is not that it is right. It usually is not. The point is that a specific claim tells you what to do next, and a general question does not. If the hypothesis is that the delivery model change added unpriced work, then the next task is obvious: compare scope on contracts signed before and after the change. That is one question, one owner, a few days. Without the hypothesis, the next task is "understand the cost base," which is not a task, it is a category, and it is how three weeks disappear. It is the same failure as a tree branch nobody can test, arriving one stage later.

This is the standard method in consulting for a practical reason: the hypothesis functions as a filter on what gets looked at. An educated guess based on limited data, informed by whatever background exists, treated as the start of an investigation rather than its conclusion. Practitioners are explicit that the hypothesis does not have to be correct, it has to be plausible. Plausible is a much lower bar than most people give themselves permission to clear on day one, and clearing it is what makes the work tractable.

The honest version of the objection is worth stating plainly, because it is the first thing a careful person thinks. Committing to an answer before gathering evidence sounds like exactly the mistake you are supposed to avoid. It would be, with one substitution: if you gathered evidence for it. The method only works because the hypothesis is written to be attacked.

Written to Be Killed

A hypothesis and an opinion look identical on the page. The difference is whether anything could change it, and that difference has to be written down at the same time, or it does not exist.

So the hypothesis comes with a second line: what would make me abandon this. If the claim is that unpriced scope on the new delivery model is driving the overrun, the disproof condition might be that contracts signed before and after the change show the same scope-to-price ratio. Write that down first, in advance, while you still have no idea which way it will go.

Writing it in advance is the entire trick. Afterwards, the mind is remarkably good at deciding that the disconfirming evidence was methodologically weak. This is not a character flaw specific to consultants. It is the best-documented pattern in the research on diagnostic error, where clinicians commit to an early diagnosis and then, in the language of that literature, prematurely stop collecting data while selectively accepting the results that support the diagnosis and ignoring the ones that do not. Premature closure is described as one of the most common errors in clinical decision-making, and confirmation bias routinely compounds it. That research is about medicine, not about analysis work, so treat it as a mechanism you can recognise rather than a proven finding about your profession. But the mechanism is not exotic, and nothing about analysis work makes anyone immune to it.

Which is why the day-one answer needs a specific social handling that usually goes unmentioned. Say it out loud as a hypothesis, in those words, with the disproof condition attached. "My working assumption is the delivery model change, and I will drop it if the before-and-after contracts look the same." The phrasing sounds like a formality. It is doing real work: it makes abandoning the hypothesis a normal outcome rather than a reversal, so you are not defending a position when the data goes the other way. A hypothesis you announced as a conclusion is much harder to kill, and the difficulty is social, not intellectual.

The same claim, twice. One has a line underneath it saying what would kill it. That line is the whole difference.

When It Is Wrong, Which Is Often

The failure everyone braces for is being wrong in week two. It is worth being clear that this is not the failure. It is the method working.

A hypothesis killed in week two cost one week and produced something you did not have on day one: a specific fact about where the problem is not, and usually a much better second guess. The before-and-after contracts look the same, so scope is not it, and while pulling them you noticed that the overrun concentrates in a particular delivery phase. That is a sharper hypothesis than anything available on day one, and it arrived because the first one was wrong in a specific enough way to be informative.

The actual failure mode is the hypothesis that cannot die. It gets rephrased each time evidence goes against it, staying always just vague enough to survive. Scope, then coordination overhead, then "execution discipline," which means nothing and therefore cannot be contradicted. Each version is a retreat, and because each retreat is small the direction is easy to miss. The check is to reread the original sentence: if the current hypothesis is broader than the one you started with, you are not converging, you are protecting a position.

There is a genuine limit worth naming. This method is strongest where a plausible guess is available on day one, from domain knowledge or a first look at the numbers. On genuinely novel problems, where any day-one guess would be arbitrary, forcing one anchors the work to a random starting point and the anchor can be surprisingly hard to shake. The move there is to spend a deliberately fixed amount of time, two days, not two weeks, on orientation before committing. The discipline is the fixed limit, not the absence of one. Open-ended orientation is how you get the folder.

Three paths for a day-one answer: confirmed, killed and replaced with a sharper one, or quietly widened until it cannot be tested. Only the third is a failure.

The Version of This You Can Do Tomorrow

Take whatever you are currently investigating and write two sentences at the top of the document, before you open another data source.

I think the answer is ______.

I would abandon that if ______.

Both sentences must be specific enough that someone else could evaluate them. If the first one comes out as "there are probably several factors," you have not written a hypothesis, you have written a description of not having one yet. Push until it names something.

Then look at what you had planned to do this week and ask which of those tasks could change the answer to either sentence. Usually one or two can, and the rest are things you were going to do because they were doable. Those are the ones to drop, and dropping them is where the time comes from.

Do this even if the guess feels premature. Especially then. A hypothesis you are slightly embarrassed by is more useful than a question you feel comfortable with, because only one of them can be proven wrong this week.

The Point

Gathering data without a hypothesis feels rigorous and safe, and it is neither. It defers the moment of judgment to whoever is in the room on the last day, with less time than you had and a folder to read.

Committing early feels reckless and is not, as long as the commitment comes with the condition for its own death. That condition is the whole method. Everything else is just having an opinion in a bigger font.

An answer you can disprove by Friday is worth more than a folder of evidence nobody has organised. Being wrong on Tuesday is progress. Being undecided in week three is not.

Coming Up in This Series

A hypothesis tells you what to look at. It does not tell you how hard to look, and that question has its own answer, which most people arrive at backwards. Analysis has no natural stopping point, so it stops when the deadline stops it, and the last two days reliably go into making a number more defensible in a way the decision cannot see. Next: the 80/20 of analysis, and how to decide what "precise enough" means before you start rather than after you have run out of time.

---

This is part of a series on structured problem-solving and senior-level communication.