Leading hypothesis: structuring a case with a hypothesis

A leading hypothesis is a precise statement, one the facts can contradict, about the cause of a problem or the best decision to make. "Sales are falling because we lost shelf space at one retailer" is a hypothesis. "There is a problem on the sales side" is not, because no data can refute it.

The approach is the scientific method applied to business: you put forward an explanation, work out what you should observe if it is true, and go and check. It is at the heart of how strategy consulting firms work, where a team never has time to analyse all of a client's data and has to choose what to look at first.

There are two ways to structure a case. The exhaustive approach opens every branch of an issue tree and goes through them one by one. The hypothesis-driven approach keeps the tree in mind but picks a branch and says why. Good candidates do both: a complete structure so nothing is forgotten, a hypothesis to decide on the order.

A hypothesis unfolds into a tree. Under the statement, you write the conditions that all need to be true for it to hold. Under each condition, the data that lets you check it. The tree gives you directly the list of what you are going to ask the interviewer for, in order.

The most useful condition to test first is the one that, if false, brings down the whole hypothesis. In the example below, if the drop is the same at every retailer, the shelf explanation no longer holds and you move on to another.

Here is the prompt. Laiterie des Monts sells yoghurts through supermarkets. In one year its market share went from 20% to 17%. The CEO believes consumers are turning away from the brand and wants to step up advertising. What do you think?

State the hypothesis. You notice two things in the prompt: the drop is fast, and the product has not changed. Consumers falling out of love with a brand is usually gradual. So you announce: "My starting hypothesis differs from the CEO's. A three-point fall in a year, with no change to the product, suggests a distribution problem rather than a preference problem. I would like to check by looking at market share retailer by retailer."

Test the condition that decides. The interviewer gives the numbers. The brand is sold through four retailers, each accounting for a quarter of the market.

The average of the four does give 17%. The whole drop comes from retailer D. The CEO's hypothesis is ruled out: consumers tired of the brand would be tired of it at every retailer.

Test what follows. You ask what changed at D. Answer: the retailer launched its own yoghurts and halved the number of Laiterie des Monts products on its shelves.

Conclude. "The drop does not come from consumers but from a decision by retailer D, which halved our presence to make room for its own brand. A national advertising campaign would fix nothing. The priority is to renegotiate with D, and to check whether the other three retailers are preparing the same move."

Three questions were enough. An exhaustive route would have examined price, product, advertising and competition before getting to distribution.

State a hypothesis when you have a clue, and stay exhaustive when you have none. The clue may come from the prompt (a date, a number that stands out, a detail the interviewer took the trouble to mention), from an answer obtained during clarification, or from your knowledge of the industry.

Without a clue, a hypothesis is just a roll of the dice, and you risk locking yourself into the wrong lead in the first minute. It is then better to present a complete structure, get a first number, and form your hypothesis from that number. In practice this is the most common situation: the hypothesis comes after the first piece of data, not before.

The approach breaks when you try to be right. Once the hypothesis is announced, the temptation is to read every piece of data as confirmation and to play down whatever contradicts it. This is confirmation bias, and it is costly in an interview: the interviewer hands you a number that invalidates your lead, you miss it, and you build your recommendation on a false cause.

It also breaks when the hypothesis is too vague to be tested. "I think the problem comes from costs" does not say which data to ask for. A good hypothesis names a mechanism: which cost, for what reason, since when.

Third trap: forgetting the map. If you set off on a hypothesis without having laid out a structure, you have nowhere to go when it falls. You improvise a second lead, then a third, and the case becomes a string of attempts.

Finally, some formats do not suit it. When the interviewer leads the case question by question, they expect an answer to each question, not a general theory announced in the first minute. In that situation keep your hypothesis to yourself and use it to interpret each new piece of data.

Announce your hypothesis as a working direction, with the clue behind it and the test that will settle it. A three-part wording works well.

1. The clue. "The drop is sudden and the product has not changed."
2. The hypothesis. "So I suspect a distribution problem rather than consumer preference."
3. The test. "To check, I would like to see market share by retailer."

Where possible, also say what would change your mind: "If the drop is the same everywhere, I will drop this lead and look at price." The interviewer sees that you are reasoning, not guessing.

When the data arrives, decide out loud. If it confirms, go down a level. If it refutes, say so plainly, return to your structure and take the next branch. Dropping a false hypothesis cleanly is a sign of rigour, not a failure.

At the end, your validated hypothesis becomes the top of your recommendation, which you present following the Minto pyramid. In the FrameLab course, structuring with a hypothesis is presented as a way to prioritise a tailored structure, never as a way to do without one. You can practise on the real interview cases: before reading the first piece of data, write down the cause you find most likely and what would prove it.

A dental clinic established for ten years in a business district has seen its number of appointments fall by 25% in eighteen months. Its prices and its team have not changed. State a hypothesis, the conditions that must be true, and the first piece of data you would ask for.