By the late 1960s Barbara Minto had an unusual job at McKinsey. She had joined the Cleveland office in 1963, by her account the firm’s first woman consultant. She was no star at crunching numbers, but she could write, so her manager started handing her the consultants’ reports to fix. Eventually, he recalled, every office in the firm was borrowing her.
Her manager put the need delicately: the writing “often wasn’t exactly what it should have been for graduates of places like Harvard Business School.”
So the job looked obvious. Tighten the sentences, fix the grammar, cut the jargon. When McKinsey moved her to London in 1966, she went on trips to its offices in Paris and Düsseldorf, and found the same reports there: findings and conclusions jumbled together, neither leading anywhere clear. Other countries, other languages, the same problems.
That was the clue. “The problem was the thinking, not the language,” she said, looking back in 2018. “People were starting to write without working out their thinking in advance.”
Which left her with a harder question, in her own words: “But how does one go about figuring out one’s thinking in advance?”
Fifteen beards
She found part of the answer in her own red pen. Editing all those reports, Minto noticed she kept rearranging the ideas into the same shape: one point at the top, a few ideas under it that the point summed up, and under each of those, the facts behind it. A pyramid.
Her book on it has an example that works on anyone. Suppose, she writes, I join you for a beer in the pub and, apropos of nothing, say I was in Zurich last week, a conservative city, and we went out to lunch at an outdoor restaurant, and within 15 minutes I must have seen 15 people with a beard or a moustache.
You start guessing. She disapproves of the hairy style? She’s going to compare Zurich with other cities? She adds that facial hair has been part of the London scene for years, and you think you’ve finally got it: London is ahead of all the other cities. “Perfectly logical,” Minto writes, “but it’s wrong.”
What she meant was that it’s incredible how accepted facial hair has become in business life. Said first, that sentence gives every beard somewhere to go. A reader, she writes, “remembers from the top down as a matter of course. He also comprehends more readily if ideas are presented from the top down.”
Start with your answer, then give the reasons under it, one layer at a time.
The rules that hold the pyramid together are few. A point must sum up the ideas directly below it. The ideas grouped under it must be the same kind of idea, in a logical order. And each group should be MECE, mutually exclusive and collectively exhaustive, an abbreviation Minto says she was the first to use for groups of ideas.
The point comes last when you think
Her book’s first chapter ends with two sections back to back: “Ordering from the Top Down,” then “Thinking from the Bottom Up.”
The pyramid is presented from the top, but it is usually found from the bottom. You start with a pile of facts, group the ones that belong together, and ask what each group says. “The pyramid is a tool to help you find out what you think,” Minto says. It forces you, she adds, to pull out of your head information “that you weren’t aware was there.”
Her course adds a step before any of it: work out the question already in your reader’s mind. The top of the pyramid is the answer to that question.
Three weeks late
Say a logistics coordinator has to tell her director that the spring stock will arrive three weeks late. Her first draft follows the week as she lived it: the supplier’s call on Monday, the port backlog, the partial shipment, the price of air freight. The request comes in the last line.
Her director may read that email on a phone between two meetings. The question in his head is what she needs from him. He gets Monday’s phone call.
She sorts her notes first. They fall into three piles: what’s late and why, what that costs in empty shelves, and what it would cost to fly part of the stock in. Only then does the top line write itself: approve air freight for the two best-selling lines by Thursday, or those shelves are empty for most of April. The three piles go underneath, in that order.
The facts are the same as in the first draft. He might now reply before his next meeting starts.
Two readers, two orders
A language model can sit on either side of the page, and the order works differently on each.
When the model writes for you, asking for the bottom line first is reasonable, as long as the thinking still happens first. Anthropic’s prompting guide, as it stood in October 2026, says that on its newest Claude models thinking is always on, so a response can open with thinking blocks before any visible text. For setups where thinking is turned off, it suggests asking Claude to think the problem through before it answers, and to put the final answer in its own tags. That is Minto’s split: thinking from the bottom up, then presenting the top.
When you write for the model, the guide suggests something a busy director would hate. For long inputs, roughly 20,000 tokens and up, put the documents at the top of the prompt and the question at the end. Queries at the end, it says, can improve response quality by up to 30 percent in tests, especially with complex inputs made of several documents. The guide doesn’t say why.
The director gets the answer in the first line, because he may stop reading after two. The model takes in the whole prompt either way, and on long documents, the guide puts the question last.