FRAME

In the middle of 1985 Andy Grove, Intel's president, sat in his office in Santa Clara. With him was Gordon Moore, Intel's chairman and chief executive. Through the window Grove could see the Ferris wheel of the Great America amusement park turning in the distance. Intel had started out making memory chips, and now Japanese rivals were taking that market by selling cheaper chips. For almost a year Intel had drifted, with no plan for its memory chips. Grove turned from the window and asked Moore a question. If the board fired them both and brought in a new chief executive, what would the newcomer do? Moore answered without hesitating: the newcomer would get Intel out of memory chips. Grove stared at him, numb. Then he asked why the two of them should not walk out of the door, come back in and do it themselves.

That autumn Intel said it would stop making DRAM, the memory chip it had pioneered. The next year it lost $173 million, its first yearly loss since its shares went on sale in 1971. By 1993 Intel was the world's largest chipmaker. It made its money from processors, the main chips in personal computers.

People usually tell this story as a lesson in courage, and that is the mistake. The decision looks right because everyone knows what came next. Suppose IBM and the companies copying its computers had switched to another company's processor. Then that talk by the window would be remembered as the day Intel's leaders gave up the business they had built. In 1985 nobody could know for sure which of the two stories would come true. Grove described the scene in a book eleven years later, when he already knew the ending.

Judging a decision by its ending is a natural habit with three faults. First, the ending comes late, often years after the choice. Second, the ending mixes judgement with luck. A careful decision can fail because a rival moved first, and a careless one can succeed because the whole market grew. Third, knowing the ending changes how we read the beginning. The psychologist Baruch Fischhoff tested this for his doctorate at the Hebrew University of Jerusalem and published it in 1975. He gave people short descriptions of events from history and medicine, each with four possible outcomes. He told some readers which outcome had happened. Often he named one that never happened. Whatever outcome they heard seemed more likely to them than to readers who heard nothing. They also believed they would have seen it coming.

So before the ending, you can judge only the reasoning. Reasoning is far easier to judge once it is written down: others can check it, and the writer cannot quietly change it later. Nobody wrote down Grove and Moore's talk. Many decisions are like that: settled in a room and explained afterwards. But some decisions are written up before they are approved. The person who approves such a page cannot know the future, but can still ask five questions. Does it name the real decision? Does it rest on the facts that matter, or on a story that sounds right? Does it say which option it turns down, and at what cost? Does it say what is unknown, and what would change the writer's mind? Does it commit to a first step, and a date to check the result? The five checks are Focus, Relevant evidence, Alternatives, Manage uncertainty, and Execute and learn. Their first letters spell FRAME.

Here is such a page, dated September 1985. Grove and Moore never wrote it. We did, using only facts Intel had at the time.

F · Focus: what exactly is being decided?

By the end of September we decide whether Intel stays in the market for DRAM or leaves it. Either way, we will call the decision a success if Intel makes an operating profit again within two years. Our other memory chips, such as EPROMs, are not part of this decision.

Before Grove's question, Intel had spent a year arguing about how to win in memory chips. An argument like that can go on for ever. With one question and one answer, Grove and Moore turned it into a yes-or-no choice. We added what their talk left out: a date, a test of success and a boundary. We need the boundary because Moore spoke of all memory chips, but the memo decides only about DRAM. The first lines of any decision memo need a yes-or-no choice with those three things.

R · Relevant evidence: which facts would change the answer?

Only one of our eight factories still makes DRAM. Our processors have been inside IBM's personal computers since 1981. Our 286 runs IBM's most powerful machine, the PC AT, and our 386 is about to come out. We assume that personal computers keep selling and keep being built around our processors.

The first fact tells the most: the decision was already half made. Robert Burgelman, a Stanford professor who later studied the episode, found out how. Factory space at Intel was scarce. Middle managers gave it to whatever earned the most profit per silicon wafer, the thin disc that chips are cut from. Under that rule, production had been moving from memory to processors for years. So the managers had started the move long before Grove and Moore decided on it.

A · Alternatives: what is rejected, and at what price?

Our managers have proposed three ways to stay. One is a giant factory just for memory, to beat the Japanese on cost. The other two are a new technology they cannot copy and special memory for narrow markets. The factory is the strongest, because it is the only one that goes straight at our costs. It works only if a new plant can make memory for less than the Japanese sell it for. None of our plants can do that today. The factory would also take money and engineers the processors need.

In his book Grove lists all three. He believed less and less in narrow markets, because memory was becoming a commodity: one product at one price all over the world. In a narrow market you can keep a high price only until rivals crowd in.

M · Manage uncertainty: what is unknown, and what would change the decision?

The open question is whether Intel can live on processors alone. If it cannot, we will have given up our oldest business with nothing big enough to replace it. Leaving is hard to undo: once the DRAM teams break up and the plant is converted, coming back would take years. We will reopen the decision if PC makers start designing their next machines around another company's processor. We will check for that every quarter.

The last two sentences matter most. In them the writers name news that has not come yet, and decide now what to do if it comes. So they choose their reaction while they are calm.

E · Execute and learn: what happens first, and when is it checked?

We leave DRAM and announce it on 10 October. Andy Grove is in charge of the exit. At the end of 1987 we will compare our results with this page and note what we got right and wrong.

Intel did announce its exit on 10 October 1985. But neither Grove's book nor Burgelman's study mentions a date for comparing the result with what anyone expected. Intel learned from the decision all the same. Grove wrote the lessons down only eleven years later, in a memoir, when he knew the ending.

What the page is worth

One warning about this memo. Intel knew every fact in it in September 1985. But we wrote it knowing the ending, and that helped us choose which facts to put in. This is what Fischhoff found, and it happened to us. That is why the page has to be written before the ending and dated. A page written that September would be worth the same whichever way the processor market went. Either way, we would judge it on what the writer saw, what they assumed and which news they promised to watch for.

The page also makes the ending useful. From one ending you learn little about anyone's judgement. From a run of endings, each set against a page written beforehand, you learn much more. In forecasting tournaments, people write down forecasts with a date and a probability, and each is scored when the outcome arrives. Over hundreds of forecasts, you can tell skill from luck. From 2011 to 2015 IARPA, the research agency of America's intelligence services, ran such a tournament. The Good Judgment Project won it. Its leaders were Philip Tetlock, Barbara Mellers and Don Moore. Moore teaches at the University of California, Berkeley, and is no relation to Gordon. A manager makes far fewer decisions than a forecaster makes forecasts, and rarely gives them a probability. So the scoring is rougher. But the habit is the same: a dated page for every decision that matters, compared with the outcome when it comes.

A common mistake is to fill all five parts with empty phrases: "the framing here is strategic", "there is significant uncertainty", "alternatives were considered". Every part is filled in, but nothing has been checked. Under the right heading, an empty sentence looks like a finished check. So take the headings away and test each sentence. Could someone else check it against something outside the page? That might be a number, an option not taken, or news that would make the writer drop the recommendation. If nobody else could check a sentence, it is only decoration.

Where the five checks come from

FRAME is our own list, but each of its checks has a source. Focus comes from decision analysis, the field Ronald Howard of Stanford named in the 1960s. In decision analysis, you start by defining the problem. Relevant evidence draws on Richards Heuer, a CIA analyst for about 45 years. He taught analysts to look for the few facts that tell one explanation from another.

Alternatives comes from Paul Nutt of Ohio State University. He followed 356 decisions in organisations, and about half failed. Failure was more common where a manager picked a solution early and stopped looking for others.

Manage uncertainty joins two ideas. In Gary Klein's premortem, a team imagines its plan has already failed and explains why. Chip and Dan Heath's tripwire is an event you agree on in advance: if it happens, you look at the decision again. The dated check in Execute and learn comes from the forecasting tournaments.

Using FRAME on the cases

Every case on this site stops where someone had to decide and asks what you would do. Write your page there, before the story goes on: a few lines for each of the five checks. The ending will not tell you whether you reasoned well. But your page will show what you saw, what you assumed and what would have changed your mind. You can compare that with what the company knew at the time.

An AI reviewer scores your answer on these five checks without seeing how the story ends. Each check gets 0 to 20 points, and the five add up to a rating out of 100. The rating is only advice: you can rewrite the answer and ask for a new rating. Below are the methods for each check; they open next to the page. Find the check you are stuck on. The last line adds two methods for checking a finished memo:

• F: clarity test, map is not the territory
• R: competing hypotheses, survivorship bias, kind and wicked learning environments
• A: inversion, steel-man, opportunity cost, second-order thinking
• M: outside view, pre-mortem, fragility audit, falsifiability
• E: two-way and one-way doors, commander's intent, decision journal
• Before you send: Decision Quality: six links, red team

References

• Andrew S. Grove, Only the Paranoid Survive, Currency Doubleday, 1996.
• Robert A. Burgelman, "Fading Memories: A Process Theory of Strategic Business Exit in Dynamic Environments", Administrative Science Quarterly 39:1 (1994), 24–56.
• Intel, "Farewell to DRAM" and "Financial Year 1986", Intel timeline.
• Baruch Fischhoff, "Hindsight ≠ foresight: The effect of outcome knowledge on judgment under uncertainty", Journal of Experimental Psychology: Human Perception and Performance 1:3 (1975), 288–299.
• Ronald A. Howard, "Decision Analysis: Applied Decision Theory", Proceedings of the Fourth International Conference on Operational Research, 1966.
• Richards J. Heuer Jr., Psychology of Intelligence Analysis, CIA Center for the Study of Intelligence, 1999.
• Paul C. Nutt, "Surprising but true: Half the decisions in organizations fail", Academy of Management Executive 13:4 (1999), 75–90.
• Gary Klein, "Performing a Project Premortem", Harvard Business Review, September 2007.
• Chip Heath and Dan Heath, Decisive: How to Make Better Choices in Life and Work, Crown Business, 2013.
• Philip E. Tetlock and Dan Gardner, Superforecasting: The Art and Science of Prediction, Crown, 2015.