The Two Ideas That Cannot Both Be Right

Two well-respected ideas about how to think clearly point in opposite directions. It seems that almost nobody has noticed. If the people who teach these ideas to executives have noticed this, I have not seen anywhere where they have said so.

The first is first principles thinking. The idea is to ignore how things have been done in the past, because habit and convention can cloud judgement. To use this approach, you must break the problem down into the basic facts you know to be true and build your answer up from those. Elon Musk, speaking as chief executive of Tesla and SpaceX, credits this method with making both companies possible.

The approach has been the subject of countless business articles, and courses. The example most people can picture in their minds is IKEA. The story goes that its founder Ingvar Kamprad asked why furniture was expensive, broke the cost into its parts, and found that the wood and fabric were cheap. What cost money was assembling it and shipping it fully built. His answer to this problem was to make the customer do both the assembly and shipping. Anyone like me who has knelt on a kitchen floor with an Allen key and a wordless instruction sheet has been part of the outcome of this thinking.

The second is what the psychologists Daniel Kahneman and Dan Lovallo called the outside view. When you want to know what something will cost or how long it will take, they propose that you should not work it out only from your own point of view, a habit they called the inside view. If you do, you will usually be wrong. You need to find comparable past cases, look at how they turned out, and start from there to solve your problem.

The Oxford researcher Bent Flyvbjerg studied large samples of infrastructure projects. In his data, rail projects ran over budget by an average of 44 per cent and carried on average about 51 per cent fewer passengers than forecast. Bridges and tunnels ran over by about a third, and roads by about a fifth. The overruns were high and they kept happening, which means that from past experience they could have been predicted. A planner who started from how earlier rail projects had actually turned out would have known this. A planner who worked only from the details of their own project would not.

First principles thinking is the inside view taken to its extreme, because it sets out on purpose to ignore comparable past cases. If planning from your own case alone is a known way to get the numbers wrong, then a method that deliberately ignores past cases would surely produce worse forecasts.

A fair objection

The obvious objection is that these two ideas are about different things. 

• Kahneman and Flyvbjerg are talking about forecasting: predicting how something will turn out. 
• First principles thinking is a design method: finding a different way to do the thing.

The objection makes sense, but every first-principles design claim has a forecast built into it. One of the best-known examples is the Tesla battery. In a 2012 interview, Tesla’s chief executive set out the company’s calculation, adding up what the raw materials in a battery cell would cost on the metals market and arriving at about $80 per kilowatt-hour, which was well below what cells then sold for. The outcome is “Cells can be made for $80 per kilowatt-hour”, which is a prediction about cost and is open to the same errors as any other forecast.

SpaceX is also a good test case. The reasoning that reusable rockets would cut launch costs made sense. The estimate of how long it would take and what it would cost was way out: getting the desired result took roughly fifteen years of engineering. The method got the end point right and the time and cost of reaching it were very wrong. That is the mistake forecasting research expects when people plan only from their own case. I have rarely seen this point made explicitly in the first-principles literature.

What the battery example shows

Lithium-ion cell prices had been falling steadily for two decades before Tesla’s calculation. Micah Ziegler and Jessika Trancik, two researchers at the Massachusetts Institute of Technology, published a study of battery price history in 2021. They find prices declining at roughly 13 per cent a year from 1991 to 2018. Every time the total battery capacity ever produced doubled, the cost per kilowatt-hour fell by about a quarter. This was a pattern that was already well researched and recorded.

Tesla’s model assumed the industry believed battery costs were fixed. The evidence never supported that. The cost was already coming down. 

How much of the subsequent fall was caused by Tesla’s Gigafactory, its large battery plant in Nevada, and how much would have happened anyway on the existing downward trend is uncertain, and it was probably some of both. But the popular version only works if battery costs were steady, and they were already falling at a fairly predictable rate.

A lot of the business literature on first principles thinking rests on batteries, rockets and a handful of other successes, picked after the results are in. Failures rarely feature. If a method is judged only on its successes, it can never be shown to be wrong.

As chief executive, Musk has tacitly conceded the forecasting point. He has said publicly that he picks the most aggressive deadline he thinks achievable and is content to be wrong about half the time. His reasoning is that a tight deadline cuts out wasted time and forces people to concentrate on the one problem holding everything up. He uses the number as a management tool rather than as a forecast. That thinking matters, because it makes this a question of judgement rather than prediction.

I have often seen the same optimism when negotiating project costs with consultants. They understand the outcome the client wants but not how complex the work is or how much of it is needed. Halfway through the project the original quote often needs to be renegotiated or, if the client’s budget cannot stretch, the scope is cut instead. The result is a project that never fully delivers what was originally agreed.

Where the method has no competitor

If you become fixated on the outside view, you will always need a comparable past case. 

Where a technology, a market or a problem is genuinely without a history, there is nothing to compare against. First principles thinking is then the only method available, and this is the one situation in which it has no rival.

There have been times in my experience when the financial industry has been faced with unpredictable market failures. In a case like this, the regulators always respond. Sometimes not immediately or pre-emptively, but they do eventually respond. 

The challenge when they do respond is that the industry is required by law to comply with the remedy, even where the market failure has no precedent and there is no past guidance to show that it will be helpful and restorative. 

Even when first principles thinking is the only available option, implementing it carries obvious risk.

How the best forecasters work

Philip Tetlock, a psychologist at the University of Pennsylvania, spent years studying people who are measurably good at predicting events. He calls them superforecasters.

Their method is essentially the same every time: start with the outside view by looking at what happened in comparable cases. Then adjust for the details of the case in front of them, where it differs from those past cases.

The order matters, with the outside view first and the inside view second. It is what the most accurate forecasters do. 

First principles thinking reverses that order.

Flyvbjerg and SpaceX

Flyvbjerg, whose data supports much of this critique, praises SpaceX in the final chapter of his book How Big Things Get Done. He holds the company up as a model of building in small, repeatable units and learning by doing. The author whose work makes the strongest case against the inside-view forecasting of first principles thinking admires the company that was built on it.

Flyvbjerg’s praise is for how SpaceX builds, and it does not extend to how the company arrives at its numbers. His prescription is “think slow, act fast”. This entails long, cheap planning, tested and revised through trial runs, followed by rapid delivery. That fits in with using first principles to come up with a possible answer and not to treat the number that comes out of the process as a firm plan.

Kahneman would probably have accepted both sides of this argument but drawn a different conclusion. In a chapter of his book Thinking, Fast and Slow titled “The Engine of Capitalism”, he argued that entrepreneurs act not because they enjoy risk but because they do not know the odds. Many who weighed them rationally would never start. In the chapter before, he had described executives making decisions on “delusional optimism”, a phrase taken from his 2003 Harvard Business Review article with Lovallo, “Delusions of Success”. Here he treats that same optimism as one of the forces driving growth and innovation.

So, at the level of the individual decision, first principles thinking is likely to often produce bad forecasts, missed deadlines and cost overruns. At the level of the economy, it is arguably what makes anything ambitious get built at all. Had Musk forecast accurately in 2002, he might never have started SpaceX.

Likewise, in competitive financial markets, looking at the numbers and the industry’s past failures, who would be brave enough to open a new bank in the belief that some unique ability or conviction would let it compete with the well-established mega-banks?

South Africa has two good examples, TymeBank and Bank Zero, both of which spent about five years rebuilding their idea of banking from scratch. Each built a bank without branches and on cloud infrastructure, rather than copying the mega-banks’ tried-and-tested model of extensive bricks-and-mortar infrastructure.

Bank Zero planned to launch in late 2018 and only opened to the public in August 2021. Had its founders known the actual timeline at the outset, the case for starting the bank would have looked a great deal weaker, which is Kahneman’s point exactly.

Forecasting mistakes may get things built, but in reality, the people who succeed are normally the ones who know they are making them.