Disruptive Leadership

Electrek has published a company-wide email from Elon Musk. It contains pithy proscriptions to cheer the soul of anyone wearied by corporate bureaucratic processes. Here are two:

– Communication should travel via the shortest path necessary to get the job done, not through the “chain of command”. Any manager who attempts to enforce chain of command communication will soon find themselves working elsewhere.

– A major source of issues is poor communication between depts. The way to solve this is allow free flow of information between all levels. If, in order to get something done between depts, an individual contributor has to talk to their manager, who talks to a director, who talks to a VP, who talks to another VP, who talks to a director, who talks to a manager, who talks to someone doing the actual work, then super dumb things will happen. It must be ok for people to talk directly and just make the right thing happen.

This is so great! Now nobody has to get permission from their boss to do anything! If you think it’s the right thing to do, do it!

One thing to keep in mind, though. People sometimes disagree about what the right thing to do is. Sometimes you can try several options and see what works best, but this is rare, particularly if money is tight or time is tight. We can probably assume time is tight at Tesla.

For example, let’s start with another part of Elon’s letter:

Some parts suppliers will be unwilling or unable to achieve this level of precision. I understand that this will be considered an unreasonable request by some. That’s ok, there are lots of other car companies with much lower standards. They just can’t work with Tesla.

Let’s say there is a supplier who hasn’t been meeting expectations. Someone from Sourcing decides to find another supplier. Someone from Purchasing negotiates with the existing supplier to reach a mutually acceptable agreement. Someone from Design decides to change the design so the part is not needed. Someone from Manufacturing figures out a quick, efficient way to modify the part during assembly to resolve the issue while still meeting cost targets.

Is it realistic that all of these solutions could be found within the same time frame? No. Is it realistic that four different people start working on different, overlapping solutions? Yes. And they probably each need to work with three or four other people, discussing options, agreeing on changes, determining who will do what, just to figure out what the end results (both costs and benefits) would be from pursuing their initial idea. So how will this group of sixteen people (an implausibly small number) react when one solution is chosen?

Four people will feel great. Twelve people will feel that they have wasted their time and efforts, and they will be less willing to help in the future – both within their sub-group of four, and within the larger group of people who were involved in looking for a solution.

So who makes sure we don’t have sixteen people working on the same problem in an uncoordinated overlapping or contradictory fashion? Who makes sure that the solution to Problem A does not cause new problems with the solution to Problem B?

To use a different example, if we are driving around town and looking for a parking spot, who decides whether we turn right or left, or continue on straight through the intersection?

That’s called “chain of command.” Chain of command means that Person A gets to decide about Topic X and we will all live with their decision.

What Elon is suggesting here is fully possible, within boundaries. Maybe any decision that costs less than $1,000 dollars, just do it – costs more and takes more time to debate it than to try, and then try something else if it doesn’t work. But ideally you would want to make sure the decision isn’t going to have a lot of unexpected consequences on other people, and, generally speaking, managers are more likely to have the experience and information to recognize if a decision is going to unintentionally affect other people too.

Speaking of boundaries, check out this part of the letter:

All capital or other expenditures above a million dollars, or where a set of related expenses may accumulate to a million dollars over the next 12 months, should be considered on hold until explicitly approved by me. If you are the manager responsible, please make sure you have a detailed, first principles understanding of the supplier quote, including every line item of parts & labor, before we meet.

Ok, let’s just try to get a very rough sense of perspective on this.

Elon wants to build 5,000 cars per week. Call it 50 weeks per year, leaving two weeks out for factory upgrades. That’s 250,000 cars per year, so any decision having an impact of $4 per car has to be approved by the CEO.

Tesla spent about $4.2 billion USD last year on automotive operations, up from about $2.6 billion in 2016. That is an increase of about 4.5 million dollars every day of 2017. Four and a half decisions per day is not too much to ask of a CEO, but usually these decision are not simple one-time snap decisions like: cream or no cream?

Tesla had 37,543 employees in 2017, according to Statista.com, and in the letter Elon mentions adding 400 people per week for “several weeks”. So let’s say 40,000 employees and figure 200 work days (Elon probably works 7 days a week but that is not likely to be representative); anything that adds a cost of 13 cents per day per employee has to be approved by the CEO. That’s either one decision for the whole year, or one thirteen-cent decision every day, or perhaps one 90 cent (roughly) decision every week, which would “accumulate to” a million dollars in 12 months.

All of these examples are only outer limits, of course; if every single employee were making decisions every day that crossed this threshold, Tesla would be spending even more money. But, for context, in one place I worked every employee could spend up to $500 on a single transaction (or for a single purpose), and it actually is quite limiting in a business context. It wasn’t enough to pay for an hour of time from the videographer.

If no manager at any level can spend a million dollars, limits will be correspondingly lower for an assembly line supervisor than for a Chief Office or Vice President. People might even have to get approval from their, well, chain of command.

Speaking of that, let’s see what happens when you don’t go through your chain of command:

I have been disappointed to discover how many contractor companies are interwoven throughout Tesla. Often, it is like a Russian nesting doll of contractor, subcontractor, sub-subcontractor, etc. before you finally find someone doing actual work. This means a lot of middle-managers adding cost but not doing anything obviously useful. Also, many contracts are essentially open time & materials, not fixed price and duration, which creates an incentive to turn molehills into mountains, as they never want to end the money train.

Clearly Elon didn’t approve these arrangements, so I have to assume that someone just got the job done, did the right thing, maybe even ignored a “company rule” that was “obviously ridiculous in a particular situation.”

You get the sense, reading this comment, that Elon thinks time and materials contracts are just obviously ridiculous to a common sense understanding. So why would anyone set up a contract that way?

Well, “time and materials” would be a good fit if you didn’t have a really detailed idea of what all the necessary work would be to get to the end result you wanted, or you don’t know all of the precise details that define every aspect of the end result. If you’re not sure or you’re not capable of explaining it well to your supplier, they aren’t going to want to promise to delivery whatever exactly it is you want for a fixed price. But if you tell them, “Hey, let’s figure it out together as we go, and we’ll pay you for your time and materials,” well, now you can get somewhere.

“Figure it out as we go” can be an appropriate strategy where time is of the essence, and it’s pretty easy to see how someone working at Tesla might feel that way.

Going back for a second look at a piece already quoted:

If you are the manager responsible, please make sure you have a detailed, first principles understanding of the supplier quote, including every line item of parts & labor, before we meet.

“First principles understanding” is an excellent approach for figuring out whether something is ultimately possible. It is a terrible way to estimate how successful your first attempt will be. And “detailed […] every line item” is kind of the opposite of “first principles,” in the sense that it has to include all of the things that are not “first principles,” like taxes, insurance, overhead, risk, etc.

What Elon has accomplished already is astonishing by any human measure. The production quality and quantity of cars from Tesla has not been perfectly on track with expectations, but nobody else has done better starting from zero like Tesla. Tesla’s burned through a lot of cash, but other people have gotten less done for the money. Elon’s been a jerk to at least some people, but other people are jerks and also totally ineffective.

I am not brave enough to bet against Elon Musk.

He’s going to die some day, I’m sure of that, and I’m pretty sure that before that happens at least one of his business ventures will collapse; his empire will be split up, and some or all will go on under different management. Eventually Alexander the Great turned back, you know, but I wouldn’t be the one to go looking to pick a fight with him.

Still, if I had to describe Tesla so far, I’d say that with company, Elon has accomplished astonishing things by spending an astonishing amount of money. Now he’s asking his company to keep accomplishing astonishing things but stop spending astonishing money. I don’t think it works like that. He’s asking for each individual to make independent, quick decisions, each decision reflecting the collective intelligence of the company – but don’t hold meetings. I don’t think it works like that.

I am not envious of anyone working at Tesla right now.


One last quote:

Walk out of a meeting or drop off a call as soon as it is obvious you aren’t adding value. It is not rude to leave, it is rude to make someone stay and waste their time.

This I agree with. This is the kind of meeting you set up when you’re not sure who needs to be involved or exactly what decision needs to be made, when you aren’t fully prepared but you just hope to figure it out in the moment. I’ve been guilty of setting up those meetings myself.

On the whole, it is better to include people than to leave them out. But that needs to be balanced with allowing people to decide for themselves that they aren’t needed after all, without any recriminations.

Toyota’s chance to score twice on a single play

Developers of autonomous driving software should realize that there is a precursor market for their technology in the material handling world—specifically for forklifts.

While the industry chatter could lead you to believe automated forklifts are available and capable today, experience behind the scenes reveals that most forklifts are not automated and most automated forklifts can only be trusted with the cleanest, clearest, least variable pallet handling tasks.

Automotive technology is rapidly advancing far beyond the limitations of current warehouse offerings. While Waymo’s automation software is making guesses about the future path of a bicyclist and driving through construction zones, automated forklifts struggle to figure out how to get forks in a pallet if the pallet is not presented with mechanical precision.

Forklifts are designed to move pallets; pallets are designed to provide a common handling interface for a virtually unlimited range of goods. In theory this should make pallets an excellent aid to automated handling. But reality gets messy.

Pallets can sag under load, to a degree that it becomes difficult to insert and remove the broader, thicker forks of most horizontal-transport lift trucks. Pallets are loaded unevenly, double-stacked, wedged together, and sometimes broken. Commonly used stretch wrap (think industrial grade version of kitchen plastic wrap) sometimes covers the fork pockets in a way that would make some sensors believe the pockets are blocked; human operators know to punch right through.

But all these challenges, and others like them, could be easily overcome with judicious use of sensors and the kind of learning algorithms, commonly called Artificial Intelligence (AI), which do best when the same basic task is performed again and again. Repetition allows the algorithm to iteratively refine its definition of what the core task is, which variations are acceptable, and which factors lead to failure—and forklift failures can be expensive.

It makes sense that major players in automated vehicles are not focusing on forklifts. The forklift market is vanishingly small compared to the automotive market. The Industrial Truck Association reports about 200,000 lift trucks shipped in North America in 2015. Automation technology replaces the operator, and it is “widely accepted that the operator makes up about 70% of the total cost of ownership of any lift truck,” according to trade publication Modern Materials Handling. Glassdoor.com reports that operators earn $30,000 annually on average. Assume that benefits will double the business cost to $60,000.

Not every forklift is doing a job that makes sense to automate; some are used only occasionally, or for moving a wide variety of non-standard and even unpalletized loads. Let’s assume that it only makes sense to automate half of the forklifts, or 100,000 units per year. If on average and operator costs are about $60,000 each, that’s a $6 billion annual market opportunity.

That’s not even on the radar of the automotive market, which the Wall Street Journal reports at $570 billion in 2015 sales for the USA. And that’s for the car itself, not for paying the operator.

But trying get the material handling market alone to pay off the R&D of autonomous driving is applying the wrong analysis to the opportunity. Automating forklifts is a secondary market – or more accurately, a precursor market – providing an opportunity to teach AI about the behavior of pedestrians and human drivers in a structured, controlled-access facility. Although some adaptations of the AI would be required for pallet engagement and lifting loads to the 40 foot heights of modern warehouses, these are minor tweaks on the core task of avoiding collisions and predicting human behavior.

Much like public roads, most warehouses have clear rules about where people should be walking and where vehicles may travel. But, also as can happen on the road, these rules are often subject to temporary exceptions. This provides an excellent opportunity to extensively test basic maneuvering and collision together with anticipating human behavior in a context where it is only semi-predictable. Warehouses are for the most part lacking significant grades (seeing “over” a hill is a problem for a scrupulously safe automated car) and they lack weather, which means that a warehouse environment is not a perfect test environment in all respects. At the same time, removing these complexities means a lower technical hurdle to get the automation truly ready for full hands-off.

Today only Toyota Industries has a stake in both the automotive and forklift markets. Toyota is lagging in the autonomous driving segment, but starting to push harder. Toyota can leverage the same investment twice, sending its automotive technology over to its forklift division – or it can wait for a start-up that originally aimed at road navigation to pivot to forklifts, as happened with fuel cells and lithium ion batteries.

Out on the open road, the big opportunity is replacing paid drivers, typically truck drivers, for the same reason as in forklifts: other than fuel, drivers are virtually the entire cost. Also, they’re getting harder to find. The chance to automate both tasks is the opportunity to dominate the logistics segment like Amazon, Google, or Apple in their respective domains.

Automation cannot and will not entirely do away with the need for humans in logistics for some time to come; they will need to work as caretakers, facilitators, and troubleshooters alongside the automatic equipment. No prior automation, from the industrial loom to the horseless carriage, ever arrived on the scene fully capable without human attention. But the well-proven preference people have for lower prices, and the increasing difficulty finding people to work in logistics jobs, means automation is inevitable, and the question is only who will be most successful in providing it.

Some businesses find themselves in the lucky position of having a marketable by-product on the way toward the product they set out to deliver. Gasoline started as waste from oil refining. Amazon’s cloud service started as a repackaging of a tool developed for internal needs. Now the opportunity exists for one of the companies on the path to autonomous driving to get some early revenue and real-world feedback — an off-highway alternative market for autonomous driving tech. Will Toyota seize their advantage? Or will it take a start-up to put things together in an unconventional yet effective combination?