As a pilot, I’ve enjoyed the ability to fly to interesting places. However, every so often less than adequate weather will leave you stranded somewhere waiting for the overcast to lift, fog to boil off etc. One of the most common causes of accidents in general aviation is untrained pilots inadvertently flying “into the clouds”. It is almost always lethal to pilots who aren’t trained for it. Without external visual references and a visible horizon to help a pilot determine what is up and down, our human physiology fails us and our sense of gravity becomes confused. On average this gives an untrained pilot 178 seconds to live. This is what happened to JFK Jnr when he crashed his light aircraft into the ocean close to Martha’s Vineyard. The solution is to get trained to fly completely by reference to one’s instruments only and obtain what is called an “instrument rating”. Obtaining this rating is so complex that it takes the same amount of time as it does to train for the original pilot’s license. While doing my instrument rating in 2015 I noticed a number of interesting parallels to the world of technology startups and data interpretation.
1. Your gut instinct can kill you
We use three senses to determine our position in space: visual, vestibular (inner-ear) and postural (nerves in skin, joints and muscles). When the human body is subjected to the forces of flight in zero visibility conditions, we loose our most important sense (visual), and the remaining vestibular and postural senses provide misleading information leading to what is called spatial disorientation. Pilots may end up thinking they’re flying straight and level, but are actually in a deadly spiral on the way to crashing.
For this reason, pilots are taught to rely exclusively on their flight instruments to determine their position and attitude. In a startup, once you find some level of product-market fit, data provided by your users should largely drive where your company and product is going. This is where you need to follow the data to override gut in much the same way a pilot needs to rely on instruments and not “seat of the pants” flying. Once a company has gained some traction and has sizeable datasets to evaluate, it becomes important NOT to just trust your gut, but instead to apply proper scientific method and evaluation of data experiments. This is sometimes a difficult transition for entrepreneurs and founders. In this complex environment, the information contained in the data supersedes the gut feel of the entrepreneur. This is particularly important when looking at conversion funnels and A/B testing experimental results. If entrepreneurs insist on believing only in their gut, very often the company ends up with its own version of 178 seconds to live.
2. Early optimization can lead to wild swings in results
When a pilot does learn to fly by instruments, there’s an additional problem; because flying instruments work with air pressure, there’s a delay between control inputs and when the resultant speed or altitude change registers on the instruments. So if a pilot sees that he’s fast loosing altitude and he sharply pulls back on the stick, it takes a while before the vertical speed instrument shows a climb. Yet only after some time has passed does the airspeed indicator finally react to the climb. The pilot now chases the dials, pushing sharply forward again, and so repeating a cycle of over-controlling the aircraft resulting in a ride that resembles a rollercoaster. This is called pilot induced oscillations, and can be violent enough to destroy the aircraft. The trick is to make smaller control inputs until there is some reaction on the instruments, and then correct from with smaller movements, careful not to overcompensate. When it comes to data and complex systems, it’s the same reason why the Fed doesn’t announce a new interest rate every morning. Interest rate inputs take time before they show on macroeconomic indicators. For this reason, software engineering guru Donald Knuth is often quoted saying:
“Premature Optimization is the Root of all Evil”
So in a startup, when you ARE using data, over-focusing on a specific metric can lead to similar wild oscillations in results, specifically when there are delays from the time the change is made to where it shows on dashboards.
3. Checklists can save you
On October 30th 1935 at Wright Airfield in Ohio, a Boeing B17 Flying Fortress took off in a demonstration flight in front of an assembly of Army brass, in an attempt to win the bid for the Army’s new long-range bomber. It was supposed to be a mere formality. The new plane had trounced designs by competitors like Martin and Douglas, and could carry five times as many bombs and fly twice as far. The B17 climbed 300ft, banked sharply, and crashed. Turns out the very experienced test pilot Major Hill forgot to remove the new locking mechanism for the elevator and rudder controls. The press dubbed the B17 “too much of an airplane for one man to fly”.
The Army responded with a new innovation — the pilot checklist. Increased complexity in flying, combined with forgetting simple steps due to distractions, made checklists necessary. Similarly in The Checklist Manifesto Atul Gawande describes how a simple surgical checklist from the World Health Organization has been adopted in more than twenty countries as a standard for care and has been heralded as “the biggest clinical invention in thirty years.”
When you’re releasing new software under pressure, when you’re making a late-night critical change on a production system supporting your first million users, produce and use checklists to avoid disaster.
4. Simulation and scenario planning helps with difficult choices
Thanks to Moore’s Law, powerful flight simulators are now available to smaller general aviation flight schools and individuals. From my study at home I’m able to practice complex approaches in a photo-realistic environment before I even step into the plane. Additionally, in a nice example of a two-sided market, trainee flight controllers offer a service whereby they provide real-time air traffic control over the internet to pilots flying in the virtual environment (this adds a complete new layer of stress in talking to a real controller).
In a startup it becomes essential to simulate growth scenarios with tools like system dynamics, financial scenarios in excel, and the expected effects on retention given a product change. Andrew Chen in his courses and talks on growth hacking stresses the importance of first building a growth model to understand the underlying dynamics of the business. Such quantitative modeling skills are essential these days in plotting strategies at a startup — or any complex business, for that matter.
Technology entrepreneurship shares a lot of the speed and uncertainly of operating a fast-moving plane in zero visibility. With real lives at stake the aviation industry has learned how to derisk these activities. Entrepreneurs could certainly learn from pilots in adopting data analysis skills, refraining from over-controlling key metrics and using checklists and simulations.