Comments on the Rise of the Strategy Machines

In “The Rise of the Strategy Machines”, Tom Davenport writes about how computers will end up running your company. He links to a more in-depth article from BCG authors about the building the Chief Strategy Robot. My management consulting friends were batting around some questions about what to call this trend, and where it is going. As usual, I had some opinions …

— — — — -

This is good stuff. I like the list of places where algorithms are making decisions. It is definitely a management megatrend.

I suggest the term “Cognitive layer”. Modern systems are constructed from a bunch of Web services that exchange, emit, and collect data . On top of this, we will have systems that analyze the data and tell us if something is wrong or provide other advice and decision making. They already look for security threats and do a lot of problem reporting. They will eventually also make strategic recommendations. I like the term “cognitive layer” for this.

I’m working on a service that I call that will look at all of the financial and operating data in a business, and hopefully make increasingly useful recommendations. This has caused me to think about some important principles:

1) Humans can make good decisions with small amounts of data. For example, they can drive a car with input from two eyes and an ear. They can also run a business. However, when that business gets large, they miss a lot of things. There is a strict limit to the amount of data that a human can process. Computers need a lot more data to start. To drive a car, they need input from at least 10 different sensors, including expensive LIDAR rangefinders. But, with enough data, computers drive better than humans. The more data they have, the better they get.

2) This means companies can get bigger. Computerized recordkeeping allowed companies to get as big as they are now. The process continues. Computerized testing and project management is allowing software systems to grow far beyond the size and complexity that, ten years ago, would have triggered the “mythical man month” problems where it becomes very difficult to test and integrate large systems. Google does not have this problem, because they have a test system that runs 100M tests per day and points out the things that need to be fixed and integrated. I believe that computerized AI-type management will allow some companies to get much bigger before they succumb to the “errors of management” that Coase described — the misuse of resources that hobbles big companies. I call these the Apex Competitors. People aren’t prepared for them.

I liked the list of places where human strategic decisions don’t work out. I wonder how much of this is because of bad decision making, and how much of this is because of the inherently random behavior of a chaotic system. I bet there is a way to separate the two effects (eg 83% chance of failure is 23% wrong choices and 60% randomness). This is important if you care about innovation. We want to remove all of the wrong choices. However, we often want to INCREASE the rate of random failure. That just means we are making bigger bets and innovating more. Failed experiments are necessary for innovation. Bigger innovations have a higher failure rate, but they also can have bigger payoffs (option value). A decision machine will optimize two factors — correctness, and option value.

SaaS entrepreneur/engineer. Founder of MAXOS, Real World DeFi. Previously founded Assembla, PowerSteering Software, on team at SNL Financial.

SaaS entrepreneur/engineer. Founder of MAXOS, Real World DeFi. Previously founded Assembla, PowerSteering Software, on team at SNL Financial.