It’s expensive to learn. But the costs of not knowing are even higher.
First, there is the cost of the mistakes you make when you don’t know what to do.
Second, there is the opportunity cost of the things that you could have done if you knew what to do.
When buying traffic (purchasing advertising to sell a product) the main cost is learning which traffic to buy and how much to pay for it. The buyer has to buy lots of traffic. Some of it pays off, some of it doesn’t. Once a buyer knows what to do she’s simply making a profit. She’s spending $1 to make $2. Not knowing, uncertainty, is bad. It’s expensive. It costs money to learn.
The best way to learn about complex, changing systems is machine learning. Humans aren’t very good at learning from complex systems. We are people of stories. Stories work well in cases of relative simplicity. Our brain’s capacity to gather and analyze facts is limited. Machines don’t really have this limitation.
Though machines ability to analyze facts is basically unlimited they do face an important challenge. Machine learning depends on getting good, clean data. You can’t learn from chaos. Neither can a machine.
It’s said that 90% of statistics in the real world is getting good, clean data. Put simply, it’s the work of getting facts. It’s hard to do. It’s expensive.
Let’s say you want to learn to predict the weather. You can’t do it by asking people their impressions of the weather. You will hear personal experiences like “Not bad,” “A good day for a picnic,” “It could have been better.”
What you need are facts. You need data about the temperature, wind speed, barometric pressure, precipitation, etc. That data comes from expensive equipment that must be installed, monitored, maintained and updated.
By analyzing this data machines can learn what happened in the past. They use that data to predict the weather with some accuracy. Without it we would be stuck with what humans can understand about weather. We would be limited to big trends (colder in winter than in summer) that we can build stories around. But those stories don’t tell you whether or not you should take an umbrella with you when you leave the house in the morning.
There are downsides to not gathering and analyzing this data.
It’s annoying to carry an umbrella on days when it is sunny. Getting soaked in the rain is also a terrible experience.
Buying traffic at a loss sucks. But not buying profitable traffic is even worse.
How do we avoid these two unpleasant outcomes of ignorance? By paying the high cost of education: getting the facts and using machine learning to make better predictions.