I have been thinking about the economics of Decision Making for a while now. And also, currently reading Prediction Machines – very interesting read.
This takes extra relevance as the economics of AI come into focus as enterprises continue to invest in AI, in search of the next engines of productivity.
Let’s start with Predictions – and here, we will take the definition from the book: Prediction is filling up the missing pieces of information. Which is broader than forecasting a future event.
To begin with, here are a few concepts:
- Prediction is an economic good – i.e. it has a cost and creates value (Utility).
- Prediction has two main features of interest:
- Expected Value (the outcome we would expect) – e.g. Sales Forecast, likelihood of a customer churn
- Uncertainty/Variance (the risk of the Prediction not turning out to be right) – e.g. Forecast error, Type-I/II errors in a customer churn
- Predictions enter into the decision process in 2 stages:
- At creation stage, when the Marginal Cost is incurred and where the Expected Value and Uncertainty express the Prediction
- At consumption stage – where it creates Marginal Utility.
- Like any economic good, a Prediction creates value so long as the Marginal Utility exceeds the Marginal Cost
All this is fine, you may say, but what does it mean? Here is a simplistic example: You can make a very low cost prediction, but if it is inaccurate, it is useless. Like any economic good. E.g. Planning creates a Statistical Sales Forecast. Procurement uses that to drive purchasing decisions based on this. But if the accuracy is low, and ends up driving higher safety stocks, the marginal cost may outweigh any potential marginal utility. And may not justify the creation of the Forecast in the first place.
The important thing is that once we define Prediction as an Economic good, we have a framework to think about comparing agents who make predictions. With AI, we have two distinct agents:
- Humans: intuitive; error-prone, with biases
- Machines: wide spectrum of Algorithms, enabled by data.
With AI, it is clear that Machines are becoming better at predictions than Humans. In pure economic terms, the cost of Machines making predictions relative to Humans is coming down faster than ever. And that is opening up a whole new world of possibilities.
Now, let’s look at the true economic engine of the modern Enterprise – Decisions. It is Decisions – big and small – that moves an organization forward.
Decisions are extremely complex, but if there is one thing that Economics teaches us, it is the power of abstraction to cut through the complexity. With that license, one way to think of this:
Decision = Prediction + Judgment
Prediction: a recommendation (or a set of recommendations) for an action
Judgment: Evaluate options, make choices and arrive at a decision. For all that machines have done and will continue to do, judgment will continue to be human. Remember that this is in the context of an Enterprise – where a human must eventually make a decision by picking an option from a set of recommendations.
Predictions and Judgment are Complementary Goods – a quick recap from College Microeconomics:
- If the price of butter becomes cheaper, your propensity to consume Bread will go up. The demand for bread will go up. More innovation will happen – you will have better bread etc. Virtuous cycle continues as the price of butter keeps on falling.
- Economic growth happens (moving the Production Frontier) when there is continuous innovation in bringing the price of butter down AND making better bread.
And this is what we need to see from AI in Decision making:
- As Machines and AI bring the cost of predictions down, the propensity to consume judgments (i.e. the number and frequency) will continue to go up. Which will in turn, reduce the decision cycle time
- As the cost of Prediction continues to reduce, more and better judgments will be made – pushing the overall decision space.
And perhaps, the most exciting part is the Value of judgments will keep on going up – in other words, far from being made redundant, humans will become even more valuable in the overall decision making process. I, for one, don’t believe that AI and machines will ever take over the human race!
With all this, how should a company think of enablers like AI? This obviously opens up a whole world of opportunities. To begin with, every organization would do well to do the following:
Predictions: Build Capabilities to reduce the cost of predictions AND improve accuracy:
- Data as an Asset: Capture, organize, learn and improve your data assets
- Algorithms as Services: Democratize Algorithms, integrate them into processes
- Process Automation and Efficiencies: Free up resources to focus on experimentation
Judgments: Help the Managers get better at decisions:
- Get the Prediction results faster into the decision making process: Invest in right tools as Decision boards; Anomaly Detection, Scenario Builders
- Feedback systems to continuously improve the Predictions