The Big Picture Of Analytics-Driven Decision-Making
People are constantly making decisions in every organization. It ranges from small repetitive and near-term decisions to less frequent and strategic but very impactful decisions. In all cases, the best way to make wise decisions (or optimal decisions, whenever they are quantifiable) is to leverage all the available data as much as possible.
Then the question is: what does it take to leverage data to support decision-making?
Understanding the big picture
Leveraging data to support decision-making is a process that can be decomposed into three major steps, typically addressed with descriptive, predictive, and prescriptive analytics.
Step 1: Understanding the current behavior of the system
Traditionally, decision-makers are expected to be those people with a lot of experience because they have a comprehensive understanding of the business. When it comes to making analytics-driven decisions, the experience is replaced with data from which we derive (as much as possible) a comprehensive understanding.
For example, suppose that Mr. Mip wants to visit a friend who lives in the middle of a big urban center. If Mr. Mip is a delivery driver who has worked in this city for ten years, he can certainly drive to his friend’s place without any help. But if he is new to this city, he needs to start by collecting data to identify potential routes. Is it possible to get there by bus? How about the subway? Is ride-sharing an option? How about driving? When are the rush hours? Is it going to rain? Are there any traffic restrictions? What are the parking rules? So Mr. Mip needs to learn as much as possible about the transportation system of the new city.
Step 2: Forecasting the future behavior of the system
Because decisions only affect the behavior of a system in the future (the decision you make today doesn’t change what happened yesterday), it’s crucial to understand how the system evolves. This is the only way to plan ahead. In addition, the further in the future we can predict the behavior of the system, the better we can plan for it. This may sound completely obvious, but you would be impressed to know how even well-established organizations tend to overlook this fact.
For example, if Mr. Mip knew from the weather forecast that there is a high chance of heavy rain early in the evening, and knowing that the traffic tends to get jammed around that time, he may consider alternative times, alternative routes, or alternative transportation modes, before departing to his friend’s place.
Step 3: Analyzing all viable options to make the best decision
Making the best decision is a very tricky thing that deserves deeper consideration.
When we talk about the best, we implicitly assume that there is a way to compare any two alternatives and decide which one is the best. But this is not always possible.
For example, suppose that Mr. Mip has two options:
Go by bus, which takes 25 minutes and costs $8; or
Take a ride-share, which takes 8 minutes and costs $25.
Which one is best? The answer is: it depends. Mathematically, these two alternatives are not comparable unless we assign weights to the dimensions of cost and time. But, depending on the context, deciding on the weights can become a problem on its own.
In addition, even when all alternatives are comparable, there may be multiple solutions that are equally good. For example, if money and time are equally important for Mr. Mip, i.e., they get assigned the same weight, then taking a bus or a ride-share does not make any difference in the example above.
Also, there may be very subjective aspects of a decision that is hard to capture quantitatively. For example, even if money and time are equally important for Mr. Mip, he might have environmental concerns and think that the bus is a more sustainable alternative.
Finally, assuming we have a quantitative way to compare alternatives, as we typically have for practical industry problems, the number of options can be a big challenge. At one extreme, there may be none or very few alternatives due to a large number of complex requirements, which is typically a very hard problem to solve. At the other extreme, the number of alternatives can be astronomic, in which case it becomes hard to identify the good ones. In both cases, computers become very handy, and advanced analytics (or artificial intelligence) finds great applicability.
A new paradigm for decision-making
As we have seen, there is a lot that goes into analyzing all viable options to make the “best” decision for a given problem. But, fortunately, analytics is well-equipped with methodologies and technologies to handle these challenges. Still, we must accept that even advanced analytics solutions are almost never perfect. After all, the real world is full of ambiguities, and ambiguity doesn’t go well with perfection.
Before data analytics (data science, machine learning, operations research, etc.) became popular, organizations across many industries--such as transportation, consumer goods, manufacturing, healthcare, and entertainment--would account for data to make their strategic, tactical, and operational decisions in a very ad-hoc fashion. This means that organizations would rely solely on the cognitive intelligence of experienced people.
The landscape started to change drastically as science progressed, technology evolved, and large organizations began to realize tremendous value from using systematic approaches (including advanced algorithms) to derive optimal decisions from whatever data they get access to.
Google, Amazon, and others have prospered not by giving customers
information but by giving them shortcuts to decisions and actions.
Harvard Business Review, Analytics 3.0
Wiser decision-making
It should be clear from the previous sections that analytics itself is not always enough to drive wise decisions. In fact, analytics is mostly about artificial intelligence (to be clear, we’re talking about Practical AI, as discussed by Mike Watson in his blog). However, the ambiguity of the real world demands cognitive intelligence as well.
There are two main scenarios where cognitive intelligence can contribute to the decision-making process. The first is in the development of analytics solutions itself. For example, human intelligence is crucial in modeling, because humans can understand the business context, and exploit characteristics of challenging problems to design the solution. The second scenario is in the deployment/consumption of solutions. For example, when recommendations from software cannot be operationalized straight away. This could be because of subjective aspects of the problem that can’t be captured by the model. In such cases, human intervention (i.e., cognitive intelligence) is required.
In the ideal scenario, it goes as follows: Humans are in control of the big picture, both from the development and deployment sides. But, as analytics evolves and becomes superior in handling certain tasks, humans delegate those tasks to machines and use the extra time to tackle the next level of complexity that artificial intelligence can’t handle yet.
Consider the analogy with driving a manual-shift car versus an automatic-shift car. Much less thinking goes into driving an automatic-shift car because artificial intelligence takes care of tracking and shifting the speed. As a result, the driver can put more focus on the road. And if the car was completely autonomous, then the driver could even read a book about AI while commuting!
Conclusion
We started with the big picture and gradually dived into some of the key challenges around making analytics-driven decisions in the real world. We pointed out inefficiencies and advantages of using advanced analytics in practical settings, emphasizing the important roles that humans have to play, both from the development and consumer sides of an analytics solution.
To summarize, we’re convicted that digital transformation is not about replacing humans with machines, it’s about synchronizing cognitive and artificial intelligence wisely.
