Business value comes when you convert business problems to data problems
One of the most difficult aspects of providing business-relevant analytics is identifying the business problem, quantifying it and building a concrete action plan on how to solve it utilizing data and analytics. (And several companies across industries often fail at it.)
In addition to data science skills, business problem qualification and quantification puts the pressure on the business line or product owner. One great tool to get started is to apply our Innolab-approach on this.
In recent years, we have gotten extremely good at solving prediction problems. We successfully offer our customers products that we expect them to buy, classify emails as junk and predict which move leads to victory in the board game Go.
We have also developed sophisticated techniques for cognitive computing: We can predict what a given picture depicts, what is the correct translation of a sentence and what is the best answer to the customer’s question.
What all these seemingly unrelated solutions have in common is that they are all predictions. And more specifically, they are all solved with supervised machine learning.
Find the business problems and give them to the algorithm - solved?
Another thing we are good at is finding business problems. Customers are abandoning their carts in e-commerce, our deliveries are running late, our machines are breaking too often or customers are switching operators.
However, none of these problems can be directly fed into a machine learning algorithm and solved. The power of machine learning can’t be harnessed until business problems are turned into data problems.
Some of the most innovative products that are labelled as artificial intelligence have their roots in an innovative way of turning business problems into prediction problems.
If you had told people 40 years ago that building a self-driving car is a prediction problem they would have told you that you’re crazy. In a similar way, automated customer service can be turned into a prediction problem: What is the answer that most likely solves the customer’s problem or asks for the required information.
Even though the above realizations are turning into revolutionary concepts, many simple issues that can help in everyday business are much easier to come up with. Automated classification of incoming documents or predicting the right expert to handle the problem are examples simple solutions that can be easily applied in many organizations.
In a similar manner, we can utilize machine learning to provide more accurate and new services to our clients.
The obvious choice doesn’t always solve the problem
Even though machine learning is not the only tool for solving data problems, the first step in becoming good at turning business problems into data problems is understanding what kind of problems are solvable with it.
In supervised machine learning, we practically show an algorithm lots of examples from the data, and let it find the rules in order to make predictions with fresh data. In the second step, the business problem is turned into a prediction problem. In some occasions this is fairly straight forward, but in many cases you need to be creative.
The most dangerous cases are those that seem straight forward but the obvious choice doesn’t really solve the business problem.
Beyond machine learning
However, there are many data problems that require different approaches. Sometimes traditional statistics is better suited for the task and sometimes we might need more complex optimization solutions to solve the problem.
Still, machine learning is a great topic to start with. It’s gaining a lot traction and for a good reason. It’s often easily productizable, it learns over time and can solve a variety of tasks and help in multiple others if used creatively.
I would argue that turning business problems into data problems is going to become one of the most important skills for decision makers, if it isn’t already. As technology takes a bigger and ever increasing role in business operations that have been previously conducted by people, two types of skills will become even more important: People skills and skills complementary to machines.
Technology enables us to do amazing things, if we only understand what we can do with it.