Imagine you found a magic lamp. One of those that if you rub it a genie appears. He comes out and says he can answer one question. What question would you ask? Wait, before you answer make sure your question is sharp, tight and defined. For example, you might ask them, will the market go up? Maybe he answers yes. But will it be up tomorrow, next week, next month? We didn’t specify. Did the genie mean the stock market, the bond market, the commodity market, the supermarket? Again, we were not clear. But what if instead, we ask, “What will be the price of Microsoft (MSFT) three days from now at 3:45 PM EST?”. That would be some useful information that we could use and take advantage of.

The example above is a good example of a good final question, but that doesn’t mean that we should not brainstorm to come up with questions and not care too much how silly they might sound at the beginning. We can prune them later.

The word brainstorming was coined by Alex Faickney Osborn in his book Your Creative Power, published in 1948. Osborn was a very successful advertising executive and business owner during his time.

This is how Osborn explains how the name “Brainstorming” came about:
“It was in 1939 when I first organized such group-thinking in our company. The early participants dubbed our efforts ‘Brainstorm Sessions,’ and quite aptly so because, in this case, ‘brainstorm’ means using the brain to storm a creative problem and do so in commando fashion, with each stormer attacking the same objective.

Classical brainstorming is a group technique to create new ideas. In this case, we are slightly changing the definition and instead of thinking of ideas to solve a problem, we are using the brainstorming process to generate questions.

How to brainstorm for questions?

The leader of the group introduces the domain. Even though we are looking for creativity, it is important to set some boundaries, so if we are looking for questions in the finance industry a question related to the biotech industry might not be relevant. The domain can be as broad or as tight as necessary.

The domain is explained in a way that all group members understand its essence.

Facts and details about the domain are given before we start brainstorming.

The group meets and begins to generate questions. Everyone just speaks out their questions. All sample questions are okay, simple ideas, crazy ideas. At this stage, we want to generate as many questions as possible. The more initial questions, the better. No group member, including the leader, can criticize any idea. Everyone is encouraged to use other group members’ generated questions to come up with yet another question.

All questions are taken down by a designated person at a place where all group members can see the questions (a whiteboard or flip board, for example).

Four Basic Rules (as prescribed by Osborn)

No criticism is allowed during brainstorming. (Evaluation of ideas after the brainstorming)

Quantity is important. The more ideas the better. (Don’t be concerned about generating only “good” questions.)

Wildness is good. Crazy questions are welcome. (Many times, the craziest ideas turn out to be the best ones.)

Combining other questions and taking another person’s question a step further or using them for yet another question is good.

A good duration for a brainstorming session is around 30 to 60 minutes. After the meeting, the list of ideas is copied and distributed to all group members.

Ranking of Questions

Questions can be evaluated later. One possible way to evaluate the questions is for each individual team member to rank them (High – 3, Medium – 2, Low – 1) on their own and then consolidate the results and see which questions bubble to the top. You can have another session to determine which ideas make the cut and determine a final ranking and final list. At this point, it is okay to discard questions or consolidate questions.

Refinement of Questions

Once the list is finalized it is important to sharpen the questions and include additional details to the hypotheses (Remember the genie?). For example, if we are looking for fraud, our initial question might be “In a credit card statement, which transactions are fraudulent?”, but asking such question will result in identifying patterns of what up to now is considered fraud. We will likely find patterns of the current fraudulent behavior. But what about bad guys that find new ways to cheat the system? We won’t be looking for those patterns. A better question is “Which transactions in the statement have anomalies?”. We can then do some further processing to determine if these anomalous transactions are indeed fraudulent. It’s a subtle distinction, but an important one.

Sub Classification of Questions

Once the questions have been refined, it is now time to break down the questions into multiple sub-questions. As an example, the initial question might be “How do we quantify sentiment analysis in a dataset of call center recordings?”. Some resulting sub-questions might be:

  • How can recorded conversations be transcribed?
  • How can sentiment analysis be performed on written text?
  • How can individual results for each conversation be consolidated?
  • What is the best way to visualize the consolidated results?

This is a very simple example, but with more complex problems just successfully breaking down a question into sub-questions can be a breakthrough in its own right. As an example, think of the CFTR gene. It has been determined that a mutation in this gene causes cystic fibrosis. A complete cure has not yet been found but identifying which gene is responsible for the disease puts us a step closer to curing it.


Many times, we might come back from a conference where we learned how to use the latest shiny hammer and we come back looking for nails or worse yet, we think everything is a nail. And lately, it seems every week there is a better and bigger dataset being advertised. However, it is very important to realize that solving business problems and questions is ultimately what pays the bills. Don’t lead with the tools or the data. Lead with the business problem and the question if you want to optimize outcomes. That’s all for now.