How to Write An Impactful Insight
Lessons learned from Bukalapak Data team
Have you ever worked on very interesting data analysis, but nobody got the insights? Whereas Data Scientists are great at performing data analysis, they are also required to discover the insight in their analysis, so they can influence the decision making process and create a data-driven ecosystem.
Discovering an insight is like solving an escape room. You have to solve the riddles, puzzles, and clues in order to find your way out before the time runs out. When you discover an insight, you have to carry out exploration to understand the problem, to look for the clues, and to find out the solution before the deadline comes. In order to do that, we have to understand the evolution of data before the insight can be discovered.
Data Evolution
Theoretically, insight is one of the forms of data that have evolved through several steps. Some of you may already familiar with this analogy:
Let me elaborate on this analogy quickly:
Data: A record of events. Raw, unorganized, unprocessed, and mostly useless.
Example: You’re a baker and own a bakery.
- January: 500.
- February: 700.
We don’t know what that means. It can be the number of bread sold, number of stale bread, or maybe total sales in a month.
Information: Data with a meaning.
Example:
- January: 500 pieces of bread sold.
- February: 700 pieces of bread sold.
Well, now this data has a meaning for you. But is it enough to make a decision?
Knowledge: Detecting patterns from the information.
Example:
- January: 500 pieces of bread sold when we only open on weekdays.
- February: 700 pieces of bread sold when we also open on Saturday.
Insight: Detecting causation from knowledge, that is typically not explicitly reflected.
Example:
- After we compared the number of items sold per day, we achieved sales on Saturday twice as much as on weekdays.
Wisdom: Action that we should take.
Example:
- Based on the insight, we should also open our store on Sunday since it will likely generate more sales.
From that analogy, we can see that wisdom can only be derived from insight, and to get an insight, at first Data Scientists must get the fundamental information from the data such as the data definition, and what that data means. Next, they should have to explore the data more further and find a pattern. By doing this step, we expect that Data Scientists will get a deeper understanding. Then, to get an insight, Data Scientists should be able to find the root cause of the problem. An insight should be able to answer “Why did it happen?”. Finally, to get wisdom, Data Scientists need to identify what action should be taken regarding the findings and insights that have been developed.
Crafting insight in the industry
Before we explore insight’s definition in industry, we should understand first why we need it in the first place. A great insight is fundamental to implement Data-Driven Decision Making (DDDM). DDDM is the process of making organizational decisions based on actual data rather than intuition or observation alone.
Making decisions based on intuition is like gambling. You don’t have any idea of the expected outcome. Hence, you will be glad if it works, and be displeased when it doesn’t; potentially you would have a huge loss as well. A great insight should be able to minimize that risk. So instead of gambling on your decision, you find out what’s the best move you should make by calculating the potential impact on each option.
In the industry, typically an insight has to be well documented, in a deck, docs, etc., explaining what problem you want to solve, how you solve it, and what the recommendations are. A great insight has a strong call-to-action which influences the decision-makers. Here are things you should have in your insight to create a great one.
Problem Statement
All of the great research should start from a solid problem. The problem you need to solve should’ve been aligned with the stakeholders' objective (usually called OKR). This is the part where you discuss back and forth with them to ensure that you solve the important problem. Stakeholders may not really understand what’s the best approach to solve their problem. That’s why a Data Scientist has the responsibility to understand the whole problem’s context before determining the analysis approach.
Explain the Analysis
In explaining the analysis, you may need to pay attention to frame the analysis for the specific audience. Commonly, there are two types of audiences with different objectives. The first objective is to explain your analysis to the stakeholders, where we expect them to understand the recommendation and the reason behind it. The challenge here is to make sure that you speak the same language as them. So you may exclude all technical terminologies since it is out of their fields and might make them lose their focus.
Questions you might want ask to yourself when including some part of your analysis are:
- Is it simple enough?
- Does your audience understand the terms you put there?
- If this part of the analysis is essential for them to understand?
- Are they still gonna understand your analysis if you take it away?
For example, if you create a fraud detection system, you may not need to explain what algorithm you are using, accuracy comparison when you are comparing several algorithms, and so on. Instead, you may want to focus on the algorithm that you choose, and translate it to a high-level definition. For example, you may want to explain what features are you using on your model, how the model performs, how the false positive rate is, false-negative rate is, etc.
On the other hand, if the objective is to present your research at a conference attended by fellow Data Scientists or to get some review from your peers, you may need to include more comprehensive details on your analysis so they can get a deeper understanding.
Recommendation
Imagine you are writing a story. This part is like the climax of your story. After you explain your analysis and the findings, what is the call-to-action? What do you expect your audience to do? A great insight should make an impact, and one way to do it is by giving strong recommendations. Those recommendations should be aligned with your findings in the previous section, so your audience understands why you would recommend those. This part is important so your analysis won’t only end up on their browser history and disappear after the next several months.
For example, if your analysis is about creating a fraud detection system, then the recommendation is to implement your model and blacklist those users who get caught. Or, if you create user profiling, your recommendation could be giving certain treatment or allocating marketing budget more efficiently based on user interests.
Impact Calculation
Just like what our chief likes to say, “what does the sky look like?”. This is important since your stakeholders need to be convinced by your recommendation. If they’re gonna implement those, what is the impact on the business? For example, if we implement the fraud detection model, we can save our promotional cost up to IDR 10 billion/month. Or, if we personalize our marketing using user profiling, we can get a higher CVR, number of transactions, and also revenue by 20% per month, while also reducing our marketing budget by 40% per month.
Wrap Up
Now that you know fundamental aspects you should write on your insight, it is ready to be presented and hopefully can influence your stakeholders to make a strategic decision. Let’s create more actionable insights!
