The world’s leading publication for data science, AI, and ML professionals.

With Big Data Comes the Importance of Knowing the Right Context

Understanding the business objectives in 3 key questions before developing the algorithms

Photo by Brett Jordan on Unsplash
Photo by Brett Jordan on Unsplash

Context, context, context. Context is everything when working in Data Science. The biggest mistake I see data scientists making is not understanding their work’s context and business objectives. Instead, they focus heavily on knowing the AI/ML techniques. Complex algorithms are great when they are used in the right place. Understanding when and where that place should be is hard.

Start with understanding your business objectives. Starting with my objectives is something I have been focusing on more closely the past two months. As I am working on a problem, I don’t want to jump right into developing solutions. Instead, I need to work with SME’s to understand the context of the situation and how they look at the data. Since starting as a consultant, my strategy for understanding my business objectives has changed. Let’s walk through some questions and what to consider when answering them.


To set the stage, let’s pick a slimmed-down version of a Kaggle competition problem and pretend this is our given business objective:

Forecast the total amount of products sold in every shop.


What does this objective ask for?

Looking at your objective, make sure you understand the question that is being asked by the business. In this case, the data is provided by one of the largest Russian software firms, 1C Company. One way to learn more about your objective is to have a presentation with your key stakeholders and walk through the objective and desired output. Discuss what you know and where you need clarifications or gaps filled.

This objective may seem simple at first glance—you need to predict the total amount of products sold in every shop—but digging deeper reveals more questions.


What data do you need to answer this question?

After initial data analysis, you can understand what you do and do not have in your dataset. With this information, you should discuss with your stakeholders what is possible with their current data. As well, you can discuss what is not. If your objectives are not achievable because of missing data, this is when you need to make that known.

For our example, let’s look at the data itself. Even though you are provided with daily historical sales data, the data can change month by month. The list of shops and products fluctuates, meaning it will be more challenging to create a model that can handle the given data. It is also important to note that we do have unique identifiers in this data. We are given shop_id, item_id, and item_category_id, giving you a unique shop, product, and item category.

Another thing to note here is the text columns. The given objective doesn’t ask about those fields. Therefore, they are not needed for this analysis. Important information includes the number of products sold, the current price of each item, and dates. All of which are available in the dataset. So far, it appears we understand the primary objective and have the available data to achieve it, but what will the end-user be doing with this information?


What is the user looking to do or achieve with this data? And can your data accomplish that?

Your user may be a business analyst, SME, or someone else. As you develop a methodology for your objective, you need to determine how they will utilize the output. Sit down with the users and have a conversation with them. What is it that they want to achieve with this data once the analysis has been done? When I have these conversations, I have found that sometimes the original objective did not cover what was actually desired. Meaning I may have to go back and determine if any additional data is needed to get the desired output.

We know our sample objective is to predict the total amount of products sold in every shop, but how will this be used? Assuming we are having this conversation with the stakeholders, here are some things that may come up:

  • Knowing how much of an item a specific shop will order can help determine how many items need to be ready for shipment next month. This output we can determine with the information that has been provided.
  • Knowing how much different items are purchased shows what items are high profit vs. low profit. Knowing this, the business can determine what to stop selling. If you went with this statement, you might need more data to understand each item’s profit margin.
  • Want a dashboard created to see trends in the items being sold based on the day of the week. Are specific items more likely to sell on different days than others? If you dig deeper into your original objective, you may find different objectives emerging from it. If this is the case, you will need to reframe your work and make sure you have the data. Our data is given by the day in this dataset, so this type of analysis may be possible to analyze trends of items sold based on the day of the week.

We can keep going, but I think you get the point. As you start to analyze your business objective, you may find the initial objective morphs or grows. This is perfectly normal. I have seen that these conversations can go on for long periods if you let them. I like to time box such discussions to an hour at most to start. With the information learned in that meeting, I can then go and revise my approach and seek feedback based on the updates. Through this work, both your and the stakeholders will better understand your data, your revised approach, and what is possible.


In Summary

Business objectives may seem very straightforward in the beginning until they aren’t. That is why I have started asking myself three questions before I jump to writing code and building models. These questions help frame the problem, understand if the data is there to solve it and figure out the output. Often, you may find out that the user’s desired outcome may not always be what they said initially. Instead, their vision morphs as you start having deeper discussions on what they are looking for and what the data can tell them. These questions are:

  1. What does this objective ask for?
  2. What data do you need to answer this question?
  3. What is the end-user looking to do or achieve with this data? And can your data accomplish that?

Do you have any other questions you like to ask when approaching a business objective? If so, what are they, and how have they helped you frame your solution?


If you would like to read more, check out some of my other articles below!

A Quick and Easy Morning Routine to Jumpstart a Productive Week

Top 3 Reasons Public Speaking Can Help You in Data Science

Top 3 Lessons Learned from Leaving a Job I Loved


Related Articles