Fall in love with the problem, not the solution!
Authors : Eduardo Quadros | Alex Souza
This is a story about a project that aimed to use machine learning to calculate the profitability index by channel automatically and dynamically.
On the project team as a whole, there wasn’t a data scientist — we had engineers and a data analyst. In the role of data analyst on the project, I thought it would be great to have a new member data scientist, as this would yield a lot of learning about how this job works.
At the beginning of the project, we had the phase of understanding the customer’s needs, where the entire team, including the data scientist, participated, gathering as much information as possible with the customer, their needs, expectations, what they already had at the time, etc. After this moment, we went to the phase of collecting the necessary data, and this was under my responsibility as an analyst who already understood how the customer’s data was arranged and worked.
Then, I met with the data scientist to talk about the problem we had — a great opportunity to learn from someone who is a reference, and the best: in a project in practice. We talked about the problem, the customer’s need, and where to find the data. His proposal was to bring a formula that calculates the traditional profitability index. I gathered all the information based on his explanation and developed a view (via SQL) that performed the informed calculation and put it in Power BI to pass it on to the customer to validate in a more visual and dynamic way as they had requested. We took this solution as our project baseline (in parentheses here: in the baseline, the data is already qualified and can be used in future cycles for predictions, prescriptions, etc.)
We sent version 1 to the customer and a few hours later we already received feedback that the calculation was correct, but they needed a better way to visualize the data. They asked for the addition of new fields that would allow filtering information by periods, sales channels, vendors, stores and specific products. We treat the view to bring the requested information, change the view and finally generate a new version (version 2) of the view and send it to the client. Here comes the “build-measure-learn” feedback loop , proposed by Eric Ries in the Lean Startup book, or in other words, do fast and fail fast!
During our project, we generated version 1 based on a solution hypothesis, and we sent it to the client ( build ), the client sent us feedback with visual improvements and filters ( measure-learn ), we made the necessary adjustments and sent the new version ( build ) to the client shortly thereafter.
So far, in this project we had 2 cycles, but these cycles can be constant, it varies a lot depending on the “problem” (product), the customer’s maturity in understanding the problem and how the solution can be evolved .
“Here is a sentence and a reflection: Fall in love with the problem and not the solution! Remember KODAK? The company thought of the solution to record important moments and memories. Do you know Uber? They think about the problem of urban mobility. And so on.”
In short, we present the customer with the solution. They validated the calculations and they were as needed, they had some requests for improvements, but only in terms of layout. In the meantime, I thought about where and when we are going to implement machine learning , and what algorithm we are going to use.
As described at the beginning of the article, I had a great desire to know more about what a data scientist’s job is like. I imagined learning to apply an algorithm and seeing a prediction at the end, but what I learned was that we should fall in love with the problem and not the solution ! We don’t always need to apply advanced statistical algorithms, techniques and resources, when a simple exploratory data analysis, a simple data analysis, or as in this case, a formula, perfectly meets the need. However, with the evolution of the solution, there will be a need to apply machine learningand so we will learn and practice these algorithms so that we can measure their effectiveness and my learning through the baseline that was created. The evolution came from the analyst together with the solution, a win/win game.
This is one of the skills of a data scientist, knowing how to distinguish whether a given solution needs to be using advanced analysis, algorithms, or just data analysis that already meets the demand. As mentioned, based on feedback loops , the solution tends to evolve to generate even more value to the business and at some point it may be necessary to use machine learning algorithms, for example (that is, machine learning appears naturally , organically).
The project also provided an understanding that the routine built in recent years has a theoretical and consolidated basis ( build-measure-learn feedback loop ). To learn more, I recommend the book: A Lean Startup, by Eric Ries .
Finally, here are a few more important points. Firstly, in many works and projects that we deliver, we receive excellent feedback and praise, but if they (the projects) are not used on a daily basis, it becomes a vanity metric. So, always monitor the solution and evolve as the customer needs. Another point is that in data, the measure is the use of information by the user. Compared to a social media like, the more information is used, the more value it generates for the organization. And last but not least, the tip is not to go around applying machine learning algorithmsfor any problem, but to understand and find the most adherent solution and evolve, whether it be with the use of machine leaning or with rules / formulas. And of course, always observing the adherence of the solution to the business need.
Sources: The Lean Startup (Eric Ries)
Thanks for reading, hope it helped!