The majority of my programming job involves working with data. While CRUD (Create Retrieve Update Delete) can be enough for many processes, sometimes the data you get isn’t the data that your users want. The biggest obstacle facing data scientists at all levels is how to get data and make it useful. What you get isn’t always what you need, and companies want data they can actually act on.
Data needs to be actionable to be useful outside of an academic sense. There’s a difference between knowing a work contains 253 of a specific character, and knowing that this means it’s the most common word and it’s used for 4.77% of the text. Most people can draw conclusions from basic data, but what happens when the complexity outgrows what an an average person can keep track of or make sense of on the fly?
Basic context can tell you some things, but is that 4.77% always the same or is there more to consider? The more you dig, the deeper you can go. Data is like poetry, the same words in the same order can mean something wholly different depending on the context. You need to understand the context or risk missing the whole point.
Information has always been a currency which can buy advantages when used right, but it can also hinder every effort when interpreted incorrectly. What do the numbers mean and how can you act on them? Actionable data is simply data which has been processed to where a decision can be made to act on it. You aren’t just presenting the raw data, you’re interpreting it algorithmically to mean something to whomever has to use it. For this to work, you need to understand the process.
Contextualizing the Process to Make Data Matter
How data is shaped determines its value. The same block of stone can blossom into a work of art or be crushed into gravel. To shape data in a way which makes sense, you need to understand how the data is being used. What is your data measuring and what does it mean to the process? It’s hard to process data when you don’t even know what it means in the process.
If you know customers are happy 94% of the time in one department, is that better or worse than if they’re happy 96% of the time in another? While it would seem obvious 96% is better than 94%, there’s more that goes into the process than that. What does each department do? How happy are customers typically? What if they were both at 95% previously?
The point of finding actionable data is to uncover these variables which are often overlooked and account for them. It’s hard to dig too deep when a person is doing the work, but it’s easy with current advances in computing technology. Inventory used to be counted by hand, now it’s scanned. A scalable process is going to require technology at most steps which means the chance to collect data and measure it.
The first step to building actionable data is understanding that there is more than numbers going into your final data without warping them. While you can work out the trajectory of a ball looking at timed snapshots, it’s easier to use basic physics. Once you begin looking at the process, you begin to understand the factors and variables which go into the system which leads to you understanding the data. This knowledge allows you to compensate for fluctuations that can lead to the wrong choices (via normalization and similar).
Understanding How Data Is Used and Acted On
The key to actionable data is being able to act on it, but why is it needed in order to act? Part of the answer shows up when you understand the process behind generating the data, but what is the data actually being used for? Are we tracking customer satisfaction between departments for layoffs or to try and streamline the process? The devil’s in the details.
The worst metrics I’ve built have come from when I didn’t understand why the data was being requested. I understood the process of how the data was generated, but I didn’t know why the teams wanted the data. Once you begin to shape the data, you are changing it. Bits are discarded and bits are modified. The goal is to make it better without tainting it.
You’re playing the same scale for your melody regardless, but different notes will be highlighted more or less than others, or even skipped. The underlying scale stays the same for the motif, but how it sounds varies. A single concept blossoms into endless interpretations.
You are the conductor to the orchestra of data and your understanding of the audience makes the difference between disinterest and satisfaction. It’s easy to understand the players for the data, but how will it sound to the audience? You need to read the room and adapt to it.
Understand why your data is being gathered to make the process easier. Most of my data gathering processes are extremely heavy on the collection process, but the actual presentation is near instantaneous. I understand the process and why it’s needed, so I can offload the boring parts to when my users don’t care. If data is continuously updated, you may not have this option and have to weigh what matters most and what doesn’t.
Optimizing By Use Case
The data I need uses extremely complex processing to paint a mundane picture, but it has to be interpreted to mean anything. The raw data is near useless, but it presents a clear picture when interpreted and collated. I could reinvent the wheel each time the data is read, but that would be both slow and inefficient since my users care about one number from the whole set.
Understanding how and why the data is needed let’s you know what to do, but how it’s accessed and used affects how you actually do it. Complex data tends to have complex calculations, and where these calculations lie can mean the difference between agile insight and frustration. Sometimes you just need the data once, other times the data is just a base for further processing. You have to understand the data, the process, what and how it’s being used, but also how that impacts the data gathering process.
One of my “simple” collection processes takes thousands and thousand of lines of code because access is the most important part of the process. I could make things easier using simpler methods, but they push the processing on the presentation layer and it just doesn’t make sense for the use case. I work with data which is write once, read often. If the data is written in the middle of the night, why would I worry about adding complexity to the process as long as it’s ready when people actually want it?
Consider this factor when working with actionable data. Your data may feed into a process, but ultimately something is being distilled for people. What is it, why does it matter, what is it used for, and how is it consumed?
Further Considerations
This article won’t make you a data scientist, but it gives you the tools to make sense of what data means and how to make it actionable. Data begins as a pure snapshot of the situation (when measured correctly), and you give it context. The same data can be painted differently to get different conclusions. Every statistics class begins with an explanation of how “9 out of 10 dentists” doesn’t actually mean 90% of dentists agree.
You can mislead with data, or you can enable actionable choices with it. If you focus on the wrong factors for the wrong people, you present something which is at best useless (or, more often than not, harmful). If you lose sight of the purpose of your work, you run the risk of working against your intentions.
Data is pure, but it’s interpretation is not. The problem is most data is near meaningless without being shaped due to subjective factors. Just look at reviews on Amazon (affiliate link), what exactly is a 5 star review worth?
Whenever you shape data (in non-trivial ways), you are making an interpretation. The more you feel you aren’t affecting the process, the higher the risk that you are. Every choice, every priority, and every decision changes the data or how others interpret it.
Once data gets past a certain level of complexity, it gets almost impossible to not need to process it. As you process data, you need to understand what you’re working with, what it means, and why it’s important so that you can turn the data into actionable data. Sheet music is all the same, but the conductor works with the performers to make it have an impact. Data is exactly the same.
Image by Matthias Böckel from Pixabay