In the world of wearable devices, user experience is dependent on metrics and insights they get from using such devices. The accuracy of those metrics is driven by the processes and methods in the fields of data science and data engineering. Our own data detective, Dario Bošnjak had the pleasure of working with the producer of women’s smart jewelry, Bellabeat. In cooperation with them, he worked on optimizing metrics that evolve the same way users evolve.
So, what is important when you deal with users in the industry of wearable devices? And why should companies like that invest their time and resources into developing correct and actionable metrics that follow and enhance the lifestyle of users?
Why do you think that data plays a major role in wearable and smart devices?
The purpose of wearable devices is to display to the user his current psychophysical state and to make recommendations for his lifestyle. This way, smart devices can motivate the user to change his or her life habits through metrics and scores in comparison with other users.
The devices create an image of the user by collecting data through sensors, for example, heart rate, breathing, movement speed, etc. Through certain metrics, they can show user’s daily life and movements. In the end, the whole user experience and satisfaction depend on the data collected, how it was processed, and presented to the user.
What are the major issues when dealing with collected data from such devices?
All companies in the smart devices industry face the same challenges. While working in the industry, I identified three major problems linked to data – accuracy, completeness, and interpretation of data.
Accuracy
Even with the advances in technology, you can not expect 100% accurate data the whole time. For example, if you are doing some physical activities, you move, so there is possible noise in the data because the device moves as well. To fix that issue you have to use various algorithms and methods to clean or correct data and calculate metrics based on that accurate data. As the noise in data may happen in completely different situations, e.g. during sleep or physical activity, you have to think about every possible scenario and be creative in dealing with such issues.
Completeness
You can not expect users to wear the devices the whole time. Sometimes they forget to charge them, so the devices turn off during sleep or some activities, or they just forget to wear them. In those cases, you’ll face missing data.
Interpretation
When you do calculations and algorithms based on data, you have to take into consideration that we as humans are different. Those differences manifest themselves through attributes like sex, age, mass, physical activities, etc. All these parameters have to be taken into account when calculating metrics from data. Also, even if some people are similar in those parameters, they can have different lifestyles, e.g. somebody needs less sleep while others need more.
Situations, where users use the devices, are different and you can never rely on just one type of data. If you want to define calorie burning, you can’t rely just on their movement speed because it’s possible they aren’t doing activities that imply quick motions (gym, uphill climb…).
Do you believe that data science and data engineering bring users a bigger end value?
Absolutely! Without those two disciplines, users wouldn’t get valuable information, or they would get incorrect ones.
Data engineering invites the possibility of collecting and combining all the data and preparing it for more detailed analytics and aggregation. Whereas data science resolves the issues from the previous question.
Can you draw from experience, what were the results of using data science and data engineering to enrich metrics and insights from wearable devices?
Using the principles of data science, we upgraded existing metrics that were shown to users. We managed to accomplish that with various experiments over data collected from test users.
Data engineering principles can always be used for data filtering, combining and aggregating for multiple users, reporting, and so on.
What were some major limitations when dealing with data from those devices?
The devices are wearable and such electronics work on a different level than others. Because of the specificity of the compact design and battery, the sampling of data is on a lower frequency. It is resolved by adjusting the sampling frequency in specific situations. We wouldn’t call them limitations, but rather the uniqueness of smart wearable devices.
From a business perspective, do you believe that your work brought certain advantages to the company?
We certainly improved existing algorithms in data correction and analysis and metrics calculation. Also, some new metrics and algorithms for data correcting and analysis were introduced. As the lifestyles of the users change, you have to be up to date with those and make sure that metrics evolve as the users evolve.
What were some methods or practices you used to define new metrics or to improve existing ones?
We used standard methods for data cleaning and recognizing noise in data, but in our work, there was never any emphasis on a certain individual method, but rather on its applicability. As we said before, users have different ways of device usage in different situations. It is imperative that you adjust the methods so they fit best in all the cases.
We defined metrics according to the data from test users and with different experiments. With that, we switched methods, adjusted parameters, and things like that. You never stick to the one thing that you tried first. No, you adjust as you go, so you can find the best method or practice to deliver the best results.
Why are data science and data engineering becoming an integral part, not only in understanding data but also in creating software solutions?
In today’s world, you can not disregard the importance of data. You now have businesses that are solely built on the usage of data. Some collect data, some generate it and some create solutions based on user-generated data. Nowadays, having a software solution that is not user-focused is a miss. If consumers see that your solution is not oriented toward their needs and wants, they disregard it. You have to deliver something of value that takes information about the users into account. And, you cannot properly utilize data without data science and data engineering. These disciplines are the starting point in understanding data, its value, and its potential. As time and technology progress, so will the need for data science and data engineering. With more and more data, companies will have to rely on someone to make sense of all the information. It has already become a way of discovering new trends and opportunities or recognizing upcoming disruptions in the market.
