Sooner or later every company that strives to be data-driven stumbles across a problem of having more data than can be analyzed without a specialized position devoted to this task. Here is how we tackled this challenge at Base.
For the past decade there has been a growing trend in software development towards focusing on and improving user experience. It’s general goal is to make software as easy and friendly to use as possible. Interestingly, this attitude is not typical for business software market, currently dominated by products that pioneered cloud solutions without much focus on end users.
At Base we decided to change this and use the data that is generated by tens of thousands of users to constantly improve our product. We adopted data democratization: everyone has access to the important data sources and can use them to improve their work. This way the value from data is leveraged on a company-wide scale to achieve a single goal – creating the easiest to use, and the most useful CRM software out there.
This is where we see one of our major strengths, and that’s why we have been heavily investing in data analytics for a long time. Once the big data infrastructure was in place, we realized that our data could generate even more value if it was analyzed more thoroughly and systematically.
It is a very typical challenge faced by IT companies once they hit certain threshold of customers and gather enough data. There are two widely-practiced solutions to it: expanding responsibilities of data science team or splitting the responsibilities into two independent teams (data science and business intelligence). We decided to go with the third one – a mixed approach. I’ll describe the three approaches in more details below.
Approach 1: Expand the data science team responsibilities
In this framework, data scientist has to be a one-man army: an expert in their business domain, highly skilled in programming, great statistician and data visualisation designer, able to present their findings in a simple way. While the idea seems to be a great solution for growing data needs without spreading responsibilities, it turns out to be impractical.
Hiring a person like that is extremely difficult, and once you finally find them, there is not enough time to use all of their skills to their fullest. Typically, a data scientist spends about 80-90% of their time programming and cleaning up data. The other 10-20% of their time is gathering actionable insights.
In other words: in this model a data scientist spends about 20% of their time creating insights that can be translated directly into profitable actions. They have to be experts in analyzing customers’ behavior spending about 5-10 hours a week on it, while also being involved in extremely complex programming, and creating reports that should be understood with no technical expertise. Managing so many different responsibilities is virtually impossible.
Approach 2. Create a new team of BI analysts
Another typical approach is hiring a statistician skilled in business analytics, whose sole job is translating the cleaned-up data from SQL tables into actionable business insights. These people are easier to find than data scientists, and can be more easily fit into existing teams to provide them with analyses they need most.
However, they are highly reliant on the data brought to them by data scientists and engineers, without sound understanding of how exactly this data is processed, or what limitations and capabilities of this process are. This way you make your chain of data-driven decision making longer (data science tasks are blockers for your analysts) and more vulnerable to miscommunication (some details in data processing pipeline might get lost in translation, resulting in incorrect interpretations).
Approach 3. Mixed approach
We decided to tackle this problem in another way: create an analytical role, enriched with basic programming skills focused on getting behavioral data about our users. We want to know where our users get stuck, which features are most enjoyable for them, which ones require more working on, etc. In order to cater for this need, the idea was to create a position dedicated exclusively to understanding how our customers use the software we create and how we can improve user satisfaction and productivity based on this data, without encountering problems of the two approaches outlined above.
This led us to creation of Product Analyst job position, an intermediary between big data and product development decision making. While not an expert programmer, a Product Analyst is supposed to be able to get any data they need with little help from data scientists, analyze them in order to reach actionable insights, and present them in a simple way to all stakeholders in the company who can use this knowledge to improve our product. It is still subject to physical limitations similar to that of the data scientist approach, but they are mirror imaged: product analyst is supposed to spend much more time analyzing data, and their programming skills are there primarily to prevent them from being blocked by data science team throughput.
As the first person hired by Base for the Product Analyst position, it has been a great professional challenge for me. Surrounded by well over a hundred of people who understand how valuable data is and who explore data on their own as a part of their work routine, I must make sure that my input is still valuable and can move the product forward. Once I got used to technical side of things and huge freedom I was given, creativity in asking the right questions has become the biggest limitation for progressing my tasks. In my opinion, this is what data analytics should ultimately be about: creativity in asking the right questions with freedom to pursue the best answers to them.