Peter Gilks & Jacob Olsufka
Peter Gilks — Senior Director of Insights, Spotify Advertising
Since 2016 I’ve led the Insights team for Spotify’s advertising business. Our role is to provide guidance that enables the product and business teams we work with to meet their objectives through evidence-based decision making and customer focus. We’re a multi-disciplinary team that covers both qualitative user research and data science, as per the Spotify model. Our work includes exploratory analysis, experimentation, metric setting, building dashboards, user testing, ethnography, and surveys.
In a previous blog post, myself and Paul Glenn covered the topic of Analytics Engineering — why we think it’s an important discipline, our reasons for investing in it at Spotify, and how we are making progress. In this post I am teaming up with Jacob Olsufka in a similar fashion to give you the low-down on our journey investing in visual analytics.
In much the same way that data preparation is part of every analysis, so too is visualization. Sometimes this may be scrappy, throw away work as we use visual techniques to quickly explore our data. Other times a visualization may be the end result of our work, either as a powerful storytelling image or as a dashboard intended for continued use by a wide group of people.
Data visualization is something that, in my opinion, is easy to learn but hard to master. Every data analyst or data scientist can produce a chart or knock together a dashboard — but to produce the visualizations and tools that drive real impact and keep people coming back for more, that takes some specialization. Whatever tool you are using for the task, it’s easy to get your first charts produced, but it can be difficult to create something close to perfect.
Just as I spotted that data preparation was becoming a bottleneck for our team, I also noticed around the same time that visualization and dashboard creation was also becoming a large time suck that delivered so-so results. N.B.: Now is probably a good time to mention that as a former Tableau Zen Master myself, my standards for visualization and dashboards are pretty high.
My hypothesis is that it’s the ‘easy to learn but hard to master’ element that was tripping us up. All data scientists were using Tableau to create visuals and dashboards, but none was an expert in the tool and we had no standardization so our results were a real mixed bag. The ease of use in getting started can quickly lead into frustration as it becomes difficult to go from good to great without clear expert guidance or the time to invest in this particular skill.
A separate but related problem we were facing at the time was due to our organization. At Spotify we place a lot of emphasis on Data Scientists and User Researchers working in an ‘embedded’ model. For a product team this means working alongside Product Managers, Engineers and Designers, focussing on the same tasks and problem spaces that they are, and building close cross-discipline relationships. This model has clear benefits for building subject matter expertise, but it also leaves a gap when work must cut across teams; there is nobody “central” to handle this. Concretely, our problem with the org structure was deciding who would take on the building of visualizations and dashboards that serve business unit decision-making, and for whom the intended audience is a leader with a broad remit.
Similar to the experience attempting to create data layers, we found that process and training were not enough to solve these two problems. We needed to hire an expert.
We created a new role to cover the following:
- Design and build ‘core’ dashboards for the business unit
- Create templates and guides to standardize and improve our dashboards across owners
- Raise the bar for data visualization in the group through mentoring and 1:1 training
- Provide visualization consulting services
Our first hire into this role was Jacob, who I knew of because of his involvement in the Tableau Community. The team has since grown, as has Jacob’s role.
It’s definitely been a success. The quality of our visualizations and dashboards has increased across the board, as has the skill level of the wider team. We are using visualization to deliver real business value as we strive to continually improve evidence-based decision-making.
As you will read below, the role has also expanded from its initial description in one simple but powerful way — we now not only provide dashboards as a service to others but also purposefully use our own dashboard creations to find new and surprising insights.
Jacob Olsufka — Senior Visual Analytics Engineer, Spotify Advertising
At Spotify, I get to work with a wide variety of roles and teams to bring data and insights to life through visualizations. It’s a very rewarding role. The purpose of my work is to make data and insights easily accessible and understandable for everyone. When I am doing my job well, my work enables evidence-based decision-making to thrive — an important part of Spotify’s culture.
Being on a central team allows me to work closely with all of the Data Scientists within “embedded” teams, and most importantly quickly learn the ins and outs of the different areas of the business. This in-depth context is necessary to build tools that allow leadership to have a go-to place to monitor the state of the high-level focus areas of Spotify’s Advertising business.
Here are a few tips that have helped us build a visual analytics culture:
Maintain close relationships with your stakeholders
In order to get folks the right information at the right time, you need to have a pulse on the types of decisions that they are making, and what their burning questions are while making them. Because my centralized team works closely with leadership, this job is easier than it would be from a more siloed position on a product team. Involving stakeholders and gathering feedback early and often is the best way to ensure that visual analytics work is effective. Even after this initial stage, keeping end users engaged throughout the process is critical. One of my favorite ways we do this is by running small trivia-style quizzes during meetings via chats. We present an insights question in multiple-choice fashion and see how many folks know the answer. We then show how a dashboard we’ve made would be used to answer the question.
Teach and inspire others as much as possible
When you’ve become an expert within your organization, try to find as many ways as possible to provide multiplicative value by teaching and inspiring others. We host several “office hours” sessions a week, where folks can get help with any part of the dashboarding process, from brainstorming and sketching in the beginning to polishing at the end. I have learned that even a small investment of time helping others can go a long way in improving the skills of a group, and makes others excited about what is possible within the world of Tableau and data visualization. This creates a virtuous cycle where we mint more experts who continue to pass on their knowledge to others.
Another way that we have created a successful visual analytics practice is by creating high quality documentation and resources, like design guidelines and principles, dashboarding best practices documents, and Tableau templates. I discussed this more in depth at a past Tableau Conference. In the video of my talk, I share these resources and a few of my favorite dashboarding tips.
Always push for high quality
Our team always talks about “spicy” dashboards and insights, where “spicy” means high quality and high impact. This attachment to a theme of “spiciness” gets people talking about high quality design and high impact insights in an approachable and fun way. Everyone is inspired to go build the spiciest dashboard or share the spiciest insight with their team. Making it fun and cool to aspire for the highest quality possible has been a contagious part of our culture.
Use your own creations
At Spotify, I have had the joy of being able to both build and maintain a suite of dashboards while also being able to scratch that insights itch that had been missing before. By being closely involved in not only the building of dashboards but the utilization of them, I’ve been able to up-level my dashboarding prowess. Through this I have learned that there are many benefits to dashboard builders using their dashboards. As the dashboard’s author, you are one of the best positioned to discover insights with your intimate knowledge of the data and how the dashboard is meant to be used.
Our team has embraced this idea of enabling everyone to find insights through something that we call Insight Safaris. These meetings bring us together as a team to explore a given dashboard, digging around for any interesting trends or data that stand out. We then share our top findings with the rest of the business, with hopes that our work can spark ideas, inspire others to look for similar insights, or generally give folks a better understanding of the power behind our data tools. Even small insights can spark big ideas!
Using your own dashboards, and watching others use them, helps you understand confusion points, design a better user experience, and uncover any bugs or issues in your underlying data. Perhaps it becomes apparent that you are always getting stuck at a certain point on your analysis journey. E.g.: you can see when a metric is moving up or down but it is not easy to understand what is causing it — is there a certain category driving this or is it a universal behavior caused by something else? Is this a seasonal trend or a true outlier? This might inform you that you need deeper levels to your analysis, either built into the same dashboard or in a new dashboard.
I absolutely love using these Insights Safaris as a way to introduce others to new or “under-development” dashboards as well. It serves the purpose of getting comprehensive feedback before putting something into production to a larger audience, while also teaching others along the way how they can be used to find insights so that when it’s released they can take full advantage of it.
Organizations can benefit from emphasizing the skill of being able to derive insights from data alongside the necessary tools to build dashboards.
If you’d like to explore career opportunities in analytics engineering or data science at Spotify, please check for postings on our job page.