Democratizing Data – Building A Data Driven Culture
What does self-service analytics look like in practice? How do you help people make sense of it and prevent false assumptions? Technical and team challenges can make it hard to make analytics truly self-service. We spoke with four data and analytics leads to hear how they approach this question.
Look at actionable metrics
Alan D’Souza, Director of Product Analytics, Lending Club
As part of their analytics strategy, Lending Club’s data team is helping people move beyond vanity metrics by making sure that anyone can drill down into data and ask insightful follow-up questions.
We’re conditioned by incumbents to want these big metrics like page summary. But those big metrics, they’re not actionable. What improves things for customers are specific questions—how is X affecting Y? What’s happening with Z? Specific questions are the things that create value, which is why we want to spend as little time as possible on these high-level, feel-good numbers and more time tackling specific things that create value day-to-day. Everyone should be able to instantly answer business questions.
Learn more about Lending Club’s analytics strategy here*.*
Data access isn’t one-size-fits-all
Harry Tannenbaum, Head of Business Analytics, Nest
Before we can answer this question, we need to define “self-service.” Data is really only self-service if it’s readily usable for the task at hand and matches the level of detail needed by the person analyzing it—raw data for analysts vs. summarized recommendations for executives.
Individuals and managers use data differently. In your analyst use case, you’re building a freeform tool, you’re building access to data. The operational people want a cube, a Tableau dashboard that has 10 different filterable parameters that they can slice and dice to get what they want. The folks at the other end just want the number, the target, the goal.
So, on the spectrum, people need pure tools on the left, and pure recommendations on the right. You should be going to your CEO and recommending a certain decision, not giving her raw data to synthesize. 95% of your output to your CEO should be English, some charts.
As an example of matching needs to access, sometimes you’ll build a cube or block for a director where it would’ve been better to provide something more straightforward. On the other hand, if you give something simple to ops, they might want the raw data instead.
As far as the data team goes, we think of our output in terms of who we’re working with. If you’re the head of analytics at a company, you should feel confident enough to go into your CEO’s office once a month with ten recommendations, ordered by revenue impact in descending order, and put that up on the wall. The data and analytics team’s output needs to ultimately empower people to do their jobs well.
Maintain event definitions and schemas
Maurice Mongeon, Data and Analytics Lead, App Annie
Before opening up data for analysis, it’s the data and analytics team’s responsibility to model and organize that data for everyone—then maintain it.
Self-service analytics only works when you have systematized definitions of your events and control over what’s stale and deprecated. Things get messy when people aren’t on the same page. Someone needs to have a lot of control about what’s being fed into the analytics tool that they’re leveraging.
Theoretically, self-service analytics should be great if everyone’s on the same page and using the same definitions. I usually see problems with this when we have two different definitions for an active user, or two different fields for a usage metric. So, using an actual example, we once had someone sending reports to customers because he thought they weren’t seeing any usage, but it was actually because he was looking at a stale field that had been deprecated.
Self-service analytics only works when you have systematized definitions of your events and control over what’s stale and deprecated.
If you approach the problem from both ends and unify definitions—in your database and across all dashboards—it works. For example, we’re Heap users. The cool thing for us is that Heap and Heap SQL/Redshift speak the same language. We have a library of 1,800 events, organized by product and page, what users did on the page, and what they’re actually interacting with. This becomes language that people get used to. They’ll see those events in reports generated by us, dashboards from the BI team, and individually they’re seeing it in Heap. There’s no confusion.
More can be more
Ville Tuulos, Head of Data, AdRoll
Even with the right tools and alignment on event definitions, data doesn’t always paint one clear or “correct” picture. You still need to understand how this insight applies to your specific business and how that translates into next steps. Here, communication is essential.
I have people doing self-service analytics arrive at the wrong conclusions all the time. I once had a very heated fight about how people should use our dashboards. Technically, it’s an easy question. But when you have enough data, you can find support for any kind of hypothesis, and then depending on the question you ask, you may get one answer that you think is the right thing to do, while someone else sees something else, and an authority wants to take a different approach, all based on analysis of the same data. And everyone goes in separate directions.
Especially as a company grows larger, it’s easy to justify any kind of thing with data. After looking at some fancy dashboard showing some kind of health metric about the product, someone can use that as an excuse to ignore feedback from individual sales people or product managers and so forth.
It’s super important to keep that communication channel open. If you have many people looking at data from different angles, you need to talk to others who are looking at things from a different perspective to get more nuance, more color.
Enjoy this post and want to make everyone in your company better with data? In our webinar on How to Build a Data-Driven Organization*,*we share a new framework for building a data and analytics strategy. Listen here.