Why you should not be teaching executives SQL
Democratizing data is often taken too literally by data analytics and IT teams to mean that everyone in their company should have access to tables in a data warehouse and a SQL terminal. While this is useful for data professionals who understand the nuances and caveats behind datasets, it is actually not what helps change a company culture toward making data-informed and data-driven decisions.
In this post, I’ll walk through some of my experiences at Facebook and Airbnb over the past 13 years and provide some helpful guidance to avoid pitfalls that I’ve seen along the way.
Many frustrated and exasperated data analysts and IT people say they wish executives at their company knew SQL. But this sentiment is actually a misguided proxy for the real problem of “I wish people would ask me fewer, low-value questions so I could focus on more important and interesting analysis.”
SQL is not the tool to democratize insights and, in fact, it is a precarious place to send less-data-savvy people who could easily be burned by not understanding the perils behind those raw datasets. Similarly, Business Intelligence tools and data viz dashboards are lacking critical context necessary to turn graphs into democratized insights.
Step 1: Ingesting the right data into the right tool stack
There is a big, complicated prerequisite for democratizing insights that are around data capture, ingestion, ETL pipelines, table metadata, tailored access, and governance. Getting these things right is a different area of focus, albeit an important one, that has been written about at length in posts like this from Towards Data Science.
Then there is the question of what tools you need for ingestion, storage, processing, and visualization that the founder of dbt wrote about the Modern Data Stack. I’m actually not going to focus much on this either and suggest there are plenty of other great posts available to read on the subject for how to empower analysts and IT people with the right data and right tools to do their jobs.
Step 2: Invest in data education early
Once all the hard work of capturing the right data and building a great tool stack is done, many companies fail to work on the last mile of data education, thereby accidentally exposing themselves to low-value questions and becoming unintended answer factories. Really great analytics and data science teams instead jump on this last-mile problem and turn themselves into insights-teaching factories.
At Airbnb, we started a whole data education curriculum and it began with a session where we walked through our company dashboard, widget by widget. I taught this class dozens of times and always made sure to focus on why Airbnb cared about the north star metric of “nights booked,” as well as what that metric actually meant, how it was defined, and why it was important for everyone to care about increasing it. Then we would walk through the other data visualization elements and talk about the main drivers of that metric and what we had learned about its elasticity in the past.
This education was not only powerful to help get new employees thinking analytically and focus on the datasets that mattered, it also was a moment to influence. Only a handful of new employees were comfortable writing SQL, but they all understood the value of key performance indicators and how their work could help drive our business outcomes.
Step 3: Ask data analysts to push insights, not field questions
There is far too much time spent answering questions, and far too little influencing the direction of analytics departments. Companies have inadvertently trained business people to ping someone on Slack or shoot an email with the title, “Hey Quick Question,” anytime they want to know something about data. 75% of people surveyed say that if they want to make a data-driven business decision, they would need to ask an analyst or IT person first. Only 10.7% of respondents would go to a self-service tool.
This is a missed opportunity for analysts to own the narrative about data and share context proactively, rather than reacting to questions. At Airbnb, we would proactively send a weekly email with a handful of graphs, each one with some context and insight about why a metric value changed. This was a great way to not only keep employees focused on the metrics that mattered, but also to fill in a story about why something changed.
Step 4: Embed analysts and ask them to influence
Analysts continue their role as being gatekeepers for data-informed decision-making. (And, by the way, they hate being gatekeepers to basic information.) It is frustrating, low-value work to pull simple stats that should be accessible to all data consumers in seconds. But from the data consumer perspective, attempting to generate insight and make a business decision without help from an analyst is a very risky proposition.
Embedding analysts deeply into product and business teams, and insisting that they pair closely with decision-makers will give them the ability to proactively influence decisions with data. Instead of grading analysts by the number of pipelines they author, tables they generate, questions/tickets they answer, or BI dashboards they build, companies should instead judge analyst performance by the amount of influence they proactively drive within teams.
Survey different departments and ask whether they understand their core metrics, how they are defined, what they mean, and how initiatives to influence outcomes. Ask teams if they have the ability to make data-informed strategic decisions with historical context. That’s the mark of a high-performance team that is data-driven.
Known limitations with current tools
Data professionals who have worked to push insights and build influence know that the tools in their toolkits are not particularly well-suited to make this happen. Generating a data table in the warehouse is only a good tool for a slim handful of consumers. Business intelligence tools do not have 100% participation because there isn't continuity around definitions and ownership, there isn't good context and story-telling about why things changed, and the impact on business KPIs is unclear. Dashboards are also prone to rot, so analysts have to copy-paste data into slide decks, which then never get looked at, let alone edited again. Data catalogs only speak the language of tables, columns, and rows, but really metrics with context are what consumers want.
Conclusion
Companies have been dedicating huge investments in data infrastructure and tooling in order to build an analytics advantage over their competitors. The dream is to “democratize data” and get employees to change their ways of working and start making decisions informed by data, not gut feelings. Those same companies have devised elaborate plans for self-service through data warehouse migrations, SQL interfaces, and business intelligence tools. But it just isn’t working to truly democratize data yet. By investing in data education and helping analysts influence, then building modern tools to support metrics, we will continue making progress toward that goal of truly democratized data.
