Rethinking the Data Science Hierarchy

The familiar "Data Science Hierarchy of Needs" pyramid—stacking logging and data collection at the base with AI and deep learning at the apex—is a useful mental model, but Shopify has found it limiting in practice. The standard pyramid tends to emphasize specific tools and subtly prioritizes "advanced" solutions over the most appropriate ones.

Shopify's data team has reworked that model into its own hierarchy, where each tier represents a distinct way the team creates value rather than a step on a ladder of increasing sophistication. The philosophy is deliberately tool-agnostic: solve problems with the simplest effective approach first, and only escalate to more complex methods when the situation genuinely calls for it. In this framing, the goal of data science isn't machine learning for its own sake—it's making an impact on the business and helping merchants make better decisions.

The approach was put to a severe test during the COVID-19 pandemic, when Shopify stood up a rapid-response task force drawing data scientists from across its embedded business units and product teams. The mandate was to surface insights about how the pandemic was affecting merchants and to support timely, data-informed decision-making.

That effort, known as Shopify's COVID-19 impact analysis, demonstrated exactly how the reworked hierarchy functions in practice—and why getting the lower levels right matters most in a crisis.

Level One: Collecting and Modeling Data

At the base of Shopify's pyramid sits the foundational work of data generation, platform engineering, acquisition, pipeline building, data modeling, and cleansing. You cannot build sophisticated models or produce sharp analysis without clean, conformed, accessible data. Shopify applies Ralph Kimball's Dimensional Modeling methodology to keep warehouse data structured consistently, making it easy for any team member to query and trust the data they use.

This isn't glamorous work, but it is critical. During the COVID-19 analysis, this groundwork paid off immediately: the team wasn't scrambling to locate or clean data. It was already structured and dependable, which gave the resulting insights credibility when decisions had to be made quickly.

Level Two: Describing the Business

With a solid foundation, the team can move to description—reporting metrics, building dashboards, and answering fundamental questions about product adoption, merchant behavior, and buyer activity. Activities at this stage include reporting, KPIs, segmentation, product analytics, and exploratory data analysis.

Descriptive analysis establishes a baseline. For the pandemic task force, the key questions were direct: What does COVID-19 mean for merchant sales? Which merchants are being hit negatively, and which are seeing an upside? Understanding the current state set the stage for deeper inquiry, though for some projects, this might be a perfectly sufficient stopping point.

Level Three: Predicting and Inferring

When the problems get harder, the team escalates to statistical analysis, causal inference, deep dives, and machine learning—including predictive analytics, anomaly detection, and classification. This is where the team starts to look forward and explain the "why" behind observed patterns.

During the pandemic, this meant looking at what was happening in different regions and extrapolating. If merchants in Italy were seeing a certain pattern of impact under lockdown, what could be expected in the United States under similar conditions? The predictions yielded a working hypothesis about what was likely to come next.

Level Four: Prescribing Action

At this level, the accumulated insight turns into concrete recommendations—both for internal product and business decisions and for guidance offered to merchants. The toolkit here includes analytics, machine learning, A/B testing, and deep dives. Crucially, the prescriptions are only as good as the understanding built at the levels below.

The COVID-19 analysis led directly to a series of product and policy pivots:

  • An extended 90-day free trial for new merchants, in response to a surge of businesses moving online during lockdowns
  • Expansion of Shopify Capital, then U.S.-only, to Canada and the UK, in response to anticipated financial strain
  • Broadened shipping options, including local delivery and buy-online-pick-up-in-store, to address delivery delays and rising ecommerce volume
  • Free gift cards across all Shopify plans and a new feature in the Shop app to help consumers discover and support local merchants

Each of these moves followed directly from observed and predicted conditions, not from a mandate to apply a particular technology.

Level Five: Influencing Direction

The top of Shopify's hierarchy is impact. Analytics, machine learning, AI, and deep dives all belong here, but so does "whatever it takes." The insight that changes a decision or shifts organizational perspective is the highest value output—regardless of whether it came from a simple query or a deep learning model.

The COVID-19 impact analysis itself is testament to that principle: it involved no artificial intelligence and no machine learning, yet it helped guide Shopify and its merchants through the crisis. In 2020, Shopify merchants generated $119.6 billion in sales—a 96% increase over 2019. The result was not a function of exotic tools but of a well-constructed data foundation, disciplined analysis, and the willingness to act on what the numbers clearly indicated. That, in Shopify's view, is the true capstone of the data science practice.

Where Impact Actually Happens

Positive influence at Shopify isn’t reserved for projects that reach the top of the Data Science Hierarchy of Needs. Work at any tier can drive meaningful outcomes. The pyramid’s apex is a reference point, not a gate: it’s the lens through which every model, tool, report, or analysis should be viewed, while the supporting layers enable the depth of inquiry that makes the higher-level work possible.

Shopify’s COVID-19 impact analysis is a case in point. Anchored by the hierarchy, the analysis produced insights that were acted on directly to support merchants during a period of acute need, while simultaneously informing company-wide business and product strategy.

The hierarchy is best treated as a mindset, not a strict sequence. Regardless of which level a team is operating at, the guiding question stays the same: how does this work enable positive change for Shopify and our merchants?