Can You Teach the Elephant to Dance? Why Culture Eats Data Science for Breakfast

TL;DR

Data science has become the bedrock of modern startups, yet it remains an "alien limb" to most traditional enterprises. In this KDD invited talk, Jonathan Becher (Chief Digital Officer at SAP) argues that the barrier to data-driven decision-making isn't a lack of sophisticated algorithms—it's the corporate antibody response. To make data science core to a business, we must stop solving peripheral problems and start tackling the culture at the heart of the organization.

The Pragmatic Gap: Why Big Tech Wins and Big Corps Wait

In the academic and startup world, we treat data science as a given. Companies like Uber, Tesla, and 23andMe aren't just using data; they are their data. However, Becher points out a harsh reality: these are the anomalies.

The "Elephants"—large, multi-national corporations—treat Knowledge Discovery and Data Mining (KDD) as:

• An Experiment: A proof-of-concept that never reaches production.
• An Afterthought: A layer of "analytics" applied to data after the business process is already finished.

The core motivation behind this talk is to understand why, despite 20 years of ubiquity, data science hasn't yet "taught the elephant to dance."

The "Antibody" Effect: Culture vs. Data

The most striking insight from Becher is that practitioners often contribute to their own irrelevance. By focusing on "safe" problems on the periphery of the business, data scientists avoid friction but also fail to influence the company’s trajectory.

When data science attempts to change the "heart of how a company operates," it triggers what Becher calls Corporate Antibodies.

• Fear of Automation: Replacing intuition-based management with data-driven logic.
• Process Rigidity: Established hierarchies that value "experience" over "evidence."
• Siloed Data: Cultural barriers where departments treat data as a source of power rather than a shared asset.

Methodology: Changing the Heart of the Business

Becher’s approach at SAP (as CDO) wasn't just about deploying more Spark clusters or better models. It was about creating a "Digital Business Unit" that operates with minimal human interaction. This is a structural change, not just a technical one.

Key Pillars of the Transition:

1. Core Integration: Moving ML models from "reports on a desk" to "engines in the machine."
2. Minimal Interaction: Reducing the surface area for human bias in routine business decisions (discovery, buying, renewing).
3. Executive Sponsorship: Becher’s own role reporting to the CEO highlights that data transformation must be a top-down mandate to survive the "antibodies."

Critical Insight: The "Small-Data" Problem in "Big-Data" Culture

The paper suggests that we often over-fetishize the "Big Data" aspect (the technology) and ignore the "Small Human" aspect (the culture). Even if the results show a quantifiable increase in efficiency, a leader who feels threatened by an algorithm will find a way to discredit the data.

Key Takeaways for Data Scientists:

• Value over Sophistication: A simple model that changes a core process is more valuable than a deep learning model that sits in a sandbox.
• Empathy for the Enterprise: Understanding the "Practical Realities" of a multi-national company is as important as understanding the gradient of a loss function.

Conclusion: The Elephant Still Needs a Teacher

Nearly a decade after this talk, the struggle persists. Large companies now have "AI Offices," but the "culture eating data science for breakfast" remains a primary reason why GenAI and predictive analytics often fail to deliver ROI in legacy sectors. The message from SAP is clear: To teach the elephant to dance, you don't just need better music (data); you need to change the elephant's mind.

Future Outlook: The next generation of enterprise data science will likely focus on "Human-in-the-loop" systems that gradually desensitize the corporate immune system, rather than attempting "Big Bang" disruptions that cause organ rejection.