N※N

People Are the Core of Successful Data Science Initiatives

85

  //science-technology.news-articles.net/content/2 .. core-of-successful-data-science-initiatives.html
  Published in Science and Technology on Friday, November 28th 2025 at 14:30 GMT by Forbes

• 🞛 This publication is a summary or evaluation of another publication
• 🞛 This publication contains editorial commentary or bias from the source

2025-11-28 221 x 224 / 7871 Bytes

2025-11-28 179 x 224 / 7663 Bytes

2025-11-28 224 x 224 / 9599 Bytes

2025-11-28 179 x 224 / 7884 Bytes

Toward Better Data Science: Mostly People, but Also Process and Technology
Forbes, 12 Oct 2021 – Tom Davenport

In a thoughtful look at what it takes to turn raw data into business value, Tom Davenport argues that the key to successful data‑science programs lies primarily in the people who drive them, but that process and technology are the essential scaffolding that allows those people to thrive. The article – found at
[ https://www.forbes.com/sites/tomdavenport/2021/10/12/toward-better-data-science-mostly-people-but-also-process-and-technology/ ] – is an accessible guide for executives, data‑science leaders, and tech teams looking to build sustainable, high‑impact analytics capabilities.

Below is a comprehensive summary of Davenport’s arguments, enriched by the additional context provided by the several links embedded in the original post.

1. The Human Core of Data Science

a. Diverse Skill Sets, Shared Purpose

Davenport begins by emphasizing that data‑science success depends on the right mix of talent: domain experts who understand the business problem, statisticians who can design rigorous experiments, software engineers who can scale code, and communicators who can translate results into actionable insights. He stresses that the most productive teams are cross‑functional and work under a single, well‑defined goal rather than a collection of siloed projects.

The article cites examples of firms that have built “data science labs” as separate units, which can experiment freely before a model is rolled out to the larger organization. These labs serve as incubators for new ideas and a buffer against the “bias of the day” that often sways production models.

b. Leadership and Culture

Beyond skill, the culture a company cultivates around experimentation, failure tolerance, and continuous learning is vital. Davenport argues that leadership must set expectations: that data‑science work is iterative, that hypotheses are tested before building, and that results are shared openly. He points out that in many enterprises, data scientists are treated as “consultants” whose work is never fully integrated into the product or service lines—a mistake that squanders talent and slows impact.

2. Process: The Roadmap from Insight to Implementation

a. Adopting Structured Methodologies

The article links to the CRISP‑DM (Cross‑Industry Standard Process for Data Mining) framework, a six‑step methodology—Business Understanding, Data Understanding, Data Preparation, Modeling, Evaluation, Deployment. Davenport argues that while CRISP‑DM is a starting point, companies must adapt it to their specific context, adding continuous monitoring and feedback loops once a model is live.

He also references the MLOps paradigm—drawing a parallel to DevOps in software engineering. MLOps encompasses continuous integration and continuous delivery (CI/CD) for models, version control, automated testing, and deployment pipelines. In the Forbes post, the linked MLOps article details best practices for tracking model artifacts, handling drift, and rolling back versions when metrics degrade.

b. Governance and Reproducibility

A recurring theme is the necessity of model governance. This includes:

• Documentation: What assumptions were made? What data was used? What metrics were targeted?
• Versioning: Tracking code, data, and model weights so that any version can be reproduced or audited.
• Compliance: Ensuring models meet regulatory standards (e.g., GDPR, CCPA, and domain‑specific rules such as the NIST AI risk framework).

The linked article on Data Governance expands on these concepts, describing policies that define data ownership, access rights, and quality standards.

3. Technology: The Enabler, Not the Hero

a. Toolkits and Platforms

Davenport lists the most common tools and platforms that teams use today:

He notes that while adopting the “right” tools is important, tool selection is usually a downstream decision once the process and people are in place.

b. Data Architecture

Underlying all of this is a robust data architecture: data lakes or warehouses for storage, data catalogs for discoverability, and data pipelines that move information efficiently. The article links to a piece on DataOps, which emphasizes the orchestration of data pipelines with CI/CD practices, automated testing, and metadata management.

4. Putting It All Together: A Practical Roadmap

Davenport proposes a pragmatic “three‑layer” approach:

1. People Layer – Build diverse teams, foster experimentation, embed learning in the culture.
2. Process Layer – Adopt adapted CRISP‑DM, embed MLOps and DataOps practices, formalize governance.
3. Technology Layer – Deploy scalable infrastructure, choose tools that fit the process, maintain flexibility for future evolution.

He warns against tool‑first mentalities. Organizations that invest heavily in high‑profile AI platforms without a clear process often fail to realize ROI. Instead, the investment should flow from the problem statement, through a well‑structured workflow, and only then to the tooling that makes the workflow efficient.

5. Additional Context from Linked Articles

• CRISP‑DM Overview: Provides a step‑by‑step guide to turning business problems into data‑science solutions. Highlights the iterative nature of the process, ensuring business goals remain central.
• MLOps Deep Dive: Explains the importance of CI/CD pipelines for models, automated testing, versioning, and monitoring—critical for maintaining model quality in production.
• DataOps Primer: Shows how to apply software‑engineering practices to data pipelines, ensuring reliability, reproducibility, and rapid delivery.
• Data Governance Frameworks: Outlines policies for data ownership, privacy, security, and compliance, essential for any data‑heavy organization.
• NIST AI Risk Management: Offers guidelines on assessing, monitoring, and mitigating risks associated with AI systems—an emerging requirement for regulated industries.

6. Bottom Line

Davenport’s article crystallizes a simple truth: People are the engine of data science, but they need a well‑engineered process and a solid technological foundation to deliver value. By aligning talent, governance, and infrastructure, organizations can create a culture where data‑driven insights are not just generated but are actionable, scalable, and trustworthy.

For executives looking to scale their analytics programs, the takeaway is clear: Invest first in the people and the process; the technology will follow, fit, and evolve around them. This human‑centered, disciplined approach is the most reliable path toward sustainable data‑science excellence.