Introduction

As a data science consultant, the projects I undertake can vary significantly, reflecting the needs and capabilities of diverse clients. This article explores the typical data science projects I encounter in my role, highlighting the evolution and current trends in the consultancy sector.

Initial Outsourced Model Builds

When I first began my consultancy, the majority of projects involved outsourced model builds. Clients often lacked internal data science teams and would seek our expertise to perform data extraction, conduct comprehensive analytics, and provide actionable insights through presentations. In some cases, the code was transformed from SAS to SQL for better implementation and scalability at the client's end. However, a more common practice was to deliver a series of PowerPoint presentations and re-run these analyses as needed by the clients.

Shifting to Higher Value Models with Internal Capacity

Over the years, as clients began to build internal data science capabilities, the nature of the projects changed. The internal teams started handling the majority of data science work, with my role transforming into project management and acting as a liaison between the internal staff and the wider stakeholders. This transition required validation and assurance that the internal team's work met high standards. Essentially, having us 'sign off' on the projects helped maintain stakeholder satisfaction without requiring constant oversight from a consultant.

Recent MLOps Projects

In more recent times, the majority of the projects I handle have shifted towards MLOps. These projects focus on implementing best practices for machine learning operations, including governance, deployment, and maintenance of models. Despite the growth of internal teams, there has been a challenge in scaling these teams effectively. Stakeholders require assurances that best practices are consistently followed without needing to assign a consultant to every single project. While this is a valuable approach, it often involves too much engineering for my comfort zone.

Unique and Unexpected Projects

It's worth noting that the consultancy landscape is filled with unique and unpredictable projects. Some clients still lack robust analytical systems, necessitating extensive data extraction and analysis. Conversely, other clients are developing advanced ML platforms, and by the time they need consultancy services, they are too far along in their development for us to meaningfully contribute. This variability highlights the importance of understanding a client's business in depth and tailoring projects accordingly.

Project Constraints and Client Standards

While projects can vary widely, there are certain constraints and standards that define typical projects. Generally, it takes a considerable amount of time to thoroughly understand a client's business needs, which limits the scope for very short projects. Clients with at least a revenue of 20 million and requiring projects of at least 100 hours are common. However, exceptions do exist, especially when dealing with two clients with very similar problems. Typically, clients engage a consultant after trying various other approaches without success.

Conclusion

The nature of data science consultancy projects is continually evolving, adapting to the growing internal capabilities of clients and the increasing emphasis on operational excellence. As the consultancy landscape changes, consultants must remain flexible and adaptable to meet the unique needs of their clients successfully.