An AI/ML project is often more challenging than an “ordinary” software project. It comes with unique AI risks that can threaten the desired business outcome. 

Given the high rate of project failures, or those that remain in “POC Purgatory” unable to be scaled into production, it’s critical for AI project owners to understand the AI risks they face to reach a successful product execution. In this article we will discuss one of the AI risks: the importance of building a team with the right skill sets. 

An AI/ML project needs more than data scientists – It takes a team

Many companies believe that if they assign a brilliant data scientist to lead the project or as the only project resource, their success is guaranteed. 

But in reality, to achieve success in an end-to-end AI/ML project, we need more than a team of data scientists. Support is needed from other parts of the organization in order for any AI/ML project to become operational, much less be maintained in production over the long-term. The full complement of roles includes various skills and experience across product, project, engineering, operations and business stakeholders to work together with the data science team. 

When we build a team, we start with foundational skills in data science like data engineering, ML engineering, data analysis and exploration, model prototyping and creation, visualization, and naturally, data scientists. We also add skills for software development, DevOps engineering, quality engineering, and security, complemented by product management and product owner, as well as UX, to ensure all the software roles integral to production release are included.  

The following image shows how software development engineering roles combine with data science to form a full team, and their association to stakeholders. Keep in mind that not every role is needed from the onset of a project, therefore we progressively bring on the right skills in the right phase.  

In our experience, it’s these adjacent and complementary roles surrounding the data scientist that enable companies to effectively operationalize the value of AI/MLprojects. In the famous words of DJ Patil, “Data Science is a team sport.” At Daitan, we agree.

How an AI/ML Project Team Should Work Together 

AI/ML projects are complex and often result in producing pivotal models tied to a company’s strategic growth. When the impact is high, the visibility is high as well. So, getting it right and putting the team on a path to execution is vital. With a strong team structure in place, we can then focus on building communication and collaboration. 

In the initial phase of a project, we use tools like ML Canvas, Impact Mapping or Value Stream Mapping to bring clarity to the project’s vision, goals, and metrics. This ensures alignment across the team, and with business stakeholders, so the project starts from a common reference point with clear measures of success. This process frequently generates documentation everyone can reference as they progress forward using normal development best practices.

When it comes to the actual engineering process, there are similarities between “traditional” software development and data science that are meant to drive quality and further integrate them into an Operations environment.

For example, both leverage experimentation and iteration to deliver effective results.  And both work towards continuous integration, delivery, and mechanisms for feedback to continuously improve the product.  

In software development, it was the emergence of DevOps(and you could say test automation) that bridged engineering and operations—improving the transition from software engineering into production environments at-scale with high-quality. In data science, the emergence of MLOps and DataOps have provided similar value aimed at optimizing the AI/ML model as data changes over time.  

Net-net, these are enabling processes—often associated with specialized training and certifications—that correlate to skilled roles on the team. For example, it’s roles like DevOps engineer, quality engineer, test engineer,  ML engineer, and data engineer that have risen to prominence in recent years because they are key to ensuring sustainable quality and continuous product improvement.  

In simple terms, this high-profile team works something like this: Data scientists focus on the science and conceptualization of an AI model into the business vision, while data engineers curate and transform the data, bridging the project to software developers and machine learning engineers who produce it for quality engineers to test and validate. Then once in production, the model is monitored for performance and optimized to maintain high-quality results. 

When you approach team building with this mindset, you build something that can release products that adhere to normal operational models and create the business value management expects.

Summary

It’s been interesting to see the impact of AI/ML over the past few years. Increasingly, it drives our conversations with senior executives—across any industry—who are interested in leveraging AI/ML, and data they generate or data they need to consume. And in those conversations, we recognize the complexity and challenges faced when building a fully staffed team with the right AI/Ml and software skills and expertise.

If you’re working on initiating data science projects, we hope this article provides some guidance for taking the necessary steps to build a team with the right skills to successfully productize an AI/ML model. If you’d like additional insights around the context of your AI/ML initiatives, feel free to reach out to me via on LinkedIn. I’d be happy to talk further about your needs and how to overcome the challenges of building the right team.  

Having the right skill sets is 1 of the 6 risks we discuss in the e-book, How to Use Risk Analysis When Evaluating an AI Consulting Partner, authored by Ivan Marin, Principle Data Scientist and, myself, Graham Holt, GM, Digital Solutions. To learn about all 6 risks, download our e-book today.  

Read our e-book to learn how to de-risk your AI project

DevOps¹ – DevOps is a set of practices that combines software development and IT operations. It aims to shorten the systems development life cycle and provide continuous delivery with high software quality. DevOps is complementary with Agile software development; several DevOps aspects came from Agile methodology, Wikipedia

MLOps² – MLOps is a practice for collaboration and communication between data scientists and operations professionals to help manage production ML lifecycle, Wikipedia

DataOps³ – DataOps is an automated, process-oriented methodology, used by analytic and data teams, to improve the quality and reduce the cycle time of data analytics. While DataOps began as a set of best practices, it has now matured to become a new and independent approach to data analytics, Wikipedia