Differences & Considerations for AI Engineering Project Management

June 30, 2023

In our own firm we have created a comprehensive, multi-stage, end-to-end AI lifecycle process - Enterprise AI - to follow when collaborating with clients. Let’s assume an organization is developing an AI-driven tool supporting a use case to assist in analysis and decision-making. Enterprise data is needed from multiple sources to satisfy the use case.  

Our AI development methodology is embedded in our Enterprise AI framework and has been adapted from an “agile” approach to developing custom AI-centric solutions for clients.  

There are several reasons why AI development is “Agile-like”. In large part it is due to the iterative nature of testing each aspect of the use case’s business objectives and requirements. Expert users determine if the results and output of the AI are correct – or not. When the AI is not providing acceptable output or expected results, analysis is conducted and corrections to the data models and/or machine learning models need to be made. Once completed, another iteration begins until the expected outputs and results meet the use case objectives and requirements. It is often difficult to estimate how long and how many times this iterative process occurs.

Above there was reference to data modelling and machine learning. These are major workstreams when engineering an AI product, and the tasks to complete this work do not necessarily coincide with a standard agile or hybrid methodology to follow conventional data input and software application programming. This is where AI development differs and impacts the methodology to be followed.  

Furthermore, the resources to assess, structure and represent the data that can be fed into an AI model are specialized as are the resources to develop the machine learning model that intakes the represented data to meet use case requirements.  

So, AI also requires its own specialists to represent data and then create computational models to meet the use case’s objectives and requirements. Estimating the number of and type of these AI specialists is also a “learning” process for project managers much less determining percentages of completion in the project plan itself.

If you would like to discuss our perspectives, our responses to the above questions for your organization, and our approach to Enterprise AI, please contact us using the form below or at [email protected].