Project Management Practices to Manage Complex AI Projects
While we have a structured PM framework which we typically use for all our traditional IT projects, managing AI/ML projects entails a lot of exploratory phases which becomes challenging for PMs to manage. Unlike our typical IT projects, there is no specific solution that the team can implement and goes through the discovery phase to ensure if this is the right implementation for the business case in hand. From the Business standpoint, organizations are aware of the potential benefits these evolving technologies bring to them in terms of increased revenue & business growth in this digital world. With this pandemic around, several organizations have adopted digital transformation to streamline their business processes & embrace new systems making AI-based projects a new norm in the foreseeable future.
However, the businesses are yet to chalk out the concrete use cases they would be benefitting from. Right from the uncertainty of the business outcome to designing and implementing the right solution, there is a lot of mystery surrounded handling these complex AI projects. With so many unknowns, it introduces a lot of risks to be handled throughout the lifecycle of the project. Since AI projects go through the stages of exploration where it goes through multiple trials, the development cost & its timelines become too uncertain and volatile to estimate. There needs to be a POC to be done before implementing the final solution. With no set framework or approach, managing these projects can be quite an uphill task. AI solutions are largely data-driven, so it’s important to have structured & right data to be fed in the AI system for it to produce close-to-accurate & reliable outcomes. The underlying expectation for the business should be that these predictive algorithms wouldn’t be 100% accurate & they can only become better with time. This also calls for bringing various specialized teams in the project like - Data preparation, Data management, Machine learning team along with already defined project teams. This gives a lot of stakeholders for PMs to manage. Also, defining the metrics for the project becomes quite elusive as the scope is ever-changing and your success criteria largely depend on the data availability, the model chosen & the extent of algorithm training that is completed.
Coming to the project management framework that is already in place. While we do have the popular agile methodologies to factor in all the uncertainty experienced throughout the project, the experimentation aspect of AI projects makes it extremely difficult to define the tasks concretely. The correct mapping of the right solution to its right business value need not happen early in the project, which makes adhering to our existing project management process a little challenging. Organizations have moved to their customized project Management methodologies that can be best suited to their needs. Disciplined Agile where teams have the freedom to choose their own way of working can be very handy in managing such complex projects.
Typical phases that the project goes through while implementing AI projects are first identifying the project goal, like the desired end behavior or any specific problem to resolve. The second is launching the base-level beta product to gauge the market response if they are willing to pay for it. This also gives a confirmation if the product needs to be built with AI or without it. Next is moving on to gathering & preparing the right data that needs to be fed into the AI system for it to work efficiently. Then it’s about choosing the right AI algorithm to be implemented on the data based on the business requirement. Once that is implemented, the system needs to be trained by feeding various scenarios for it to produce close-to-accurate results. Once the results are in alignment with the desired business outcome, they can be deployed in production.
It is inevitable to raise the maturity of the project management practices to keep up its pace with the ever-evolving technology. While the tailored project management framework needs to be worked upon, below are a few practices organizations look to adopt for such complex AI projects:
Fail Fast: Since these projects pass through multiple exploratory cycles, it becomes imperative for the project team to adopt the “Fail fast” philosophy. The incorrect approach needs to be discovered early & needs to be rectified before realizing it at the later stages of the project lifecycle. Data gathering & preparation, implementing an algorithm on it, & verifying its outcome with the expected business solution needs to be executed in the shorter feedback loops only to fail fast & implement the right solution at the end.
Foster Collaboration: This is essentially required at the Data gathering stage. One of the pillars for AI’s success is having the right data. It needs to be gathered from multiple teams like the Business team, Engineering & Data-science team & clients only to feed & train the system with the right structured data. AI projects typically fail when there is a lack of team collaboration during this stage, resulting in data silos & inadequate data to develop the system with.
Apart from the regular Project Management skills, AI practitioners need to scale up their skill-sets in terms of building statistical knowledge, AI/ML related technical knowledge of coding, AI-powered business applications & their project workflow only to sharpen their implementation skills. Organizations need to impart specific training to ensure the team is equipped enough to manage such complex projects. This is a new challenge PMs would be expected to surmount in this age of technological evolution.
Thanks for reading . Please feel free to share your comments and feedback.