Over the past years, we’ve seen a significant increase in how traditional industries leverage technology to improve their businesses. One of the most disruptive I’ve seen is using Artificial Intelligence (AI) and Machine Learning in everyday business operations. What used to be associated with science-fiction is now disrupting most, if not all, industries—from mundane tasks to groundbreaking insights.

AI and Machine Learning take large amounts of data, analyze them as they come in and identify patterns at a fraction of the time. This is incredibly useful for decision makers as the solution learns over time and presents the data as digestible insights. As more industries adopt AI as its new norm, it’s important for Project Managers to know the implications of AI projects and how they differ from traditional software development.

The success of traditional software development projects is determined by the quality of the software that is produced and the timeframe to develop the solution. Regardless of if a project is run with Waterfall or Agile methodologies, the basic steps that go into development remain the same: define the solution, design it, develop, test, and deploy.

Traditional software development projects take prioritized requirements to scope out the duration of the project. AI projects, on the other hand, do not rely solely on developing quality software but on interpreting data to extract patterns and insights as well. For instance, when machine learning is embedded into a software product to learn and identify patterns in data, the quality of the data and the algorithm used take precedence in the project. This becomes an obstacle for AI project managers, who cannot use past examples or past projects to accurately determine the duration of the project.

The data component also makes the development different than a traditional project as the data has to be cleaned and prepared for usage first. This process may take weeks or months depending on the volume of the data. This poses a big risk on AI projects and it may be the reason that, on average, 85% of AI projects fail.* There is considerably less ambiguity in traditional software development projects. That’s why it’s easier to determine the duration of the project and its complexity.

Another factor to take into consideration is that AI projects leverage Machine Learning. Machine Learning algorithms learn from data rather than through explicit programming. AI is constantly asking questions of the data its analyzing and that data is constantly changing. As new data sources are added, the outcomes can change as well.

For this reason, it’s difficult to determine an end to an AI project, because the end is not clear. That is, of course, if stakeholders are not pulling the plug for the project because the project is taking longer than expected. Being an AI project manager means living with the “no end” reality. Embracing ambiguity must be acceptable to company management and project stakeholders as well.

An important factor in deciding whether or not to take on an AI project is establishing a shared understanding between management and stakeholders that as data changes, the outcomes can too. Aligning everyone’s expectations is part of the process in accepting uncertainty as part of the AI project evolution.

Questions to ask are: When can we conclude the AI project is finished? How do we realistically manage these “no end” realities with our stakeholders who DO have a defined timeline?

The goal that we have determined in the AI projects I manage at Capmation is reaching a 95% acceptance of AI outcomes that stakeholders or end users can use. Once the 95% mark is reached, there is usually consensus that the AI output is accurate enough to use on a day-to-day basis and sign-off to deploy to production.

In traditional software development projects, this sign-off by stakeholders would mean the solution goes live and the development work is over. That isn’t necessarily the case for AI projects. Remember, we reached a 95% acceptance, and the outcomes are always influenced by the data it analyzes. As time progresses and data changes, the AI solution will begin to have a decline in output accuracy. At this point, the solution will need to re-calibrate to deliver value to end users. However, this falls under the solution’s maintenance plan, and that’s a blog post for another day.

As AI disrupts many industries, it’s important for digital project managers to prepare and learn how to manage not only the AI project itself, but the many uncertainties and risks that come with them. Uncertainty is a significant risk in these types of projects and, unfortunately, it’s inevitable. However, making sure everyone’s expectations are aligned from the start of the project will help in overall project success.

*IndustryWeek.com; “AI Has a Poor Track Record Unless You Clearly Understand What You’re Going for”

Have an AI project you would like to discuss? Connect with us now.