For many executives, October contains the scariest day of the year: Q3 earnings calls that illuminate how distant EOY goals are. Though companies can achieve these goals (reducing costs and maximizing returns) with AI systems, many are spooked when trying to implement the unfamiliar. I’ve found 3 steps help ensure successful AI projects:
1) Outline the problem.
This should be a high-value problem—testing AI’s abilities on a problem with no business impact means the solution will likely not be used, even if it is successful. If the company doesn’t see value, they are less likely to make future investments, and will fall behind exponentially as early investors race ahead.
The issue should also require complex analysis rather than something that could be determined by simple regression. AI excels at finding subtle patterns in data sets under complex or changing conditions, and is not necessary in cases where there’s a clear cause and effect relationship with only a few variables.
2) Gather appropriate data for this problem.
The data provided must be mapped to the problem to be solved. It also needs to be data in which you can reasonably expect to find a pattern (AI couldn’t protect you from failures caused by human error, for example, unless employee fatigue was part of the dataset given).
“Data” can take different forms for different use cases. It could mean a pipeline of historical and current sensor data, a labelled set of images, or something else entirely.
Finally, don’t give data scientists nightmares: clean data that is organized and labeled yields the best results.
3) Define success.
What are the desired outcomes? What is the impact of solving this problem for employees, customers, and workflows? Who will the end user be? How will results be measured? Thinking this through on the front end helps to show returns and cultivate buy-in throughout an organization.
In the right context, AI demonstrates clear value, capturing and utilizing information previously limited to a few experts. Though there will likely be bumps along the way—issues we’ll address in future newsletters like change management or coordinating data across business groups—well-implemented AI will deliver treats, not tricks.
