Three Critical Success Factors for AI Projects
What to consider when planning your next effort
Jan 21, 2021

Today, we find ourselves in the midst of an AI renaissance, with artificial intelligence finding its way into almost every aspect of human life. In the midst of this AI boom, businesses across industries are racing to take advantage of the technology and integrate AI into their operations.

While some businesses have seen successes in AI implementation, many organizations are struggling to launch and effectively leverage AI efforts. Here, Software Engineer Sumeet Shah discusses three common components that differentiate the AI success stories from the failures: clear objectives, data, and expertise.

Clear Objectives

As you set off on your AI journey, you must have a destination in mind. While this may seem like a no-brainer, you would be surprised at how often projects get launched without clear outcomes and expectations. Beware – the AI industry sings an alluring siren song promising insights and innovation that has enraptured many an unsuspecting decision maker. Do not be led astray! Clearly define your business objectives and focus on how AI may (or may not) help you achieve them.

From the get-go, primary stakeholders should be able to clearly describe in detail the goals of the project and the measurable benefits they expect to observe as a result of success. These business objectives are the compass that will guide your team. Deciding on your objectives at the start enables a planned, targeted approach. If your organization is new to AI try to pick modest objectives that are achievable, but impactful. Think on the scale of months, or even weeks, rather than years. The experience that you pick up combined with the growing comfort and enthusiasm built around AI with stakeholders will set you up to pursue more ambitious AI projects in the future.

You may learn that in order to achieve your goals, AI is not the right approach. That’s okay! Some problems lend themselves to being solved with AI while others simply do not. AI is not a panacea, and after objectives are articulated it may become clear that using AI will yield little if any benefit. If you can solve the same problem in a less expensive way, you should. Without clear business goals ready at the outset, you may come up with a terrific solution/answer to the wrong problem.

Data

You have clear objectives and have decided that using AI is the best way to achieve them. Now you’re going to need some data. Any AI expert worth their salt can tell you about the importance of data. The availability of pertinent data is a key obstacle that has tripped up many AI projects. Even the most sophisticated AI model will fail if it does not have adequate training data. Thanks to the pervasiveness of technology like the smartphone and the internet, we are constantly generating and gathering user data. In spite of this, we often still find ourselves short of data. As new AI use cases emerge, new types of data become necessary.

Without an abundance of good data, or the ability to generate or obtain it, an AI project will be dead in the water. Just having data is not enough. You need data pipelines that standardize, filter, label, clean, and enrich your data. Many organizations today have tons of data, and want to jump right into training models without putting in the effort needed to properly prepare their data. Bad data leads to bad models with bad biases that make bad predictions.

For example, a few years ago a major tech company attempted to train an AI model to help them review resumes and find the best new hires. They had an abundance of data from all of their previous applicants and hires. Unfortunately, they neglected to clean and normalize their data. As the model ingested the historical data, it noticed that the majority of successful candidates were male, and began to ignore other factors.

Because the tech industry has historically been male-dominated, the model taught itself to discriminate against female candidates regardless of merit and qualifications. Luckily this behavior was caught early on, but this incident is just one of many that illustrates how critical data curation can be!

Organizations that take the time to build robust data pipelines reap the rewards of performant, effective, and reliable AI models.

Expertise

So you have all the high-quality data you need to power your AI. Now all you need is a savvy crew. Like many fields of study, some of the most revolutionary concepts in AI were developed long before it was feasible to implement them. As a result, we now have several mature, tried and tested algorithms and the software tools needed to use them. While the bleeding edge of AI may still be confined to elite research groups at the top tech companies, there are several tools and code libraries that enable the training, testing, tuning, and deployment of cutting edge AI.

The fact that the tools are available doesn’t mean that you’re ready to go. You need a team that is trained and educated on how to apply these tools properly. AI is neither a one-size-fits-all solution nor an exact science. These general purpose tools are not built specifically for your problem, and you may need to tweak them, or even build tools of your own.

Models often yield unusual and unexpected results. What types of algorithms might perform well for your particular challenge? Why might our model be performing poorly in a given sector? What parameters can we tweak to improve performance? Being unable to answer these types of questions will obviously hinder your AI efforts.

Like any tools, the effectiveness of AI tools is determined by the abilities of the person wielding them. In Bruce Lee’s hands, nunchucks are whirling, elegant weapons capable of dispatching 20 evil henchmen to the hospital in short order. In my hands, nunchucks will send one person to the hospital, and that person is me. However, to my credit, I bet I could use AI tools far more effectively than Mr. Lee. In both cases, the difference is not the tool itself, but expertise and experience. Whether from internal enthusiasts or external consultants, having a solid AI knowledge base will allow your organization to confidently drive AI projects forward.

Conclusion

The demonstrated potential of artificial intelligence to revolutionize the way businesses operate makes AI a highly attractive technology for organizations to pursue. In many industries, it’s believed that those who fail to integrate AI will be left in the dust by those that succeed. Every AI project will be unique, and bring its own challenges and difficulties. But all AI projects need the same basic components to even have a chance at succeeding: clear objectives, data, and expertise. The absence of one of these puts any AI project at severe risk. Together, these three factors will serve as the foundation that will enable and empower your organization’s AI endeavors and set you up for success.