5 common pitfalls when implementing AI, and how to avoid them

The expectations for artificial intelligence (AI) are endless. Yet, while AI can indeed boost business and create new opportunities, implementing such initiatives always implies change, and change never comes easy.

Despite all the excitement, many organizations are not particularly well poised for an AI implementation. In my experience helping clients apply AI in their business, the greatest challenges I see are not rooted in technology or mathematical skill, but rather stem from organizational cultures, silo mentalities and old practices.  

Following are five common implementation challenges, along with suggestions to overcome them:

1. Forgetting about the customer

AI is not a single tool you can start using like a new customer relationship management (CRM) system. From a technical point of view, it is a collection of advanced analytics and automation solutions that can be used for so many different purposes that it will cross industry boundaries and create new disruptive business models.

As a result, it is important to focus on the business value of AI to the end user. For example, people buying insurance policies are not just buying insurance, they are buying security. Those calling taxis are not just buying rides, they are buying a reliable, easy and fast way of getting to their destination.

Solution: Even if you are starting with a minor pilot project, always bear in mind the real value of the product or service the customer is buying.

2. Sticking to old paradigms

Established companies―even the best in their fields―frequently get stuck in old paradigms. Things are done the way they always have been because old ways seem easier than learning new ones. Even today, some people still print out and fax documents because it is “easier.” Well, most certainly it isn’t.

If old habits are not discarded, AI shouldn’t be used to optimize inefficient procedures. Optimizing a horse-drawn cart won’t give you a car, even if AI is used to select the best horses in the world.

Solution: If a process is a relic, modernize it before looking to AI.

3. Asking the wrong questions

I believe that our data analysts likely could predict any recurrent event if they had sufficient data. Finding an answer is not always the most difficult part of the job. What’s often harder is asking the right questions.

It is fundamentally important to formulate research questions correctly in the initial stage of an AI project. If the findings remain “nice to know” and do not spur any vivid discussion, the issues to resolve were probably based on the wrong questions. Without sparking further discussion, AI becomes nothing more than “dormant analytics” that will not benefit the business.

Solution: Ask the questions that will spur discussion.

4. Not engaging the right people

What if you’ve made sure the issues to be examined are the right ones, and you think the results are impressive, but you can’t get people excited or make things happen? Maybe it is because you didn’t engage the right people.

It is only when the right people are involved that AI can become useful and working solutions can be scaled for wider use in the business. If it seems the results do not make much of a difference to anyone, you probably did not engage the right people―often these are management or other decision makers.

In the worst-case scenario, if no one is interested in the results of the AI project, the solution will be forgotten, and the status quo will prevail.

Solution: Find and engage the right people at the very beginning of your project.

5. Failing to substantiate the value of AI

It goes without saying that AI should provide some added value to the business from the very beginning. If the benefits are not obvious to the decision makers, the AI pilot will never make it to production.

Solution: Provide justifiable business case calculations, such as measurable savings. For example, if you can decrease customer loss by 10% using AI, it is a fairly straightforward task to provide a savings calculation. Be sure to monitor pilot results and update calculations accordingly.

As you can see, these common pitfalls are organizational and cultural. Many AI projects don’t progress beyond the proof-of-concept stage because of these challenges, not technical ones.

CGI helps clients overcome challenges to adopting AI and other technologies. We can help you drive new business models, services and products and reinvent customer touchpoints. I invite you to visit our emerging technology page to learn more.

(initially published by Samu Paajanen)