Nacho is the founder and CEO of BairesDev, the leading nearshore tech solutions company, and of BDev Ventures, a VC fund for B2B businesses.
A lot has been said about artificial intelligence’s potential to revolutionize virtually every company in the world. The internet is filled with success stories about businesses that reimagined their workflows and boosted their processes by strategically adopting AI. What’s more, a recent Deloitte benchmarking study found that leaders in AI implementation show an average of 4.3% ROI for their projects.
What all those facts aren’t showing is that you can’t expect such a level of success just by “plugging AI into your process.” If that were the case, then there wouldn’t be so many companies that struggle to get real value from AI. And, unfortunately, there are plenty of such companies out there. In fact, the likelihood of success is very limited for companies just starting out on their AI journeys. The same Deloitte study noted companies that are taking their first AI steps get only a 0.2% ROI from their implementations.
There definitely has to be a reason why there’s such a rift in ROI between companies. And there is! In my years of experience working with a software development company, I’ve seen plenty of companies fail in their AI implementations. The main reason: not knowing how to avoid the common pitfalls that prevent companies from getting real value from AI. That’s why I made a list of what I believe are the most important risks in AI adoption along with suggestions on how to avoid them.
1. Failing To Define The Use Case
As I said above, AI isn’t a “plug and play” technology. Yet a lot of companies treat it as if it were. In other words, many businesses believe that they can invest in any AI solution, implement it in a specific process and sit down to watch how it delivers. While I’d love for AI to be that simple, that’s naturally not the case. AI has highly different use cases that need specific implementation strategies to provide real value.
That means you have to analyze your needs and thoroughly understand what use you’re going to give to the AI solution. It isn’t the same to adopt AI to help you with customer support, identify marketing trends or set dynamic prices. Those use cases are very different from one another and imply different approaches to AI. That’s why you can’t start an AI implementation without a proper understanding of what you need it for and what you expect to achieve with it.
2. Forgetting Who Will Be Using The AI Solution
Knowing where you want to get with AI is an important first step in AI implementation, but it won’t mean anything without the step that follows it: understanding who the users of the AI solution will be. This might sound like a simple detail to solve, yet it’s more complicated than that. Broadly defining the user base (or simplifying your perception of that user base) can lead to AI-derived processes that don’t address the users’ true pain points.
The challenge is evident: You need to develop and implement AI solutions that actually boost your talent, not create obstacles. For that to happen, you will have to engage with the people who will be using your AI solutions and solicit feedback about your plan. These people will surely provide you with insights into their everyday issues and the solutions they imagine for them. What’s more, they can help throughout the implementation, pointing out improvements and adjustments.
3. Neglecting The Quality Of Data You’ll Be Using
AI solutions need data to work, but to truly offer valuable results, that data needs to be of the highest quality possible. You can only achieve that quality level by performing a series of practices for all your datasets. However, many companies forget to carry those practices out or perform them superficially. The result is obvious: datasets that aren’t clean enough, that include irrelevant data or that are plagued by incorrect information. When companies introduce that data to the AI solution, they end up getting poor outcomes.
The solution is simple to understand yet hard to develop. First and foremost, you have to understand that preparing data for AI is a continuous process. You can’t expect to work on your cleaning methods once and forget about them. You’ll have to move forward with a data management strategy that contemplates data preparation and governance and that includes adjustments to the process when you need them.
4. Believing That A Perfect AI Solution Is Possible
Many people dream of the perfect AI solution — that is, a solution so developed that you only need to feed it good data to get high-quality results. What’s more, some people achieve something closer to that solution: an AI algorithm capable of providing insightful outcomes on a constant basis. However, that doesn’t mean they have achieved that perfect solution. The reason is simple: An AI model will lose its accuracy and predictive capabilities as time goes by — mainly because data, environments and scenarios change.
If your AI solution fails to change with them, then you’ll quickly have a useless algorithm in your hands. To prevent that from happening to you, you need to do two things. First, establish a continuous improvement process that tweaks and adjusts the solution frequently to reflect the changes in scenarios that will inevitably occur. And second, you need to get out of your head that the perfect AI solution is possible. All AI implementations have flaws, so you’ll always have to keep an eye on how to improve what you have.
Successfully implementing an AI solution isn’t precisely easy, and getting it to provide real value is a huge challenge. However, if you strategize the implementation right and pay attention to these pitfalls, you’ll be one step closer to doing so.