Surprisingly, according to a survey of more than 2500 executives conducted by MIT Sloan Management Review, BCG Gamma, and BCG Henderson Institute seven out of 10 companies report minimal or no gains from their AI initiatives.
It wouldn’t be right of us to shift the blame on AI. We need to remember two things – one, AI is essentially a ‘machine’ that learns from the input you provide; it has no inherent intelligence of its own. Two, AI doesn’t have the potential to solve every conceivable problem; it has limitations on what it can and cannot solve. There are various other factors that hinder the success of your AI project. Let’s explore them in detail.
When someone mentions AI or machine learning, scenes from sci-fi movies like Terminator or Wall-E flash in our minds. This is far from reality right now. Will it be possible in the future? We don’t know yet. However, we are so tuned into such expectations that we want our AI models to deliver such advanced results. We need to understand that building an AI model is a long process of trial and error. You need to work patiently through it.
As I mentioned earlier, not all business problems can be solved by AI. You need clearly define the problem you are trying to solve. Your goal should be precise. It should bring value to your business and improve your KPIs. You need to explore whether you have the required data to solve it. On the other hand, just because you have data doesn’t mean you should set out to create an AI project.
If you are familiar with AI, you should know that data lies at the heart of AI. It is crucial for the development and training of any machine learning algorithm. Hence, before you begin building an AI model, you need to build a firm data strategy – what kind of data would you need, where can you find this data, is the data well labeled and structured? What about data governance?
One of the results of a lack of data strategy is using incomplete datasets. If you feed the AI model with an incomplete dataset, you can be sure to expect incomplete or incorrect results from it. While compiling your dataset you need to be careful that you don’t rush the data strategy or planning process and ensure that you have everything you need before kick-starting the development of the AI model.
An algorithm is like a recipe that lists the steps/instructions that the machine needs to follow to come up with an answer. There are many things that can go wrong here.
Developing an algorithm is influenced by the person creating the algorithm. This means that the developer is most likely to transfer their biases into the algorithm and can result in incorrect output by the AI model.
If you are a data scientist and are communicating the details of the project with the business team, make sure that you cut all the technical jargon out. They are most likely not interested in the technical aspects of your project. They want to know the problem you will be help solve and the results (read as return on investment) they can expect from these efforts.
Organizations usually prefer to rope in new graduates or those with very little experience. The reason is obvious – to save costs to the company. This is a huge mistake. In exchange for cost-saving, you’ll receive excuses from your data hires on why the project is getting delayed. You need a data expert who has the experience to handle the AI requirements of the organization.
It’s easy to get ambitious and take up a complicated project from the start. It’s one thing to be ambitious and a whole other to be over-ambitious. Always begin with a simple project and then expand or diversify.
Another reason for AI project fails is the setting of wrong expectations. When you want to sell the idea, you might want to quote a lesser time frame and this may not align with how long the AI project takes in reality. This could be due to one of the many factors discussed earlier in this post.
If you pay close enough attention to the points mentioned above, you can thwart the failure of your AI projects. Before anything else, make sure your entire organization is aligned for this kind of digital transformation.
Have you faced such issues or have experienced something different from what’s listed here? Let us know in the comments.