5 Reasons Why Your AI Projects Are Not Succeeding

Continuing our 5 Reasons series, this month we talk about why your AI projects are not reaching the level of success you were hoping for. We all get a great deal of pressure to implement AI, and it’s exciting to hear all the intriguing success stories from our colleagues–but what if your own experience is poor? We analyze  5 reasons for why you may not be successful, or not as successful as you thought you could be, and give you solutions to apply to each of them.

1. You don’t have a solid knowledge corpus

You implemented a spiffy chatbot and it comes up with… not much, or random things, or outdated advice. A likely cause is that the chatbot does not have an appropriate corpus of data, meaning a large enough set of validated data. For many complex-support organizations, this means a well-rounded, well-managed knowledge base. If yours is tiny or last updated three years ago, the chatbot will reliably come up with bad advice.

Knowledge management can itself be assisted by AI tools, but it’s essential for the success of AI retrieval tools.

Lesson: The unflashy work of knowledge management is essential to the success of search and generative AI projects. Don’t neglect it.

2. You expected perfection from the get-go

Early days with an AI tool can be underwhelming. The tool can miss seemingly obvious solutions and come up with ideas that are glaringly non-sensical. Think of AI tools as toddlers exploring the world. They are just trying things and learning from their experiences. (They are not as cute as toddlers, granted.) Monitor their progress and you might be amazed of how quickly they move from incoherence to wisdom. And if they don’t, ditch them.

Lesson: AI tools are (very fast-learning) toddlers: give them time

3. You turned on automation too soon

Related to the point above, turning on automation means giving your toddler the run of the house. Danger! Using the tools in supervised mode is essential to start, especially if automation will be visible to customers. Once you have reach a high level of reliability, you can turn on automation.

Lesson: Test and test and test again before turning on automation

4. You just changed your processes

AI tools are super-powerful pattern matchers, and they match to history. For instance, if you just made major changes to your case categories, an automated case-classification tool will, for some time, continue to use the old categories–until it learns the new ones. This is normal, and an excellent reminder that change takes time. (Automated tools, on the whole, do learn and change faster than humans.)

Lesson: if you make significant changes in underlying processes, there will be a lag in the performance of AI tools during the learning period

5. Your top performers are not adopting

Your most skilled team members can often do their work faster and more accurately than with any copilot tool since they have years of experience and can effortlessly recall and use their deep knowledge of your products and organization. As a result, they may skip using the tools, and, consciously or not, discourage others from using them.

On the other hand, novices usually benefit greatly from copilot tools, at least as a first line of defense, before they consult your experts. They are the ones who will derive the most benefit from copilot tools so work with them to start. Your experts may find that they, too, can benefit when they are working on rare or esoteric problems, but that will only be a rare occurrence.

Lesson: Focus adoption programs for copilot tools on the novices and mid-level team members.

What are your AI challenges and what have you done to overcome them? Please share what’s working for you in a comment