Over the years, interest in artificial intelligence has cycled through periods of intense hype followed by mass disillusionment. To achieve genuine business value from AI, organizations must move past the hype and focus on pragmatic deployment – but that also requires a shift in mindset. Focusing on the following six critical elements will provide a solid foundation for doing so.
Define a Clear Business Objective
The first step is to define clear business objectives, ensuring that they align to problems that AI can realistically address and outcomes that organizations can anticipate. Companies with nebulous goals like “implement a chatbot for customer service” tend to struggle, while companies that are more specific (e.g., “reduce customer churn by 15% using predictive analytics,” or “build a prototype AI solution to quantify the business value of a new inspection system”) are much better positioned for success. The more specific the desired outcome and the “why” behind that outcome, the less risk there is of misalignment. Minimizing the possibility of misalignment maximizes the likelihood that the product you build will achieve the desired outcome.
Specify What the AI Model Should Do
Continuing on the theme of specificity, it’s important to be as exact as possible when articulating what the AI model should do to achieve the desired outcome. One good way to accomplish this is to have the product owner simply write out paired prompt responses in as much detail as possible, e.g., “If I present this input to the model, here is/are the expected output/outcomes.” At my company, we are seeing increasing numbers of product managers write out 10, 50, or even 100-plus paired prompt responses, and the results correlate strongly with product success. Specificity helps the entire product team – governance, security, and user communities – to maintain a clear, shared vision of how the product should perform.
Build a Multidisciplinary Team
The third element is to ensure that your team is multidisciplinary. Successful AI deployment requires a combination of technical expertise, like that of data scientists and engineers; domain knowledge, like that possessed by business analysts and subject matter experts; and change management skills – in other words, diversity in all relevant dimensions. A cross-functional team can ensure that the AI solution is technically sound, aligned with business needs, and effectively integrated into existing workflows. However, it is also critical to recognize that assembling this team is a “long pole in the AI tent.” Whether hiring, training, or sourcing the team from a third party or partner, organizations must start early.
Prioritize Data Quality and Availability
Fourth, prioritize the quality of your available data. Without high-quality, relevant, and representative data that’s free from bias, most models will struggle to produce accurate, acceptable results. Consequently, data is typically another long pole in the AI tent. We advise clients to invest in data cleaning, preparation, and integration early in the development process. This means making sure the data is accessible, properly labeled, and representative of the real-world scenarios the AI system will encounter.
Focus on Incremental Progress with Practical Implementations
Fifth, focus on practical, incremental implementation, starting with small, well-defined pilot projects that address specific business needs. This is an approach that allows for rapid iteration, learning, and refinement. Projects that take this approach have the greatest chance of leading to successful deployments, as they can be scaled gradually, building organizational confidence and demonstrating tangible ROI with each step.
Critical to this process is being able to demonstrate from the first step that the model fulfills its purpose and makes progress toward the desired business outcomes. Scaling this effort requires focused change management and the understanding that people – those who design, build, deploy, use, and govern the AI system – are critical to any scaled AI solution. Driving wide organizational buy-in and, ultimately, adoption requires all business functions to be clear on the benefit and to see the data demonstrate that the model is meeting its promise – and will continue to do so while being scaled.
Prioritize Responsibility
Finally – but certainly not “last” – prioritize “responsible AI,” which includes elements of security, transparency, explainability, ethical use, diversity and inclusion, accessibility, and bias detection. At my company, we bring this to life via our responsible AI committees, which provide guidance and oversight across all these areas, promoting scalable AI projects.
By keeping the six critical elements outlined above in mind, organizations can successfully move beyond the hype cycle and unlock the true potential of AI to drive meaningful business outcomes.