Achieving AI Success In 2024: A Blueprint For IT Leaders

Nearly two years after ChatGPT’s launch, we’ve seen a sort of haves and have-nots emerge in AI: Some organizations are getting tangible business value from the tech, and the rest are far behind. If you keep an eye on the big global enterprises, you’ll see how companies like Klarna are leveraging GenAI to great success—in their case, replacing the workload of 700 customer service agents. But these big players seem to be in the minority.

It’s not for lack of trying. Organizations across the board are interested in LLMs and other AI technologies. McKinsey’s recent survey found that 72% of organizations are leveraging AI in 2024. But according to data from Harvard Business Review, as many as 80% of AI projects fail. Companies are struggling to get a real business impact from AI.

Success stories like Klarna and others prove that AI can be tremendously beneficial to enterprises. The question is, how does everyone else harness the technology effectively and avoid falling into the 80% of projects that fail? The keys are to start small, clearly define goals and iterate quickly.

Identify a small but meaningful use case.

The first step is choosing the right use case for your AI initiative. The most promising use cases are often simply alleviating nagging challenges that have long troubled an organization. These are the problems that have persisted for years despite previous attempts to solve them, and they are typically aligned with overarching company priorities.

Here are some things to consider when selecting a use case:

• Identify high-impact areas. Focus on improving efficiency or effectiveness in known problem areas, where even small improvements are worthwhile.

• Secure data access. Ensure the availability of clean, relevant datasets as the foundation for your efforts. If data isn't ready, look to a different use case.

• Assess feasibility. Choose challenges that are feasible with current tech and resources. Balance ambition with practicality to avoid resource strain.

If organizations are looking for a systematic approach to prioritize use cases, they should consider employing a simple yet powerful tool—a "value vs. effort" matrix. This framework allows you to chart potential projects based on their anticipated value against the effort required, helping leaders quickly identify and focus on quick wins (high value, low effort) while steering clear of less rewarding endeavors.

Prioritize good data.

The proliferation of data (especially web data) over the past decade or so is the x-factor that’s made AI possible in the first place. Data is the lifeblood of AI, but it’s also the cause of a lot of AI failures. So it’s essential that companies ensure data is accurate, clean and relevant.

They should start with minimal viable data: That means gathering just enough data to effectively support a given use case. This boosts the efficiency of AI projects and speeds up the iteration cycle, allowing for quicker adaptability and feedback. When lacking organic data, it’s important to consider using AI-generated data or synthetic data sources, such as AI-generated text and images. This will broaden analytical capabilities and uncover more insights.

Overall, enterprises should challenge their preconceived notions about what data can be leveraged. AI systems can handle a broad variety of data assets, turning overlooked or underutilized “dark data” (for example, PDFs, presentations and reports) into valuable insights.

Agree on what success looks like.

It’s critical to define what success really looks like to make sure your efforts align with tangible outcomes. What matters to the business? How will you know if your AI initiative is working or not?

This begins with setting realistic goals. Organizations need specific, measurable goals that sync up with their wider business objectives and vision for success. How do you define these goals? It’s about metrics: Decide which key performance indicators (KPIs) will track your AI's effectiveness and its impact on your business. Select metrics that not only reflect the performance of the AI system but also how it boosts real-world business outcomes.

Enterprises need to set up a routine to monitor these KPIs closely and tweak their strategy when needed. Leaders can’t take a set-it-and-forget-it approach to AI. They have to be agile to ensure their project continues to meet its targets and enable them to adjust to new data or challenges.

Build, test and learn quickly with users.

Too many organizations get mired in isolated, lengthy development cycles when beginning an AI project, spending months and months building AI and never truly deploying it. Instead, they should work toward a quick prototype that they can share with users and then iterate from there.

Good feedback is essential—and getting it sooner can avoid countless wasted hours. Stakeholders need to deploy prototypes swiftly to end users so they can make tweaks accordingly. This iteration is needed to deliver a solution that meets the actual needs of its users.

It’s fundamental here to learn by doing. The purpose of AI isn’t theoretical; it’s to solve business problems. Organizations must be able to adjust course promptly based on early feedback.

Transparency is also key. Leaders will foster trust by sharing both successes and setbacks across the organization. As successes build, they can expand the scope of the given use case or transition to related use cases. This collaborative approach ensures continuous improvement and successful scaling.

Leverage AI strategically.

AI can seem intimidating for enterprises experimenting with it for the first time. However, any organization can deploy it effectively as long as it breaks down the process into manageable steps, focuses on strategic implementation and fosters an agile development environment.

The ultimate goal of AI is to augment human capabilities, transforming challenges into opportunities and creating new capabilities for your business that weren’t previously possible. Getting AI wrong can be costly. But in the long run, not leveraging AI at all will be even costlier.

Source: forbes.com