Some cracks are starting to emerge in the hype cycle surrounding generative artificial intelligence (GenAI). After reaching the “peak of inflated expectations” in August 2023, many companies have started to encounter challenges when it comes to extracting real, meaningful value from the technology. Now is the time to address those concerns head-on or face a long journey of losing and slowly rebuilding trust in tools that leverage GenAI.

To get a better feel for where the world is right now when it comes to overall perceptions of GenAI and specific concerns about its implementation in the professional setting, we recently surveyed 158 C-level executives, vice presidents and directors “engaged in strategy, technology and business process at the top 20 nonbank lenders, top 100 insurance carriers, and tier 1, 2 and 3 financial institutions.”

Perhaps not surprisingly, given all of the attention surrounding the power of GenAI to transform knowledge work, the vast majority of professionals are excited about its prospects. All told, 91% of respondents have launched GenAI solutions of some type over the past year. Specific business functions getting the most attention are product development (93%), customer care/experience (82%), human resources (82%) and corporate strategy (75%).

The GenAI revolution has a data problem.

Despite the widespread enthusiasm, however, many businesses are starting to confront challenges when it comes to extracting real-world value from these initiatives as they move from the pilot project phase to full-fledged enterprise solutions. Many companies are reporting a significant drop-off in the number of users of their GenAI tools, and others are suggesting that their GenAI pilot projects are not meeting their goals for return on investment.

In nearly every example, the root cause of the disconnect between the promise of GenAI and the reality of implementing it in a real-world business setting is a lack of seamless integration between AI-powered solutions and enterprise workflows. In fact, according to our survey, fewer than half of respondents said that any one of their AI-enabled business functions is integrated with at least one other business function. The reason for that is data silos.

Nearly three-quarters (74%) of respondents who have implemented AI pilot projects in their organizations said data silos have been the primary barrier to enterprise-wide AI integration. Among that group, 33% said data is siloed within each business function, and 41% said data is siloed in some business functions but shared among others. In both cases, the result is the same: The number one thing keeping GenAI initiatives from reaching their fullest potential inside large corporations is data.

AI results start with data strategy.

By now, it has become conventional wisdom that data is crucial for the accuracy of AI outputs and is the foundation upon which all successful AI is built. The more reliable the AI outputs, the more valuable the solutions are for end users.

While everyone seems to understand that concept, very few businesses have managed to do the data governance and data management work necessary to enable the free flow of data across business functions and software platforms that make it possible for GenAI to synthesize large swaths of information and support companywide initiatives.

This process will not automatically happen when companies blindly purchase an off-the-shelf chatbot technology and set it loose on customers. It requires a great deal of planning and internal choreography between customer-facing personnel, data and analytics teams, and software development teams to bring together all of the disparate data assets necessary to deliver a complete, 360-degree view of the customer that can be used to train the system.

Take, for example, customer service copilot tools, which have been one of the most common GenAI use cases that have been showing promise in the insurance and financial services sectors. These solutions, which deploy a GenAI-powered agent-assist technology to monitor live customer support interactions and simultaneously search through the company knowledgebase to offer real-time guidance to customer service personnel, are helping to cut down on time spent searching for answers and improve the overall customer experience. A simple example is when a customer calls in to file an insurance claim and the AI copilot instantly pulls up that customer’s full claim history, details on other bundled policies, information on local repair resources and previous records of customer service interactions.

In the old world of insurance claims, each of these pieces of information would have been housed in different systems with different levels of administrative access and different formats. As a result, the customer would often be put on hold, handed off to multiple representatives and forced to repeat everything each step of the way. GenAI-powered copilots have the power to remove that barrier, putting all of the relevant information at the fingertips of a representative in real time.

However, it only works when the data backbone behind the scenes can support that level of seamless information transfer. Getting to that point requires a focused effort and close collaboration to link all of these disparate datasets into a centralized technology architecture. Without that, the GenAI copilot will be no more effective than a human customer service representative who needs to transfer a caller to another department.

For companies that understand the critical role that data management plays in making the promise of AI a reality, the efficiency and customer experience gains they’re seeing are immediate. For those that rush to build GenAI tools on top of a broken data foundation, disillusionment awaits.

LeackStat 2024