Überblick
Artificial intelligence (AI) offers massive economic potential, yet businesses face challenges in harnessing its full power. The transition from AI proof of concept (POC) to production is fraught with difficulties, leading to issues like ineffective marketing and less-than-ideal customer experiences. Overcoming these challenges starts with identifying some of the biggest barriers to unlocking AI value.
Surmounting these barriers can lead to improved profits, productivity, and customer experiences, as shown by one Teradata customer who successfully scaled AI initiatives. Learn how this customer drove a 244% return on investment (ROI) over three years, saw a 3x rise in model development, and delivered close to $80 million in referral program profit.
AI is poised to transform business across industries in the coming years, with more than $25 trillion in total economic potential.1 From financial services and telecommunications to supply chain management and healthcare, AI can unlock valuable insights aimed at driving greater productivity and business value. Yet many business and analytics leaders still struggle with how to harness the full potential of AI.
Put simply, moving AI initiatives from POC to production isn’t easy. And the inability to operationalize and scale AI can lead to a range of negative downstream effects, from ineffective marketing engagement and less personalized customer experiences to supply chain friction.
In this article, we’ll take a closer look at the top three barriers to unlocking AI value, including establishing a solid data foundation to scale AI, removing bottlenecks in data scientist productivity, and overcoming data movement’s impact on model cost and security. And, we’ll offer insights from a commissioned 2024 Forrester Consulting Total Economic Impact™ study detailing how one current Teradata customer significantly scaled AI to drive profits, boost productivity, and optimize customer journeys.
Barrier 1: Creating a solid data foundation to scale AI
While AI is experiencing exponential growth, many enterprises fall short in launching AI and machine learning (ML) initiatives. In fact, over 40% of AI projects never make it from POC to production.2 Among the biggest reasons for this are spiraling data growth and data preparation inefficiencies that restrict model development.
Uncontrolled data growth
AI requires lots of data. To accommodate this, many organizations simply add to their tech stacks. This can lead to a tangled web of systems and increase technical debt, which can result in unmanageable data growth. To streamline AI development, enterprises need harmonized data that optimizes workload efficiency and prepares data faster.
Data preparation inefficiencies
One study shows that about 80% of AI project time is spent preparing data—not creating business value.3 Why? To train models, large amounts of data are often accessed and moved from different systems across an organization, from secure terminals to developers’ personal laptops. These inefficiencies can require more employees as the number of models increases. The result can be a “model plateau,” with organizations either limited to creating fewer models or forced to add new hires to increase development. In either case, this can hamper efforts to scale over the long-term.
Barrier 2: Productivity bottlenecks from model maintenance and governance
Once models are built, they need to be maintained and governed so they work optimally. But Teradata has found that enterprises and business leaders often underestimate how much time their teams need to devote to model maintenance. Maintaining version control, preventing model drift (where a model’s predictive ability can degrade over time), and updating feature banks all require employee time. And those costs increase as more models are built, which can negatively impact AI innovation.
For example, a data scientist may spend up to 10% of their time simply doing model maintenance. If that same employee builds 10 models, their schedule is fully booked. If valuable data scientists are dedicated to simply model maintenance, they’re prevented from focusing on new AI initiatives, optimizing the customer experience, and growing the business.
Barrier 3: Costly data movement
Harmonized, streamlined data is key to accelerating AI innovation. Storing information in separate areas not only strains storage budgets, but also requires that tools be built and maintained to allow systems to integrate. That drives more costs and complexity—and can increase security risk. This holds true for any organization that deals with sensitive customer or business information—like those in healthcare, financial services, or retail.
In the end, costly data movement can impact the speed of model development and deployment, slowing both AI insights and time to market for AI initiatives.
1. “The economic potential of generative AI: The next productivity frontier,” McKinsey, 2023.
2. “Gartner Identifies the Top Technology Trends for 2021,” Gartner, 2020/
3. “How Companies Can Use DataOps to Jump-Start Advanced Analytics,” McKinsey, 2021.