Jeff Fettes is the founder and CEO of Laivly.gettyEvery week, I talk to contact center leaders asking the same question: Which AI use case should we deploy next? However, that question won't lead them to the results they actually want. The brands pulling ahead in the new AI race view AI as a tool to help execute today's plans and get ahead of tomorrow's problems. They're building AI as infrastructure to allow quick and easy deployment, testing and iteration of future tech ideas, knowing that any specific use case in 2026 might have a short shelf life.​We surveyed 200 contact center leaders for our March 2026 "AI Deployment Index" and found that 65% of respondents called their most recent AI initiative successful. However, 43% reported projects being delayed or stalled, and 53% reported projects exceeding original budgets. Deployments are happening, but the results often aren't.​The gap comes down to framing. Deploying AI use case by use case made sense a few years ago, but now that approach works against you. The technology evolves faster than any deployment can keep pace with, and today's cutting-edge capability is tomorrow's baseline. At that speed, you can't start from scratch each time.​The AI Use Case Trap​The traditional model for contact center AI has been to identify a specific use case, build toward it, deploy and declare victory. It seems logical enough, but to use an analogy, when wiring a building, you don't run a dedicated circuit from the breaker panel every time you plug in a new lamp. You wire every room, wall and future outlet so that adding something new means plugging it in, not rewiring. The infrastructure exists before you know exactly what you'll need.​Building toward one outcome with no structure behind it often leaves you unable to adapt when it falls short. If the use case works, great—but you've built a point solution, not a foundation. If it doesn't work, which is more common than most will admit, there's nothing to fall back on. You're starting from scratch on the next attempt.​Contact center AI pilots fail at scale for a predictable reason: The small, engaged group you test with behaves very differently than a broad customer base. What works in controlled conditions often doesn't survive when exposed to real volume. If your entire AI strategy is built around making one thing work, a single failure point can set you back a year.​What AI Infrastructure Actually Means For CX​AI infrastructure in the contact center means the ability to insert AI into your workflow at any point and move, remove or replace it just as easily. If leadership identifies an opportunity to use a large language model (LLM) at a specific moment in an agent interaction, that capability can be activated quickly, tested honestly and adjusted based on real data.​This matters more in customer experience (CX) than almost anywhere else because the environment is uniquely constrained. An office worker can explore a new AI tool and decide for themselves if it's useful. An agent can't. Every tool has to be tested, vetted and built into the workflow before it ever reaches them, all while meeting compliance and consistency standards. The infrastructure has to support that with real security and change management at scale.​The contact center AI landscape is also moving faster than most organizations are built to absorb. For decades, large enterprises wore their caution as a competitive virtue: rigorous, secure, established. Today, by the time a use case clears procurement, pilots and full deployment, the underlying technology may have already moved on. The organizations playing catch-up will be the ones that never built the infrastructure to move faster.​​Why Contact Centers Still Default To Point SolutionsContact centers often still hesitate to adopt an infrastructure strategy because a specific use case is easier to explain, fund and measure. Asking for budget to reduce handle time or automate a particular workflow has a clear business case attached. "Build an AI foundation" sounds like a larger, more abstract commitment involving significant upfront cost, complex implementation, organizational disruption and unclear ROI.Dependency is also a legitimate concern. With technology changing so quickly, no organization wants to bet long-term on the wrong platform. However, good AI infrastructure should increase optionality, not reduce it. If adopting a platform makes it harder to change models, replace capabilities or adjust workflows, it's not really infrastructure—it's another form of lock-in.However, companies shouldn't have to fund an open-ended transformation and trust that value will eventually follow. In my experience, the best approach is starting with a real problem that can produce a measurable return and building that first deployment on reusable rails. The first use case should earn its keep while establishing the groundwork that makes additional use cases faster and less expensive.Providers also need to meet contact centers where they are: working with existing systems rather than requiring costly replatforming and giving operations teams the ability to make changes without IT resources. Commercial models should support incremental adoption, with providers transparent about portability, data control, implementation effort and the true cost of scaling.​The Right Question To Ask About The Value Of AI​Most leaders think about AI ROI in terms of specific outcomes: Did this chatbot deflect enough calls? Did this tool reduce handle time? Those are reasonable questions but the wrong ones to focus on.​Think about how you'd answer if someone asked whether your personal AI assistant delivers ROI. Most people can't quantify it precisely. They just know it makes them more capable. It's embedded in how they work—helping draft emails, time-blocking their schedules, keeping track of to-do lists. The value isn't a single use case but, rather, the accumulated capability.​That's the new goal for contact center AI: a foundation that makes every future deployment faster, cheaper and easier to adjust. The leaders who will get the most out of AI over the next few years are the ones who stop looking for the right use case and start treating AI as infrastructure.​Forbes Technology Council is an invitation-only community for world-class CIOs, CTOs and technology executives. Do I qualify?