Is ‘Almost’ Ever Good Enough? The Customer Experience Debate in the Age of AI

When contemplating customer service and experience, the phrase “almost good enough” might seem trivial or even dismissive. But in today’s fast-paced tech landscape, particularly with the emergence of generative AI, this debate has gained serious traction. Is it acceptable for our support systems to be 80-90% accurate, especially when they’re not human?

The Role of Generative AI in Customer Service

Generative AI has emerged as a game-changer in the customer support arena, promising quicker responses, cost efficiency, and around-the-clock service. Yet, as enterprise customer experience (CX) leaders dive deeper into its capabilities, they’re realizing that it might not be the silver bullet they hoped for.

Gintautas Miliauskas, CEO and Co-Founder of Mavenoid, points out a key distinction: “If you have an 80-90% accuracy rate on generative AI, companies are kind of okay with that as long as they have escalation channels into human support.” However, the narrative shifts dramatically when we’re discussing human agents; the margin for error suddenly feels substantial.

This discrepancy highlights the complexities of deploying AI in CX settings. While generative AI offers an innovative response model, the question arises: does “almost” suffice in the realm of customer support where precision is often critical?

Why Generative AI Scales But Doesn’t Always Deliver

Generative AI undoubtedly possesses flexibility. It can engage in conversational flows and handle open-ended queries seamlessly. Yet, as Gintautas points out, there are limitations—especially with edge cases or complex problem-solving scenarios. This inherent unpredictability poses real risks for brands, especially in safety-critical sectors.

Mistakes in this realm can equate to financial losses, compliance violations, or, in extreme cases, endanger lives. “There are just some circumstances that purely require a business or a brand to know what it’s going to say every time,” he explains. Thus, many organizations find themselves hesitant to rely solely on generative AI, fearing the repercussions of miscommunication.

Another limitation to consider is user experience. Unlike professional prompt engineers, most customers lack the expertise to craft specific queries. Simple statements like “doesn’t work” can derail entirely, leading to frustration and ineffective support trajectories.

The Case for Deterministic AI

Contrasting generative AI is deterministic or curated AI, which emphasizes controlled, pre-defined responses. This approach serves well for high-stakes interactions—think safety guidelines, troubleshooting instructions, or compliance-related queries. Yet, deterministic systems also present challenges, particularly regarding scalability.

Gintautas notes: “Every new issue requires content creation and ongoing maintenance.” As a result, these systems can struggle to cover long-tail or multi-topic inquiries, demanding ongoing investment from support teams to keep content fresh and relevant.

The Trade-Off: Accuracy vs. Scalability

Historically, the choice for CX leaders has felt binary: deterministic or probabilistic AI? Gintautas calls this a false dichotomy. “Many people fall into the trap of conflating generative AI with conversational AI,” he states. Instead, the focus should be on resolution effectiveness rather than just conversational capabilities.

In complex environments—especially those involving physical products—both accuracy and adaptability are essential. Take multi-step troubleshooting, for example. Customers benefit from clear instructions combined with visual aids, which deterministic systems may provide. The key, Gintautas suggests, lies in blending the two.

The Hybrid Approach: Merging Strengths

At Mavenoid, the goal is to fuse deterministic precision with the flexibility of generative AI. This hybrid approach allows organizations to provide high-confidence responses in critical areas, while generative AI can efficiently handle less significant inquiries.

For instance, Gintautas describes how, with every new product, there’s a corresponding manual. “From generative AI, you can upload that manual and instantaneously address known queries,” he elaborates. However, by curating the experience with images and optimizing step sequences, brands can ensure that queries are resolved effectively.

This dual model not only enhances scalability without compromising accuracy but also fosters customer trust. Continuous improvement emerges from analytics, ensuring that the system adapts to evolving customer needs. Gintautas advocates for gathering extensive feedback—escalation rates, resolution rates, and impact metrics—to make informed adjustments over time.

A Problem-First Approach for CX Leaders

For those navigating the AI landscape, Gintautas underscores the importance of clarity: “Look at the top contact drivers in your business and what it takes to resolve those queries.” This involves selecting the right tools—whether text, images, or actionable steps—to facilitate effective resolutions.

The focus should shift from merely adopting the most conversational AI to implementing a problem-first strategy. By prioritizing accuracy, safety, and adaptability, enterprises can move beyond the notion of “almost” being acceptable and strive for excellence in customer service.

Gintautas emphasizes that deterministic AI offers the reliability and control enterprises require, whereas generative AI introduces necessary flexibility. The era of simple, one-size-fits-all solutions in the AI domain is fading; the need to blend these approaches is stronger than ever. Understanding how to balance deterministic and probabilistic AI will empower CX leaders to deliver reliable, scalable, and trustworthy support, creating significant advantages in a landscape defined by rising expectations and profound consequences.