AI in Customer Service seems to be all anyone can talk about. But when it comes to talking about what’s not working, most organizations have been suspiciously quiet.
The reason behind that silence was revealed a few weeks ago when a report from MIT unearthed a stark truth: despite billions in investment, 95% of enterprise AI pilots are failing to deliver meaningful business impact. The MIT NANDA report found that only 5% of pilots lead to rapid revenue acceleration, yet the organizations getting it right are seeing transformational results. (If you missed it, one of our data scientists did a breakdown of the State of AI in Business 2025 report here.)
Against this backdrop, Customer Service stands as both the biggest opportunity and perhaps the most brutal reality check for AI deployment. Here’s a deeper look at what separates the winners using AI in customer service from the wishful thinkers, and why your customer experience strategy needs to evolve, fast.
The Breakthrough Zone: Where AI in Customer Service Actually Delivers
Smart customer service teams aren’t just adopting AI – they’re using it as a catalyst for change and growth. The most successful deployments are rewriting the playbook on what great support looks like:
- Smart Routing Revolution: 52% adoption rate isn’t just impressive. It’s game-changing. Companies are eliminating the dreaded customer ping-pong effect, where unresolved customer issues just bounce from channel to channel, with some reporting 40% improvements in first-contact resolution. When customers reach the right agent with the right context on the first try, magic happens.
- Virtual Agent Dominance: Nearly 60% of enterprises now deploy virtual agents, but the winners aren’t just automating – they’re orchestrating. Advanced AI agents can now converse with customers and execute complex multi-step workflows, from processing payments to fraud checks to shipping coordination. We’re talking about AI that doesn’t just chat, it acts.
- The Domain Expertise Edge: Here’s where the plot thickens. According to the MIT report, AI providers that focused on a particular pain point or workflow saw greater success than more generic Generative AI options. As they put it, “The standout performers are not those building general-purpose tools, but those embedding themselves inside workflows, adapting to context, and scaling from narrow but high-value footholds.”
Likewise, that same study found that 90% of employees regularly use generic, consumer-grade AI tools like ChatGPT, but incorporate them for specific parts of their workflows, demonstrating that combining AI with domain-specific knowledge can create successful outcomes, even as larger enterprise pilots flounder.
The main takeaway is that focused application of AI produces the clearest and highest likelihood of success. And that focused application requires you to have a deep understanding of your processes and where AI can create measurable value in the shortest amount of time. So, whether you start with basic use cases like smarter routing or more complex use cases like AI agents and workflow improvement, domain knowledge must come first. After all, if you don’t understand how to improve your customer experience operation, AI won’t either.
The Danger Zone: Where Good Intentions Go Wrong
The flip side is brutal, and the data doesn’t lie:
- Trust Deficit Crisis: In the MIT report, “model output quality concerns” was ranked as the second highest barrier to scaling enterprise AI, only after “unwillingness to adopt new tools”. But both of these speak to a fundamental truth. That enterprise workers don’t trust AI outputs. And they don’t trust AI systems to make their lives easier. Because they are the ones who have to pick up the pieces when things go wrong, like explaining to an angry customer why their “intelligent” system either completely misunderstood their problem or provided patently false information.
- Data Governance Chaos: Having clear data boundaries isn’t just “the right thing to do” for your customers. Gartner predicts that, “by 2027, 60% of organizations will fail to realize the expected value of their AI use cases due to incohesive ethical governance frameworks”. This means that taking shortcuts like mixing training data with sensitive customer information will come back to bite you. Privacy violations, compliance nightmares, and customer trust erosion are the inevitable outcomes.
- The Over-Automation Trap: Organizations who prioritized automation, rather than getting to the right outcomes, are finding that AI is not the magic wand they thought it would be. In the MIT report one respondent shared, “It’s useful the first week, but then it just repeats the same mistakes. Why would I use that?” Companies are discovering that speed without substance is just expensive frustration that isn’t worth the time or investment.
This is where Loris fundamentally differs from the AI systems driving these failures.
- CX Intelligence That Shows Its Work:While many Generative AI platforms operate as black boxes, Loris builds trust through complete transparency. Every conversation analysis comes with clear reasoning and specific evidence. For example, when Loris flags a conversation for customer confusion, teams see the customer messages that support that claim. Even with Ask Loris, we reference real conversations as supporting evidence so you know the analysis is based on real data and have the ability to drill in for additional context.
- Data Governance That Actually Protects:Loris gets ahead of potential compliance issues by automatically detecting and redacting personally identifiable information (PII) prior to analysis. This approach helps you preserve conversational context for meaningful insights while also meeting your information and data security requirements.
- Unified Quality Assurance Across Human and AI Agents: As organizations deploy both human agents and virtual agents, traditional QA can’t scale to keep up with the volume of interactions. Loris provides unified quality intelligence across your entire agent ecosystem with consistent evaluation frameworks for both human and AI performance, cross-channel conversation tracking, and AI agent optimization insights that create feedback loops between human and virtual agent improvements.
The Build vs. Buy Verdict: Case Closed
For years, enterprise teams debated building AI capabilities in-house. Data from the MIT study settles this argument definitively:
- 66% of successful AI deployments used external partners, citing faster time-to-value, superior model performance, and dramatically lower total cost of ownership
- Conversely, 33% of successful AI deployments came from internal development
- Pilots using strategic partnerships were almost twice as likely to be adopted by employees compared to internally built
In addition, a recent HP study found that organizations who took the DIY approach to their AI projects took anywhere from 12 to 24 months to reach production versus 3 to 9 months for externally bought solutions. Besides the upfront investment in these resources, this study presented another risk of taking too long with internal development: “In fast-moving industries where AI provides immediate competitive advantage, even a 6-month delay in deployment can result in significant market share loss.”
The Rapidly Closing Window for AI Decisions
Enterprise AI spend reached $13.8 billion in 2024 – a more than 6x increase from the prior year. But the smart money is going to proven platforms with domain expertise that can support that scales without sacrificing quality. And the window of opportunity is rapidly closing to partner up and start extracting more value out of AI. As one CIO from a financial service firm stated in the report, “Once we’ve invested time in training a system to understand our workflows, the switching costs become prohibitive.”
The lesson is crystal clear: internal AI expertise doesn’t automatically translate into deployment success, especially in high-volume, high-stakes customer environments. The value comes from deploying AI that is proven in the use cases that help you truly understand your customers, your agents, and your business outcomes – not just AI that sounds impressive in vendor demos.
With 95% of AI Projects Failing, You Must Rethink Your Approach
Agree with the theory, but need tangible steps to put it into practice? See how to operationalize this approach in our latest whitepaper.
Why Loris Makes AI Adoption Safer – and Smarter
The reason most AI deployments in customer service fail isn’t the AI itself. It’s the lack of alignment between what the AI model or models do and what the business needs. Loris bridges this gap by combining deep customer service expertise with a library of purpose-built AI models designed to deliver real value from day one.
Here’s how our most widely used models reduce the common risks of AI failure and deliver measurable impact:
- Sentiment That Understands Context: Most sentiment models treat “no” as a negative. Loris knows better. Our proprietary Customer Sentiment model have been proven on hundreds of millions of customer service interactions to capture five levels of sentiment, from very dissatisfied to very satisfied, and interpret them based on domain-specific context. Paired with our Sentiment Target Classification, which identifies whether negative emotion is directed at the agent, the company, or both, these models help teams not only understand how customers feel but also why. This drastically reduces misinterpretations and allows leaders to focus on the root cause of dissatisfaction instead of chasing surface-level complaints.
- Conversation Quality at Scale: Our Conversation Quality (CQ) model acts as a predictive CSAT score for every interaction – no survey required. That means you don’t need to rely on the 5–10% of customers who actually fill out feedback forms. CQ analyzes the entire conversation, based on the conversation path, customer sentiment at the end, and presence of negative conversation markers, giving you consistent, automated insight into whether a customer felt heard, helped, and satisfied. This not only increases visibility but builds trust in the AI’s ability to reflect true customer experiences.
- Contact Drivers Tailored to Your Business: AI becomes irrelevant when it fails to understand your business. That’s why Loris builds custom Contact Driver models for each client. This takes all available conversations and uncovers the main topics in customer conversations and structures a hierarchy that maps to your products, policies, and team ownership. It then automatically categorizes every customer interaction with the reason customers are reaching out, revealing patterns in needs and concerns. These models are developed in weeks, not months, and allow you to unlock Voice of the Customer insights that connect directly to operational decisions, product roadmaps, and retention strategy.
- Transparent AI, Not a Black Box: Every model output in the Loris platform is evidence-based and easy to inspect. Whether you’re reviewing agent performance, tracking an emerging customer issue, or asking our Ask Loris feature for a trend analysis, we show you the real conversations behind the insight. This transparency builds internal confidence in the data and ensures that AI isn’t replacing judgment – it’s enhancing it.
- Domain-Trained and Continuously Improved: Many providers in this space rely solely on open-source models trained on generic internet data. Our models are trained and evaluated on real customer support conversations and improved through continuous feedback from human agents. That includes models for Polite Refusals (so we don’t mislabel “no thanks” as negative sentiment), FAQ extraction, confusion detection, and more. (Case in point, we can spot the difference between an upset customer using profanity and one who didn’t like the jerk chicken they ordered.) And we don’t stop improving after implementation – our data science team manages and fine-tunes models continuously, without requiring dedicated client-side resources.
The result? AI that understands your customers and scales your team’s impact because it’s already been proven on over half a billion interactions. That’s all unnecessary risk.
The Bottom Line: AI Isn’t Magic – It’s a Multiplier
AI won’t solve all your customer service problems overnight. But when applied with precision and purpose, it becomes an incredible force multiplier. The best CX teams are pairing human judgment with machine intelligence, using AI not to replace human insight but to give their teams superhuman capabilities.
AI agents will reshape demand for software platforms in 2025 and beyond, but the organizations that thrive will be those that choose specialization over generalization, transparency over black boxes, and business outcomes over technological novelty.
The customer service AI revolution is here. The question isn’t whether you’ll participate – it’s whether you’ll lead it or get left behind.
