IBM’s Global AI Partner on Fixing CX Bots: The ‘Human Touch’ Is Just Good Data

As foundational LLM model quality improves and powerful agentic protocols emerge, the technical barriers to AI implementation are lowering, shifting internal conversations away from "What's possible?" toward "How do we execute?". In customer service settings, the best solutions center around an obsession over how to capture the benefits of AI automation without sacrificing the very human connection that defines great service. If done wrong, the pursuit of an intelligent solution leads to a profoundly unintelligent outcome. But when done right, customers benefit from the speed and scale of technology in harmony with hyper-personalization and hyper-specialization of knowledge that feels deeply human.

We spoke with Raj Kadiyala, Executive Partner for Global AI Innovation at IBM, who argued that the answer lies not in a more complex algorithm, but in a simpler truth most companies overlook. Kadiyala’s perspective is shaped by a career on the front lines of digital transformation, with leadership roles at consulting giants like Accenture and Deloitte and as the founder of his own successful tech consulting firm, Hextura. Now, through initiatives like his "AI Simplified" content channel, he’s on a mission to demystify the technology for everyone from students to executives.

• The human component: "I've seen the most customer service implementations fail when the data coming in is not good, hasn't been curated well, and is biased," Kadiyala said. "And surprisingly, the data is still the most human-reliant component."

For Kadiyala, the promise of AI in customer experience rests on three pillars: creating ease of use for the consumer, using data for predictive analytics, and building efficient application workflows. But Kadiyala also identified a trio of hurdles: the technology, the process, and the people that cause AI models to hallucinate, become too opinionated, and drift from their intended purpose. These failures are not just technical glitches; they are expensive business problems.

• AI's sweet spot: The good news, Kadiyala noted, is that the solution to this data problem lies in AI's current sweet spot. "A lot of AI capabilities today are actually in content and how you curate information from content in terms of process efficiencies, like supply chain or CRM," he said. The process for achieving this is a clear, repeatable lifecycle. "You have your data, you create the data sets, then you curate the data sets and put them into a vector database. Your AI models start pulling information from that. That's the lifecycle," he explained.

• Engineering empathy: This methodology is often powered by frameworks like RAG systems that help models pull from verified information sources. This technical process is the key to engineering what feels like a human touch. "AI models can be inherently built to show a sense of empathy," Kadiyala said. "You train them using data sets based on a pattern of questions. For instance, with bad hotel reviews, a human responds in a certain way. AI models can do that too, and you can tune them to have that sense of care and recognition."

This opportunity isn't just for large corporations; it extends to what he called "middle tier" smaller companies like law firms or real estate firms who "can't afford the billing rates" associated with enterprise solutions. In any setting, mastering data quality today is the ticket to leveraging the next wave of innovation. By focusing on this foundation, businesses aren't just chasing a futuristic ideal. They are investing in what Kadiyala said is one of the most successful, real-world applications of AI today.

• Beyond avatars: "The next 18 months will be defined by innovation in agentic AI." This future, he clarified, won't come from a single, all-in-one product. "One thing in the AI ecosystem is that there's not one product or one company that does it all. It's a matter of integrating different products," he said. "You can duplicate humans and create avatars, but that's a different product stream. You actually have how you get the data curated for the content, which is a different product ecosystem."

• Connective tissue: This integration of specialized, agentic systems is the ultimate payoff. It creates a "connective tissue" that allows high-quality customer service data to break out of its silo and inform other parts of the business, transforming a cost center into a proactive engine for strategic intelligence. "I would say that agentic AI is going to take on a lot of responsibilities for helping create workplace efficiency."