Why enterprise AI depends on better conversations

Enterprise AI is entering a new phase. Over the past few years, organisations have invested heavily in foundation models, copilots and autonomous agents, and many are now moving beyond experimentation to deploy AI across everyday business operations. Progress, however, remains uneven: Deloitte recently found that only 14% of organisations have deployable agentic AI solutions, and just 11% are actively running them in production.

 

As enterprises work to close that gap, technology leaders are discovering that enterprise AI creates the greatest value when participating in the key conversations that drive the business: those with customers.

 

When context flows seamlessly from one interaction to the next, customers, employees and AI all build on what came before. When context doesn’t flow, customers repeat themselves, employees waste time hunting across systems for answers, and AI enters every interaction at a disadvantage, only having a fraction of the story.

 

 

Context as a strategic asset

Organisations have spent years building customer databases and integrating enterprise applications. Those investments remain essential, but AI raises the importance of another enterprise asset: conversational context.

 

Context extends well beyond customer records. It includes the conversation so far, the decisions already made, and the information needed for every interaction to pick up where the last one left off. An AI agent can retrieve information, summarise previous interactions and automate routine work before handing the conversation to an employee when judgment or empathy is needed. Working from the same shared context, AI and human employees each contribute different strengths, allowing AI to handle routine coordination and details, making time for humans to focus on connection.

 

 

Orchestrating AI across the enterprise

Consider a bank customer who receives an alert about a potentially fraudulent transaction on their account while travelling abroad. An AI agent reaches out, authenticates the customer and gathers the necessary information. If the customer disputes the charge and additional verification is required, the conversation moves to a human specialist who already has the full interaction history rather than starting the questioning over. In parallel, backend systems complete the remaining steps needed to resolve the issue, updating the account, triggering a card reissue, and logging the case.

 

From the customer’s perspective, it is a single, uninterrupted conversation. What they don’t see are the multiple AI systems, employees and business applications that worked together to deliver that experience. Customers don’t think about which AI model resolved their issue. They remember whether the interaction was effortless and whether the problem was solved the first time.

 

Fraud detection is a customer-facing example that is readily understood, but some orchestration challenges show up inside the enterprise that are much tougher to untangle. Many organisations have decades of accumulated workflow logic embedded in legacy contact centre and interactive voice response (IVR) systems that predate today’s AI tooling entirely. Rebuilding that logic by hand to run on modern infrastructure is slow, expensive and risky, which is one reason large-scale system migrations so often stall or get abandoned midway.

 

AI can bridge this gap: one model can translate legacy workflow logic into plain language for human review, then a second model can use that reviewed logic to generate an equivalent workflow on modern infrastructure. This kind of human-checked AI pipeline preserves institutional knowledge instead of discarding it, and offers a useful modernisation pattern for any organisation sitting on legacy systems it can’t easily walk away from.

 

 

Building for flexibility

Most organisations are no longer evaluating a single AI solution. Different teams are adopting different technologies based on their needs, and today’s technology choices will inevitably evolve as new capabilities emerge. That makes interoperability and orchestration essential for the future. Better to build AI environments that can adapt over time rather than those that require a rebuild every time a better model becomes available.

 

Betting an entire AI strategy on a single provider limits flexibility in a market that’s evolving rapidly. A more durable strategy is a layered approach: build an orchestration and data foundation, so AI models become swappable components rather than the foundation itself.

 

Building inflexible AI capabilities also introduces long-term complexity. Earlier this year, Gartner predicted that more than 40% of agentic AI projects could be abandoned by 2027 because organisations struggle to demonstrate business value or manage implementation complexity. That reinforces the importance of approaches that connect AI across existing systems and workflows rather than treating each deployment as a standalone initiative.

 

 

Preparing for the next phase of enterprise AI

Technology leaders should evaluate whether their communications and data environment can preserve knowledge across channels, support collaboration between people and AI, and remain flexible as new models emerge.

 

Organisations that strengthen these foundations will be better positioned to adopt new AI capabilities as they emerge while delivering more consistent and high-quality customer and employee experiences. As enterprise AI continues to evolve, the greatest value will come from connecting conversations across people, AI systems and business workflows. Preserving conversational context throughout those customer interactions gives organisations the flexibility to adapt while continuing to deliver better outcomes where they matter most.

 

---

 

David Funck is Chief Technology Officer at Avaya

 

Main image courtesy of iStockPhoto.com and Akarapong Chairean