Enterprise AI has reached a turning point. The technology is robust, accessible, and increasingly embedded across the organization. Yet, for many companies, meaningful returns remain elusive—not because AI doesn’t work, but because it struggles at the last mile.

The AI last mile is where models meet reality: complex systems, imperfect data, human judgment, and real customer consequences. And few environments expose that gap more clearly than enterprise support.

Support is central to customer trust, recurring revenue, and product complexity. It is where AI must be accurate, explainable, and actionable—not just impressive. That makes enterprise support one of the clearest lenses for understanding AI Economics™, and how AI actually creates (or fails to create) durable business value.

This dynamic is at the heart of a recent TECHtonic episode, The AI Last Mile: How AptEdge Is Redefining Enterprise Support, where TSIA Executive Director Thomas Lah spoke with Kusal De Silva, CEO of AptEdge, and Aakrit Prasad, AptEdge co-founder, about what it really takes to deploy AI in complex B2B support environments—and what that reveals about pricing, services, and long-term value realization.

Key Takeaways

• Enterprise support reveals whether AI can operate effectively inside real-world complexity.
• Speed and deflection matter less than accuracy, resolution quality, and productivity in high-stakes environments.
• AI is reshaping pricing and services models, making ongoing value realization central to profitability.

‍

Why Enterprise Support Is the AI Last Mile

In many enterprise functions, AI can deliver incremental value even when it is imperfect. Support is different. Here, approximation turns into risk.

B2B support teams operate across sprawling product portfolios, frequent releases, and customer environments that are rarely standardized. Small changes can cascade quickly. Prasad illustrated this reality during the TECHtonic conversation, when he described a moment from his experience scaling AppDynamics: “We changed something in the user experience of the product. Then the next morning, we got 5,000 tickets.”

That example captures why support is such a powerful stress test for AI. When AI lacks context—whether about the product, the customer environment, or the downstream impact of a recommendation—it doesn’t fail quietly. It creates noise, escalations, and customer distrust at scale.

This is why enterprise support surfaces the true economics of AI faster than most functions. It forces leaders to move beyond experimentation and to ask a more complex question: Can this AI be trusted in contexts where the business is most exposed?

Related: The AI Last Mile: How AptEdge Is Redefining Enterprise Support

Why Context and Action Matter More Than Speed

Much of the first wave of AI investment in support focused on efficiency. Faster responses. Fewer tickets. Higher deflection rates. Those metrics are easy to measure, but they tell an incomplete story in complex environments.

In enterprise support, the cost of a fast but incorrect response can exceed the cost of waiting for the right one. That’s why De Silva emphasized a different priority during the episode: “We're not trying to be the fastest answer here. We're not trying to be another chatbot here. What we are trying to get to is the right answer.”

This distinction matters economically. Accuracy reduces rework, repeat tickets, and escalations. It protects customer trust and shortens time to true resolution, not just time to first response. Over time, this shifts support from a cost-control function to a contributor to retention and expansion. This is a fundamental shift: value is created by reliability rather than volume.

Why Data Quality Is No Longer the Real Constraint

For years, data readiness has been the most common reason enterprises delay AI adoption. While data quality remains essential, the discussion highlighted how much the landscape has changed. Modern AI systems are increasingly capable of working with unstructured and inconsistent data. As De Silva put it: “Data has been the scapegoat… But the reality is that if you have the right systems… we're fine with your data being messy.”

What actually slows adoption now is trust. Leaders hesitate to put AI into customer-facing workflows until they are confident it can handle edge cases and complex scenarios without creating risk. This reframes the problem. The question is no longer whether the data is perfect, but whether the AI can earn trust in the most demanding parts of the business.

How Trust Is Actually Earned in Enterprise Support

In enterprise support organizations, credibility flows from the most experienced engineers. If they don’t trust the system, no one else will. That’s why AptEdge focuses early deployments on earning buy-in from senior engineers who understand the edge cases and failure modes better than anyone else.

De Silva explains the importance of that validation step: “If we can get their trust and benchmark against those accuracy rates that they would deem as appropriate… It’s an incredibly integral part of the tuning process.”

Prasad reinforces who these users really are: “We call them support engineers, not support agents… they're really engineers, they're solving really complex problems.”

This is the operational reality of the AI last mile. AI becomes valuable only after it’s been tuned, validated, and accepted by the people closest to the work.

AI Is Changing Support Work, Not Eliminating Its Importance

The episode also addresses a tension many support leaders feel: the promise of AI augmentation versus headlines about workforce reduction.

Lah puts the question directly on the table: “Does the concept of just level one support just go away?” The answer isn’t a simple yes or no. Instead, the work itself shifts upward.

Prasad reframes how roles evolve: “The way we define level one today might be very different than what it means 10 years from now.”

Lah adds urgency for anyone waiting on the sidelines: “If you are not leaning into leveraging AI… you're not going to be relevant to the company.” From an AI Economics perspective, this is critical. AI increases productivity, but it also raises expectations. Support teams that adapt become more strategic, not less, because they can deliver higher-quality interactions at scale.

Pricing AI Remains Unsettled, But the Direction Is Clear

Few topics generate more uncertainty than AI pricing. Everyone agrees that seat-based models don’t fit the AI era, yet outcome-based pricing is still challenging to operationalize. Lah put the problem plainly: “User-based pricing is just not going to work.”

De Silva captures the challenge with a question many leaders are quietly asking: “How do you invoice a robot, basically?” 

The discussion lands on a pragmatic middle ground. Outcome-based pricing is compelling, but enterprise buyers still need predictability for budgeting, renewals, and expansion. That’s why hybrid models are emerging as a bridge. This pricing tension is a central theme of AI Economics™—and one that the industry is still actively working through.

Related: Outcome-Oriented AI Services

Why AI Expands the Role of Services

If AI requires tuning, validation, and continuous optimization, services don’t disappear. They become more critical.

De Silva offered a simple way to distinguish where services remain essential: “If it scales linearly with people, then it's a service… If it scales with applicable systems and data… that's software.”

Lah connects this directly to TSIA’s core position: “AI is not the end of services. It is the Era of Services.” From an economic standpoint, this is where profitability lives. Ongoing value realization—keeping AI aligned with changing workflows, data, and customer needs—creates durable expansion opportunities that pure software models can’t deliver alone.

‍

Related: AI Pricing Models Supercharge the Drive to Value Realization

Why the AI Last Mile Is Where Profitability Is Decided

Enterprise support strips AI down to its fundamentals. It reveals whether AI can earn trust, handle complexity, and deliver meaningful outcomes. This is the AI last mile. Solving it is not about more automation—it’s about accuracy, context, services, and continuous value realization.

For a deeper look at how these dynamics play out in real-world support environments, listen to the full TECHtonic episode featuring Thomas Lah, Kusal De Silva, and Aakrit Prasad. And for leaders ready to move from experimentation to profitability, TSIA’s AI Economics™ research provides the frameworks needed to turn last-mile complexity into a durable advantage.

FAQs

What is the AI last mile?

The AI last mile is the challenge of deploying AI effectively in complex, real-world environments where accuracy, trust, and outcomes determine value.

Why is enterprise support such a strong test case for AI?

Support combines technical complexity, customer impact, and revenue risk, making it one of the most demanding environments for AI to prove real value.

How is AI changing pricing and services models?

AI is breaking seat-based pricing and increasing the need for hybrid and outcome-aligned models, while expanding the role of services in ongoing value realization.

‍