Human-in-the-loop servicing shouldn’t be the end goal for travel

Human-in-the-loop is quickly becoming the accepted model for complex travel servicing: artificial intelligence (AI) does the heavy lifting, but a person validates it, approves it and, in some cases, takes action.

Human-in-the-loop should be a transition toward full autonomy, not the end state. 

The starting point differs across travel. Many online travel agencies (OTAs) already want servicing automated end to end but remain skeptical that travel’s infrastructure allows high enough coverage and accuracy. For travel management companies (TMCs), human service remains more closely tied to the white-glove proposition. But travel doesn’t have to keep designing servicing around the belief that complex means human.

Because as long as every complex refund, exchange or disruption event still requires human review, travel will remain constrained by queues during disruptions, operating costs that grow with volume and a servicing experience that doesn’t match the speed and ease of booking. 

Previous automation hit an execution gap, so humans filled it

Behind every exchange or refund sits airline policies, fare rules, penalties, taxes, waiver codes and regional regulations. Add multiple passengers, partially used tickets, ancillaries, multiple forms of payment, virtual interlining and multiple content channels, and the number of possible paths multiplies quickly. 

The challenge exists in two areas. First, bringing together fragmented information from across different airlines and systems, stored in different formats and often written for humans rather than machines. Then executing that decision across equally fragmented airline, content and agency systems. 

Until recently, solving both reliably across every edge case, cost more than low-cost manual labor.

That is why humans have remained essential to complex servicing. They became the cheaper integration layer connecting information, rules and execution. 

Human-in-the-loop risks being the next execution gap

Human oversight will always matter when a traveler’s intent is unclear, they don’t know what they want, or empathy or judgment genuinely improves the outcome. But requiring human validation for every complex servicing decision risks creating a new and an unnecessary execution gap.

AI, combined with deep domain knowledge can find the information, interpret the rules and determine the right action in seconds. If that action then waits for a human to validate, approve or even execute it, the bottleneck has moved, not disappeared.

During a major disruption, thousands of decisions can be made instantly but only executed as quickly as people can validate them.

Complexity should no longer determine where humans are required. Human judgment, empathy and reassurance should.

Human-in-the-loop makes servicing more efficient. Autonomous execution is what makes it scalable. 

AI changes what’s possible. Domain expertise makes it work.

Recent advances in AI have solved part of the interpretation problem. They have not solved air servicing on their own.

Much of today’s automation relies on ATPCO Categories 31 and 33, which provide the structured rules for exchanges and refunds but only cover 40-50% of cases. Category 16 goes deeper, with free-text conditions that apply more broadly but have historically been difficult for machines to interpret. Large language models can now parse that information and combine it with booking context as part of an automated decision.

But an LLM alone does not close the execution gap.

Reaching the edge cases requires combining that deeper policy understanding with booking, ticket and segment context, tax refundability, airline waivers and regional regulations. And it requires connectivity across content sources, execution channels and agency systems.

It also needs experience. Air servicing is a problem of millions of edge cases. Every correctly resolved case adds to the understanding of how a particular combination of context, rules and conditions should be handled. At sufficient scale, those individual decisions become recognizable patterns.

At Wenrix, reaching the coverage required for autonomous execution has taken eight years of solving edge case after edge case, alongside verified real-time agent actions across $50 billion in real servicing data from more than 60 global OTAs and TMCs.

AI changes what machines can understand. Accumulated domain expertise turns that understanding into reliable execution. 

Autonomous execution has to earn the trust of the agency

Removing the human checkpoint raises the bar. Autonomous servicing has to deliver both accuracy and coverage: accuracy without coverage leaves the hard cases with people, while coverage without accuracy creates unacceptable risk.

The real benchmark is whether it can handle enough complex cases, accurately enough, that human validation becomes the exception rather than the default.

We are seeing that achieved in production. One leading global OTA is now autonomously executing more than 93% of servicing cases at 99.8% accuracy, including scenarios beyond those traditionally associated with automation.

The consequences of getting those decisions wrong are real. A refund calculated incorrectly by $20 can result in an agency debit memo (ADM). Technology providers therefore need to stand behind their execution. At Wenrix, we guarantee customers against ADMs caused by our automated technology.

Travelers are more ready for autonomy than we think

There is still an assumption in travel that human involvement means better service, particularly in managed travel where “white glove” has traditionally meant access to an agent.

But travelers increasingly judge service by the outcome. They want their problem resolved accurately, quickly and with minimal effort. Our research supports that shift: only 7% of travelers said they would never want servicing handled automatically, while 72% of managed travelers were comfortable with AI managing a rebooking during disruption.

For many servicing requests, autonomy can deliver the better experience: instant, frictionless resolution rather than waiting for human review.

White-glove service doesn't have to mean faster access to a human. It can mean not needing one in the first place. 

The automation gap will become a competitive gap

Many of the world's largest OTAs are already moving beyond human-in-the-loop toward autonomous execution. As they prove that complex servicing can be resolved instantly without human validation, that experience will increasingly shape traveler expectations.

That expectation will extend into managed travel. As instant, autonomous resolution becomes normal in leisure travel, waiting for human validation when traveling for work will increasingly feel like friction.

Travel has seen this pattern before. From online booking to NDC, innovation adopted by OTAs has eventually influenced TMCs. 

TMCs shouldn't simply replicate the OTA model. Corporate travel brings additional complexity, from corporate policy and approval workflows to duty of care. But they can build on what OTAs prove at scale, adapting autonomous technology to the requirements of corporate travel rather than starting from scratch. 

Autonomy changes more than the cost of servicing

Once human availability is no longer the constraint, servicing can become more than an operation to make it efficient. It changes what OTAs and TMCs can build around it.

During disruption, that could mean resolving problems proactively instead of waiting for thousands of travelers to make contact. Or being able to continue selling complex offers, without the worry it will create more work for customer care.

A servicing request can also signal new commercial intent. If a traveler moves their return flight two days later, that creates an opportunity to extend the hotel or car booking and make relevant offers when the need emerges.

Fast resolution can create loyalty too. Our research found that 83% of travelers who experienced disruption booked with the same supplier again when it was handled quickly and without friction.

Finally, there is operating leverage. When air volume can grow without servicing headcount and operating costs increasing proportionally, growth no longer requires businesses to build the servicing organization it previously demanded. 

The question travel should be asking

Complexity alone should no longer justify the need for a human.

As the execution gap closes, the question changes from:

“Where do we need a human in the loop?” to “Where does a human genuinely add value today and can AI deliver that outcome as well or better?”

When it can, people can move onto where they add greater value. This should be iterated until the end-goal is achieved. 

That continuous improvement enables you to ask a bigger question: What post-booking experience would you build if you no longer had to design around the human checkpoint?