Travel Tech’s Foundational Shift Towards Agentic (AI) Architecture

Graham Anderson
Oct 2
9 min read

The global accommodation marketplace, valued at approximately $1.2 trillion in annual revenue (Skift Research, 2025) is fragmented by design. Each part works efficiently: hotels optimise revenue; OTAs refine conversion; wholesalers reliably connect systems. Yet this localised efficiency creates a massive systemic burden - billions of daily transactions generating huge volumes of replicated data.

This is not inefficiency; it is the rational outcome of thousands of independent operators querying each other because no one controls the whole stack. The underlying architecture of this ecosystem continues to work. However, the rise of agentic AI will fundamentally change the economics of how queries flow through this proven infrastructure.

Precision Becomes the New Default

When AI agents search on behalf of travellers, they do not want hundreds of hotel results across multiple pages. They want 2-3 verified options representing meaningful trade-offs. This shift from broadcast queries ("show me everything") to precision requests ("show me only verified matches") has profound implications.

Current Model: The Platform Broadcast

For illustrative purposes:

Consumer facing OTAs each process millions of consumer searches daily.
Each consumer search triggers 200-500 upstream B2B queries from the OTA (accommodation provider to the OTA may connect to multiple more levels to source inventory).
B2B Platforms at each level accumulate data, control interfaces, optimise for their conversion.

The B2B volume dwarfs consumer OTA volume because of the multiplication effect, one consumer search query generates hundreds of infrastructure queries as each layer independently validates availability and pricing.

Emerging Model: Agent Precision

In a potential optimised distribution flow:

Agents query with detailed, contextual (and potentially multi-modal) requirements.
3-10 targeted queries replace 200-500 broadcast searches. There is an expectation that the upstream connection will bring back specific / close matching options to meet personalised request.
Travellers own comprehensive preference data through their agents.
Result: reduction in search volume; higher conversion rates.

The industry's fragmentation persists, agents simply navigate it far more efficiently.

The Conversational Traffic Reality

The narrative around agentic efficiency assumes a simple equation, fewer searches equals lower infrastructure costs. The reality is more nuanced.

What Actually Changes

Agentic systems do not eliminate traffic. They transform it from exploratory broadcasts to continuous conversations.

Current Traffic Pattern

Human searches on the way to work, at lunch, during evening browsing sessions and at weekends.
Burst load, heavy queries returning hundreds of results.
Idle periods, minimal traffic between search sessions.
Infrastructure optimised for peak human browsing hours and leverages cloud elasticity, spot instances, etc.

Agentic Traffic Pattern:

Agents operate 24/7, continuously monitoring on behalf of travellers.
Constant load, lightweight verification queries, status checks, monitoring.
No idle periods as agents never sleep.
Infrastructure must handle sustained conversational exchanges, not likely to be burst searches.

The Micro-Conversation Reality

A single booking might involve dozens of verification exchanges

Pre-Booking:

"Show me verified noise level data for this property"
"What's the actual upgrade likelihood based on my status?"
"Confirm this price includes all fees"
"Verify real-time availability before I commit"

Ongoing Monitoring:

Continuous checks: "Has reliability score changed?"
"Are there better options now available?"

Post-Booking:

"Confirm room assignment received"
"Monitor for service issues at this property"
"Track refund processing status"

What This Means for Infrastructure

The efficiency gain is not in total query volume, it is in query value.

From:

Wasted exploratory searches (show me everything).
High bandwidth, low conversion value.
Concentrated during human browsing peaks.

To:

Targeted verification and monitoring.
Lower bandwidth per query, but higher frequency.
Distributed across 24/7 as agents continuously validate.
Much higher value per query (conversion rates improvements expected).

The Cost Structure Inverts

Current platforms pay for:

Massive compute during search bursts or to sustain the overall volumes of searches per day.
Idle infrastructure between sessions if fixed size infra implemented.
Low conversion ROI on most queries.

Agentic platforms pay for:

Constant lightweight compute across 24/7.
Always-on monitoring infrastructure.
Higher conversion ROI per query.

Total infrastructure costs might stay similar, but the value extracted per dollar spent increases dramatically because queries serve actual booking intent rather than exploratory browsing.

What This Means in Practical Terms

Do not start planning for "100x traffic reduction", that's misleading. Plan for infrastructure transformation:

From burst-capacity search infrastructure to sustained conversational APIs.
From idle-heavy compute to always-on lightweight verification.
From broad exploratory queries to precise validation requests.
From low-value traffic to high-conversion interactions.

The platforms that redesign infrastructure for conversational patterns, not just reduced search volume, will capture the actual efficiency gains whilst maintaining service reliability.

Data Ownership Inverts

Here what changes fundamentally, data ownership inverts.

The Current State

Platforms accumulate user behaviour data across millions of travellers.
They infer preferences from fragmented observations.
Hotels see booking outcomes but not decision-making context.
Travellers control nothing.

The Agentic System

Comprehensive preference data lives in personal agents (the "digital twin").
Queries contain requirements, not personal identity.
Suppliers never see who is asking (until booking confirmation), just what's required.
Better personalisation with better privacy.

This creates a paradox, i.e. the model providing superior personalisation also provides superior privacy because personal data never leaves the agent. Only anonymous requirement specifications get transmitted until a booking is made.

Specialised Intelligence The Mixture-of-Experts Architecture

The future is not a monolithic AI. Instead, there will be contextual specialists, a mixture of experts, operate at each layer exchanging data, interpreting intent and building deterministic actions.

Personal Agent Specialists

Evolving examples may be:

Preference Interpreter. Translates vague desires ("a cosy spot") into precise requirements.
Loyalty Optimiser. Calculates real-time value across all loyalty programs.
Trust Network Analyst. Weights reviews from travellers with aligned preferences.
Contextual Adapter. Provides context of travel needs and supporting data points e.g. business trip vs vacation requirements.

Distribution Intelligence Specialists

These will be at the platform level interacting with the Traveller’s agents and the Supplier’s agents:

Verification Agent Tests supplier claims against reality.
Pricing Analyst Identifies "too good to be true" patterns.
Availability Validator Scores booking confirmation reliability.
Risk Manager Monitors supplier financial stability.
Arbitrage Agent Constantly working for optimised price/margin for maximising profit.
Fraud Prevention Agent Provides levers and controls to manage security and fraud prevention at the platform level for each micro transaction.

Supplier Intelligence Specialists

These will interact with platforms and also directly with the Traveller’s agent:

Capability Assessor Verifies data about actual service delivery.
Demand Predictor Identifies which agent requirements drive bookings.
Reputation Manager Converts operational excellence into algorithmic preference.

This distributed intelligence efficiently navigates permanent fragmentation while allowing each operator to maintain independence.

Operational Excellence Becomes the Algorithmic Moat

This transformation builds on decades of proven operational infrastructure.

What Stays:

Hotels own inventory independently.
Regional specialists provide local expertise.
Regulatory compliance remains jurisdiction-specific.
Transaction integrity requires distributed coordination.

What Changes:

Query patterns Precision vs broadcast.
Data ownership Traveller-controlled vs platform-accumulated.
Competition basis Verifiable excellence vs marketing claims
Platform role Infrastructure provider vs consumer gatekeeper

The "boring" operational excellence, i.e. guaranteed availability accuracy, transparent pricing, reliable booking confirmation, settlement between parties, etc. becomes the algorithmic moat. Marketing fluff becomes poison when agents verify every claim.

Transaction Integrity Remains the Foundational Moat

The AI efficiency hype overlooks the fact that distributed coordination is genuinely hard.

Current platforms own entire transactions. One point of failure equals one point of accountability. When bookings fail, you know who to call. Agentic systems distribute coordination creating multiple micro-failure points at a massive scale.

The scenarios that matter will not change:

Medical emergency requires immediate cancellation and rebooking.
Hotel claims no reservation exists despite confirmed booking.
Payment processed but inventory unavailable.
Which system's record is authoritative when they conflict?

The architecture requirements:

Clear accountability models Determine who is liable at each step.
Compensating transaction protocols Automated rollback when multi-step bookings fail.
Human escalation paths The phone still rings at 2am.
Financial reserves Agents (Travellers/Platforms/Suppliers) must guarantee bookings they facilitate.

The companies solving transaction integrity whilst achieving efficiency gains will capture disproportionate value. Those achieving only efficiency will discover that adoption stalls without proven booking reliability under stress.

Balancing Optimisation with Discovery

Pure optimisation risks predictable boredom. An agent always booking the "safe" choice eliminates the discovery of a beloved boutique property.

The solution:

Discovery as a Parameter The user must be able to adjust their preference for optimisation versus exploration.
Curated Inspiration Presenting optimised "safe" choices alongside one "inspirational" choice.
Feedback loops Learning when novelty adds value.
Context-aware recommendations Exploration vacation vs routine business trip.

Suppliers differentiated by genuine experience quality, not just operational reliability, maintain value even in algorithmic markets.

Three Strategic Paths Forward

Platform Resistance (Short Lifespan)

Major OTAs attempt to block agents via restrictive APIs. This fails because suppliers embracing agent-friendly interfaces gain algorithmic preference.

Hybrid Transition (Most Likely)

Dual models coexist: broadcast for humans, precision for agents. Winners build the best verification infrastructure and transaction integrity guarantees.

Complete Inversion (Possibly)

Agent-mediated booking is dominant. Platforms become infrastructure providers guaranteeing integrity. Suppliers compete on verifiable operational excellence.

The Realistic Timeline

The technology arguably exists today. The economic pressure on companies to have AI somewhere in their strategy is real. Claiming certainty about when mainstream adoption happens would be dishonest.

What We Actually Know

AI agents capable of complex travel booking exist now.
70% of travellers express interest in the concept.
Only 2% trust AI enough to grant full booking autonomy.
That is a 35x trust gap that will not close quickly.

2025-2027: The Learning Phase

This is not necessarily when positions get established, it is when the industry discovers which challenges actually matter:

Early experimental implementations reveal technical gaps.
Platform resistance strategies become visible (API throttling, legal challenges).
First transaction integrity failures teach expensive lessons.
Some suppliers build agent-ready interfaces and learn what works.
Adoption might reach 5-10% among early adopters in constrained use cases.

Companies learning fastest during this experimental phase - understanding what breaks, what scales, what users actually trust - position themselves for what comes next.

2027-2030: Infrastructure Emergence (Maybe)

If the experimentation phase proves viable economics and solves core trust problems:

Industry standards begin forming (probably painfully, through market failures).
Transaction integrity protocols mature through iteration.
Legal frameworks clarify through case law and regulation.
Platform business models either adapt or face genuine disruption.
Adoption might reach 15-25% as mainstream travellers consider agent-mediated booking for routine trips.

2030-2035: Mainstream Transition (Possibly)

Agent-mediated booking could become normal for certain traveller segments and trip types. However. this assumes solving problems not encountered yet.

The Honest Assessment:

We do not know at this stage if this transition takes 5 years, 10 years or stalls at 20% adoption because distributed transaction integrity proves unsolvable at consumer-acceptable cost. What we do know is that there will be a technological evolution in the fulfilment of accommodation sales and distribution. We know that because, the technology capability exists now but moving from "technically possible" to "mainstream trusted" requires solving operational, legal and trust challenges that do not have clear solutions yet.

The Grounded Reality Check

Before planning revolution, acknowledge context:

Current tech stack is proven at massive scale This is a strength, not a weakness. The industry processes trillions of transactions reliably.
AI costs remain uncertain Current pricing may be subsidised. AI costs replacing traditional compute may not stay "cheap."
Capability plateaus are possible Current AI tech delivers remarkable results. Whether we are at an evolutionary pause or a breakthrough cusp, the operational transformation required remains identical.
Transaction integrity is genuinely complex Distributed system coordination does not solve itself. The companies investing here build sustainable advantage.

Build Your Algorithmic Moat Now

The question is not whether your organisation is ready. It is whether you are building the verification infrastructure, transaction integrity protocols and agent-friendly interfaces whilst your competitors wait for clarity. Your strategic response depends entirely on which role you play in this ecosystem:

If you are an OTA or Aggregator Platform

Your path requires transformation from consumer gatekeeper to agent infrastructure provider.

Transform your infrastructure Billions of broadcast searches become millions of verified, guaranteed transactions. This is your moat.
Invest in transaction integrity systems Build compensating protocols to make agent efficiency reliable.
Develop agent-friendly APIs Handle legitimate burst traffic without throttling.
Establish clear accountability models Define financial liability, service levels, escalation paths before agent-mediated bookings begin. Your guarantee becomes your value proposition.
Leverage your data verification capability You already validate supplier claims across thousands of properties. Position this as essential infrastructure agents need.

If you are a Supplier (Hotel, STR, Experience Provider)

Your opportunity is to reduce dependency on distribution intermediaries by becoming agent-ready. Note, your distribution partners are enhancing their solutions to meet the agentic conversation and they have the smarts to do this at the scale required. They will still have a very important role to play.

Audit data honesty Marketing fluff becomes poison. "Quiet room" needs documented noise levels; "likely upgrade" requires historical confirmation rates.
Invest in real-time accuracy Agents need guaranteed availability, not probabilistic estimates.
Develop agent-friendly interfaces Enable direct agent queries, reducing distribution costs / offset against your own increased technology investments to support.
Build verification infrastructure IoT sensors documenting service quality, transparent pricing with no hidden fees, real-time inventory accuracy.
Prepare accountability frameworks When agent bookings fail, who's responsible? Have clear answers before participating in agent networks.

If you are building Agent Platforms

You must solve the problems OTAs solved over decades: operational trust and reliability.

Transaction integrity before efficiency. You must match incumbent platforms' proven reliability whilst delivering efficiency gains.
Establish financial reserves Will you guarantee bookings you facilitate? This determines whether you're trusted infrastructure or an experiment.
Build comprehensive data portability Set out secure Digital twin infrastructure where travellers own complete travel history, preference and intent data.
Design for graceful degradation When coordination fails (and it will), can travellers still complete bookings manually?

The Strategic Reality

The competitive advantage is who makes distributed coordination reliable when the human cries.

Data will be relocating. From platform databases to personal agents. From aggregated inference to verified precision. From broadcast optimisation to personalised validation. From centralised accountability to distributed coordination, with all the operational complexity that entails.

The industry's fragmentation persists. What changes is how intelligence navigates that fragmentation on behalf of travellers. The companies building both the efficiency and the reliability that mainstream adoption requires will establish the standards everyone else must meet.

Treat what’s left of 2025 and through next year as learning investments, not revenue transformation. The companies that survive are not those who bet correctly on timing - they are those who build adaptive capacity to respond as the market reveals what actually works.