ThailandOnline travel (OTA)Scale · $100K
Agoda

Millions of room-nights, dozens of languages, one 2 a.m. cancellation panic at a time: agentic traveler support for Agoda

Give Agoda an agentic layer for the booking-amendment and refund conversations that dominate its multilingual support cost — resolving in the traveler's language with live reservation and property context.

Entry use case
Booking amendment & cancellation resolution
Expected outcome
In-conversation date changes, cancellations, and refund-status resolution across top language pairs, measured against cost-per-contact and CSAT baselines
Recommended next step
Propose an A/B pilot on one language pair (e.g., Korean or Japanese amendments) with cost-per-resolved-case as the primary metric.
What we understand

Agoda's operating reality

Agoda is a Bangkok-headquartered online travel agency within Booking Holdings, strong in Asian accommodation supply, operating its platform and customer support in dozens of languages for travelers worldwide.

Public fact

Agoda has a famously engineering-driven, experimentation-heavy culture — decisions follow A/B evidence, and in-house build capability is substantial.

Public fact

Booking amendments, cancellations, and refund-status chasing dominate OTA contact volume, with hard language-coverage costs across 24/7 shifts.

Reasoned inference

Property-side coordination (confirming amendments with hotels) creates a second conversation leg that multiplies handle time per case.

Reasoned inference

An agentic layer that negotiates simple amendments directly with property systems or front desks could collapse the two-leg resolution loop.

Seller hypothesis — validate

Validate with the account team before outreach: In-house conversational-AI roadmap and where an external platform is welcome · Contact volume and cost by language pair · Property-extranet API depth for agent-executed amendments

Build vs buy

Why we have a right to win here

Co-build target

Strong internal engineering — sell infrastructure, models, governance, or selected capabilities

Evidence: Agoda's engineering-driven culture under Booking Holdings builds much of its own CS tooling and ML — it will never buy a turnkey stack. But frontier models, long-tail multilingual speech, and evaluation infrastructure are components it rationally procures and A/B tests against its own baselines; that is the co-build wedge.

Why they won't build the full stack: What Agoda won't build: carrier-grade telephony, dozens of languages of production voice, and frontier-model R&D — Booking Holdings' capital allocation favors experimenting on top of vendor models, not competing with them.

What management is signalling

Parent Booking Holdings has publicly emphasized generative-AI investment across its brands alongside continued margin discipline.

Reported factQ3 2025 earnings call · Nov 2025

GTM implication: Enter as an A/B-testable component — cost per resolved case on one language pair — inside a group already committed to AI-driven support economics.

What already exists (don't pitch this)
  • In-house customer-service platform and chatbots
  • Extensive experimentation (A/B) infrastructure
  • ML-driven personalization and pricing systems
  • 24/7 multilingual support centers
What customers still can't do end-to-end (pitch this)
  • →Frontier-model reasoning over rate conditions and amendment policies
  • →Agentic property-side outreach that collapses the two-leg resolution loop
  • →Production voice quality across long-tail language-hour combinations
  • →Step-change in cost per contact on low-volume language pairs
Opportunity map

Where agentic communications pays off first

WorkflowWhy it matters hereValueComplexitySpeedChannels
Amendment & cancellation resolution
Policy-accurate amendments executed in-channel in the traveler's language, with property confirmation handled agentically.
The single largest contact class; policy complexity (rate types, property rules) makes it expensive in every language.
Friction today: Agents parse rate conditions manually; travelers wait on hold at 2 a.m. local time in the wrong language.
In-app chatVoiceEmail
Refund status & payment disputes
Live refund-pipeline status pushed proactively; disputes triaged with grounded transaction context.
Refund limbo drives repeat contacts and chargebacks — pure cost with negative CSAT.
Friction today: Status spans acquirer, property, and OTA ledgers; agents read opaque states to anxious travelers.
In-app chatEmailVoice
In-stay issue resolution
Priority detection, immediate property outreach by the agent, rebooking alternatives offered live within policy.
A traveler standing at a front desk with a missing booking is the moment Agoda's brand is decided.
Friction today: Urgent cases queue behind routine ones; property phone-chasing is manual.
In-app chatVoice
Watch the change

Booking amendment & cancellation resolution: today vs the agentic model

Scenario: A Korean traveler needs to shift a Phuket booking by one day at 1 a.m.; the agent reads the rate conditions, executes the amendment, confirms with the property, and closes the case in Korean in four minutes.
Today
same interaction, two worlds
Agentic layer
First-contact resolution
deferred via tickets
in-conversation actions
Platform capability
Cost per contact
$13.50 median assisted
$1.84 median self-service
Benchmark
QA coverage
1–2% sampled
100% scored
Platform capability
Sources & assumptions
  • · First-contact resolution: Process design: agent acts in systems of record
  • · Cost per contact: Gartner customer service cost benchmarks, 2024
  • · QA coverage: Platform capability: every interaction logged and evaluated
  • · Gartner, customer service cost benchmarks (2024): $13.50 median assisted vs $1.84 self-service per contact
  • · McKinsey, digital-first collections research: 20–25% NPL reduction among leaders; up to 40% opex reduction with gen AI
  • · Baymard Institute: ~70% average cart abandonment (meta-analysis)
  • · IAMAI–Kantar via IBEF (2025): 900M+ Indian internet users; 98% consume Indic-language content
  • · LeadSquared and vendor funnel studies: 78% of students choose the first institution to respond (directional, vendor data)
  • · HDI / ITSM operator benchmarks: $15–25 per L1 ticket; 40–60% of L1 volume is resets/status (validate per customer)
  • · Conventional-flow wait times and volumes are typical operator patterns — assumptions to replace with the customer's own baseline
  • · Agentic-flow behaviors (context retention, 100% logging, in-line policy checks) are platform capabilities, not projections
Recommended solution

One integrated stack, opinionated for this account

Channels · Tilicho Labs
In-app chatVoiceEmail

Voice & channel orchestration, telephony, conversational execution, session/state, routing, integration build. Capability coverage validated during implementation.

Intelligence · Google Cloud
Gemini reasoningEnterprise groundingWorkflow agentsMulti-agent orchestrationGoverned actionsEvaluation & analytics
Systems · Agoda
Reservation platformProperty extranetPaymentsPayments pipelineProperty contactsInventory

API access to these systems is the critical-path dependency.

Trust & languages
EnglishThaiMandarinJapaneseKoreanBahasa IndonesiaVietnameseHindiArabic

Identity-bound sessions, policy-bounded actions, 100% audit logging, human approvals at defined points, in-tenant intelligence.

Business case

The economics, with your numbers

Addressable monthly interactions5.0M
Seller assumption — replace in discovery
Current cost per interaction ($)$1.8
Industry benchmark scale — validate
Automation / assistance rate (%)55%
Seller assumption — pilot proves this
$9.0M
Current operating cost / mo
$47.9M
Modelled gross benefit / yr
0.0 mo
Payback on Scale
413%
3-yr ROI (modelled)
Automated/assisted interactions per month2.8M
Modelled AI run-cost per month (usage + cloud, system estimate)$962K
New monthly operating cost$5.0M

All figures are modelling estimates from the labeled inputs above — nothing here is customer-provided yet. The pilot's first job is replacing these assumptions with Agoda's measured baseline. Package price covers implementation only; recurring usage billed separately.

Recommended package

Scale — $100K implementation

Scale · $100K · A multi-channel production deployment10–14 weeks to production across priority workflows

Why this package for Agoda: Multilingual amendment, refund, and property-liaison workflows across chat and voice at OTA volume are inherently a multi-workflow, multi-language deployment.

Included
  • 4–6 channels
  • 2–3 priority workflows
  • Multiple enterprise integrations
  • API credential & security setup
  • Advanced orchestration
  • Multilingual support
  • Agent Assist / human escalation
  • Production analytics
  • Expansion roadmap
Not included
  • ✕Usage & consumption (billed separately)
  • ✕Enterprise-wide governance build-out
  • ✕Multi-BU rollout
Recurring costs (separate from the package)

Packages cover implementation and integration only. Recurring costs are billed separately: Tilicho Labs platform usage (~$0.15/call-min indicative, usage only), Google Cloud consumption, telephony/carrier charges, managed operations, and support & optimization. No package includes unlimited usage.

Customer resources required
  • · API access + credentials for: Reservation platform, Property extranet, Payments
  • · A named business owner for booking amendment & cancellation resolution
  • · Security review counterpart and policy sign-off (Reservation platform scope)
  • · Baseline metrics for the pilot's success thresholds
Executive messages

What to say to whom

CEO

“Support cost per booking and CSAT trade off against each other in every budget cycle. Agentic resolution is the rare intervention that moves both in the right direction — and it A/B tests cleanly.”

CIO / CTO

“This slots into your experimentation culture: define the metric, split the traffic, and measure agentic resolution against your current stack per language pair. Keep what wins.”

COO

“Language-coverage scheduling across 24/7 shifts stops being the binding constraint — one deployment covers the long tail of language-hour combinations you can never staff efficiently.”

VP Customer Experience Group

“Amendment cases close in one conversation instead of a traveler leg plus a property leg — handle time collapses and CSAT moves at exactly the contact class where it is weakest.”

Chief Risk / Compliance Officer

“PCI-scoped payment handling stays in your systems; the agent orchestrates with tokens and logs every step — GDPR/PDPA obligations enforced per traveler jurisdiction.”

CFO / Procurement

“Model the fully loaded cost of a 2 a.m. Korean-language amendment today versus an automated resolution — then multiply by your annual amendment volume. That is the business case in one line.”

Outreach

Pre-built offer emails for Agoda

Written from this account's own research — the strategic signal, the capability gap, the entry workflow, and the modelled economics — not a mail-merge template. Pick the moment and the persona, edit anything, then copy or open in your mail client.

Moment in the deal
Cold outreach — no prior conversation
Who it's addressed to
Cares about: The workflow itself and its daily failure modes
Register
Length
Draft — edit freely before sending
Open in mail client

Customer-safe by construction: drafts are composed only from customer-facing fields. Account tier, build-vs-buy classification, priority score, internal routing, and partner-commercial detail are not inputs to the composer, so they cannot appear in a draft. Money figures are always framed as modelled from the customer's own volumes. Read before sending — you own what goes out.

The pursuit

Tier 2 — high-potential incubation

Why this tier

Bangkok-headquartered global OTA under Booking Holdings serving travelers in dozens of languages at enormous scale — customer-service cost and multilingual coverage are core P&L lines, but a deeply technical in-house culture means the bar for proof is high.

Recommended next step

Propose an A/B pilot on one language pair (e.g., Korean or Japanese amendments) with cost-per-resolved-case as the primary metric.

Entry: Booking amendment & cancellation resolution · Scale package · 10–14 weeks to production across priority workflows. Human fallback throughout; success thresholds agreed before build.

Start the pursuit

Research-based priority-account universe assembled from public information, market scale, communication volume, and solution fit. This is NOT an authoritative list of top Google Cloud customers; existing Google Cloud relationships are noted only where publicly reported. Validate every account with the account team before outreach.