In luxury experience travel, every customer message can be the start of a high-value booking — and every minute of delay can push that customer to a competitor. Marmoris Yachting, which operates two premium brands — Marmoris Yachting and Fly Vietnam — serving multilingual international travelers across peak seasons in both Phú Quốc and Nha Trang, chose to invest seriously in an AI Agent architecture rather than scaling headcount to chase seasonal demand. Six weeks after going live with Filum.ai, three numbers tell the story most concisely: nearly 60% of conversations resolved by AI on its own, nearly 87% goal completion, and a customer care team that got its space back for the work that genuinely needs a human touch.

"When we digitalize, I want to optimize fully — to reduce the repetitive interactions that need a human, so the team can focus on the things that truly require nuance and relationships."

— Phạm Huyền, Vice Chairwoman of the Board, Marmoris Yachting (AI Agent kickoff, 11/2025)

About Marmoris Yachting

Marmoris Yachting is a luxury yachting brand in Vietnam, operating across two key coastal destinations: Phú Quốc and Nha Trang. Part of the same ecosystem is Fly Vietnam — a brand developing the sport of paragliding, running tandem flight experiences combined with mountain-sky-sea tours. The two brands serve very different audiences: Marmoris leans toward travel agencies (~70% in peak season) plus high-end individual travelers of mixed nationalities — Vietnamese, British, Russian, Chinese, Korean — while Fly Vietnam is nearly 100% individual experience-seekers.

The moment Marmoris decided to invest was also a moment of industry tailwind: according to the General Statistics Office of Vietnam (2025), Vietnam welcomed nearly 17.2 million international visitors in the first 10 months of 2025, up 21.5% year-over-year — with Phú Quốc and Khánh Hòa among the fastest-growing destinations. For a premium brand like Marmoris, this was both an opportunity to expand its international base and an operational pressure: how to respond promptly to a stream of multilingual inquiries across multiple chat channels at once, without doubling the team for every season.

The challenge: Multichannel, two brands, and peak-season pressure

Before November 2025, Marmoris's customer care operation was a multichannel but fragmented picture. The team had to switch constantly between different chat platforms — Messenger, Instagram, Zalo OA, WhatsApp Business, and even KakaoTalk for Korean customers. Each channel lived on its own tab, sometimes its own device; the care team had no unified interface to track the full customer flow of both brands at the same time.

Behind that interface was a manual tagging process with many labels per customer — status, product interest, language, and other criteria — all applied by hand so the marketing team could segment and remarket. "Right now when sales closes a customer, they also have to classify the customer — quite a lot of information to enter," said Phạm Huyền (Vice Chairwoman of the Board, Marmoris Yachting). In peak season, every unanswered message risked becoming a customer who switched to a competitor.

The pressure on response time isn't just a feeling. According to research from Lead Connect / InsideSales / MIT (2025), 78% of customers buy from the first company to respond to their inquiry — and responding within 5 minutes raises conversion odds 21x versus responding after 30 minutes. In luxury experience travel — where customers often compare several tours at once — every minute of delay carries a high probability of a lost order.

Marmoris had tried options before Filum to solve this multichannel and manual-tagging problem. A third-party solution had been used but didn't support customer classification well, forcing the team back to managing each source page directly. An in-house AI built by the technical team for Fly Vietnam's Messenger was piloted for two or three weeks, then stopped because it didn't meet real operational needs. The problem remained — there just wasn't a good enough answer yet.

"Once the AI was on, customers asking for basic information were all supported by AI, so I could flexibly go deep with select customers and only step in when the AI flagged me."

— Ngô Hoàng Thảo Nguyên, Customer Care Specialist, Marmoris Yachting & Fly Vietnam

The solution: Two brands, two AI Agents, one workspace

In November 2025, after a careful evaluation, Marmoris decided to deploy Filum.ai. The decision didn't come from a single feature — it came from three requirements Phạm Huyền (Vice Chairwoman of the Board) emphasized from the very first evaluation: operational autonomy (the Marmoris team can configure and adjust the AI Agent themselves, without sending small requests through a partner each time an advisory flow needs to change); gradual rollout (a mechanism to raise or lower the percentage of conversations the AI takes on); and one workspace that holds both brands, while each brand's AI Agent stays independent in knowledge, tone, and advisory scope.

Five channels unified into a single interface

All of Messenger, Instagram, Zalo OA, WhatsApp Business, and Facebook Comments — across both brands — were unified into a single workspace through Filum Inbox, with each conversation clearly labeled by channel and brand, so the care team could follow the entire customer flow without juggling tabs or devices. In one operational review, the Marmoris team itself proposed bringing WhatsApp Business onto the same interface — a clear sign that the unified inbox had become the team's primary way of working.

Two independent AI Agents for two brands

Marmoris and Fly Vietnam operate as distinct brands — luxury yachting differs entirely from adventure paragliding in tone, advisory scope, and product specifics. Two independent AI Agents were deployed in the same workspace, each with its own identity, prompt, and knowledge base. The Marmoris Yachting AI Agent advises on yacht tours, seating classes, and water sports; the Fly Vietnam AI Agent advises on paragliding tours and accompanying hiking experiences. Both have a handoff mechanism to the human team at the payment step or when a question falls outside automated scope.

Control by design: AI and humans trade roles when needed

In the early phase after go-live, the biggest risk for a premium care team is letting AI answer a high-value customer incorrectly. Marmoris needed a rollout that allowed the AI's share of conversations to increase gradually — starting at 10%, then 25%, then 100% — so the team could monitor answer quality at each step. Alongside it: a "Take over" / "Reassign to AI" control. When the AI veers off, the care team takes over a conversation with a single click; once resolved, the conversation is handed back to AI. That smooth role-trading between AI and human is exactly what lets Marmoris keep both response speed and experience control.

According to McKinsey (2025), AI-enabled self-service can reduce request volume by 40-50% and cut cost-to-serve by over 20% — but the prerequisite is a smooth handoff between AI and human. This is precisely the value zone Marmoris prioritized.

Automated tagging — freeing the marketing team

After a closely supervised pilot phase, the automated tagging mechanism was rolled out jointly by Marmoris and Filum. The AI Agent scans the content of each conversation and automatically applies classification labels according to logic the Marmoris team defines — customer status, product interest, language. The process that previously consumed most of the marketing team's time after each chat — tagging by hand — is now condensed; the team can filter audiences by criteria in minutes instead of hours.

Results

In the first six weeks after go-live (from 19/11/2025), across both Marmoris and Fly Vietnam brands:

• Nearly 60% of conversations resolved by AI independently — with no human intervention; a rate many larger companies take months to reach.
• Nearly 87% AI goal-completion rate once the knowledge base stabilized — for every 100 conversations, AI carries most of them fully from greeting through to capturing closing information.
• Nearly 98% full-resolution rate within the FAQ scope — general company information, booking & payment processes, policies & procedures, services & amenities, safety & incident handling.
• About 85% completion within the premium product-advisory scope — more complex because customers explore many variants, yet a strong number for an early stage.

The number that impressed the Marmoris care team most isn't on a dashboard. It's that the team can go deep with one customer on one channel while heading out into the field to photograph a paragliding customer in flight — and when they return, the AI has neatly answered the basic questions from other customers, escalating to a human only when a situation calls for nuance. In Fly Vietnam's particular operation, that's how one care specialist does the work of two without losing experience quality.

What's next

Marmoris's journey shows something many Vietnamese businesses are still wrestling with: deploying AI isn't a race of "AI replacing people" — it's about redesigning operations so people and AI work better together. The coming Phú Quốc and Nha Trang peak seasons will be the real test, and the data so far shows Marmoris has built a foundation solid enough to step into them — with Filum's 1-on-1 Customer Success Manager model continuing to stand alongside the team through every operational beat.

Do you also run multiple brands, multiple channels, and a customer care team that needs to be freed from repetitive interactions to focus on higher value?

Book a demo with Filum →

See more success stories at filum.ai/customers.