About KINGLED

KINGLED is a Vietnamese brand of decorative and residential LED lighting, operating a B2B2C model — distributing through a regional dealer network across the country's North, Central, and South, while engaging directly with end customers through digital channels.

The product catalog spans from recessed downlights, spotlights, and magnetic track lights to indoor and outdoor wall lights — each line carrying dozens of SKUs with four to five technical attributes that must be combined correctly. In this context, the quality of interaction on chat channels becomes a critical link in the buying journey — alongside product quality and the distribution network itself.

The challenge: a complex-product brand running on a small team

LED lighting is not a category that can be sold through simple Q&A. Each KINGLED product line has four to five technical attributes that must align: cutout size, wattage, form factor, color temperature, and housing color. A customer asking "can I get a quote for a 90mm recessed downlight" is just the starting point — the right consultation walks them through several steps to land on the correct product at the correct price. A generic chatbot without operational training risks either responding incorrectly or with too much information, causing leads to drop. The Lead Response Management Study (Dr. James Oldroyd, MIT, 2011) found that 78% of deals go to the company that responds first — meaning a single off-target reply is enough to lose a customer to a competitor before they come back.

Meanwhile, operational pressure was rising. Each day, KINGLED's Marketing team — just two people — fielded hundreds of conversations across three channels: Facebook Messenger, Zalo OA, and the website. The team worked business hours, while end customers messaged across all hours. Before AI, the average wait time for a customer reply approached three hours. According to the Vietnam Customer Experience Report 2025 (Filum.ai × SOI.Pro), Vietnamese consumers have reset their expectations of response speed across industries — they expect buying a LED light to be as responsive as ordering food online or making a digital payment. That three-hour gap was the chokepoint in KINGLED's chat-to-lead pipeline.

The solution: building strategic AI operations with Filum.ai

Filum.ai's 5-week implementation model set a clear requirement from day one: the Marketing lead at KINGLED would participate directly in the AI training process — learning how to write prompts, configure source documents, and evaluate conversations. This was designed so that AI wouldn't just run at go-live, but could be self-tuned by KINGLED after handover. That was exactly what the CEO proposed from the start: "I want us to walk this together so my team also learns the recipe — to self-tune later."

One interface for every conversation

The first step was unification: messages from Facebook Messenger, Zalo OA, and the KINGLED website were brought into a single platform where AI and human agents work side by side. The Marketing team no longer jumps between apps; messages, customer history, and notes for one customer all show up in one place. When an agent takes over a conversation, AI automatically pauses; when the agent is done, AI is ready to handle the customer's next message — no scenarios where both reply at once.

One AI, multiple skills, one set of guardrails

KINGLED's AI is not a stack of independent assistants — it's a single AI trained on multiple skills, each handling a specific cluster of tasks: product consultation that understands LED variant logic (cutout → form factor → color temperature), FAQ handling for installation and warranty, complaint intake and handoff to a human when needed.

A skill-based design lets one AI maintain consistent personality and knowledge across the full conversation, while still reacting precisely to specific situations. When a customer types "quote for a 90mm recessed downlight", the AI asks follow-ups about wattage and form factor, then proposes the right product with images — not a long catalog dump.

Structured leads, distributed to regional dealers

After consultation, the AI invites the customer to share a phone number and address, then stores the lead with structured fields plus urgency classification (high/medium/low) and intent strength, before routing to the regional dealer in charge (North/Central/South) for a personalized quote with discount — matching the B2B2C distribution rhythm. Lead handoff has moved off personal Zalo accounts: the flow is now structured, classified, and traceable.

Automatic conversation summaries for managers

Every conversation gets summarized into key takeaways. Instead of re-reading each chat to understand what customers care about, the KINGLED team can skim summaries and get a weekly view — what topics customers ask about most, what they complain about, what converts to leads. This is a class of qualitative data that simply didn't exist in the operation before.

The results: AI carries 87% of chats, wait time drops 15×

After roughly three months in production:

• AI handles 87% of all inbound messages across deployed channels. 45% of conversations are resolved end-to-end by AI, with no human intervention. KINGLED's 2-person Marketing team shifted role from manual chat duty to AI operators: monitoring performance, evaluating conversations, and refining prompts when AI misses the mark.
• Response time dropped 15× — from nearly 3 hours down to just over 10 minutes. For messages handled directly by AI, wait time is only 15–30 seconds. The Short Life of Online Sales Leads (Harvard Business Review, 2011) found that companies responding within the first hour are 7× more likely to qualify a lead than those responding later — and 60× more likely than those waiting 24 hours.
• Nearly 500 structured leads captured in the first 3 months — including name, phone number, product interest, regional location, urgency rating, and intent strength. Sales reps and regional dealers can prioritize hot leads instead of processing in arrival order.
• An unexpected upside — message consistency across channels. Before Filum, there were cases where sales reps quoted one discount, Marketing chat sent another, and the customer received conflicting signals. Now the AI only quotes list prices during consultation, handing off discount negotiation to the dealer — the process and message reaching the customer are now consistent.

"AI has hit about 85% of what we hoped for — it freed up our shift coverage, stayed on-script, and handled image analysis from customers surprisingly well."

— Nguyen Hoang Giang, Marketing Lead, KINGLED

What's next

The KINGLED story reveals a set of preconditions for AI to actually operate inside a Vietnamese business with a complex product: source documents that are structured cleanly, a regular conversation evaluation loop, and an internal champion who participates in AI training from day one. KINGLED is now expanding the AI's skill set into new task clusters — after-sales support, warranty handling, and seasonal topic analysis of customer interests.

Want results like KINGLED?

Book a demo with Filum →

Explore more success stories at filum.ai/customers.