About ZUS Coffee
When ZUS Coffee went looking for technology to keep pace with its rapid growth, it didn't start with the question "how do we cut headcount cost?" It started with a different one: as the store count climbs every year, how does a 30-person Customer Happiness team make sure every customer gets the right answer, from the right person, at the right time — across Malaysia, the Philippines, and Singapore?
Founded in 2019 in Malaysia with a single store, ZUS Coffee has expanded to 7 countries with more than 1,000 stores in under a decade. That growth was no accident: ZUS positions itself as a "tech-driven coffee chain" — a brand that treats technology as a core competitive advantage, not a supporting feature. From ordering systems to the loyalty app to internal operations, technology is the backbone of ZUS at every touchpoint.
The Customer Happiness team sits behind every support experience. Thirty people cover every intake channel — from menu questions and order lookups to complaint handling and responding to Google Reviews from dozens of stores. When ZUS decided to accelerate its geographic expansion, the question wasn't how many more people to hire — it was how technology could absorb the increase: how do 30 people serve an ever-expanding set of markets?
The challenge: when expansion outgrows the team
Before 2025, Customer Happiness at ZUS ran on three tools that had served for years: a chat platform, a ticketing system, and a chatbot. Each was stable within its own scope — but none talked to the others. At the end of every conversation, an agent created a ticket, picked a category, wrote notes, and manually synced across systems. The steps weren't always completed: nearly half of all tickets were missing a category — not from carelessness, but from a lack of time between back-to-back conversations.
The old chatbot in Malaysia couldn't resolve real customer requests; it only stretched out the wait through unnecessary steps before reaching a human. Instead of relieving the team, it added unplanned work. And AI, instead of supporting the team, had become an unnecessary middle step.
The Philippines — ZUS Coffee's first market — had no AI at all before Filum.ai. On a normal day, the team handled around 200 conversations; on peak days, that figure could climb toward 750. Thirty people, no automation.
The real pressure came when ZUS looked ahead: Malaysia was about to open new channels, Singapore was preparing to go live, and the chain was targeting further expansion. The old operating model wouldn't keep up.
"Finally, Filum AI is ready to launch — and to me, there's no beautiful slide or report that can replace a working product."
— Ruby, Regional Head of Customer Happiness, ZUS Coffee
The journey with Filum.ai: AI that does the real work
In April 2025, ZUS Coffee began working with Filum.ai with a specific requirement: not another chatbot, but an AI Agent that could handle conversations end to end, look up orders in real time from ZUS systems, and deploy across multiple markets with channel-specific configurations. Alongside building the handling logic, the two teams migrated more than 32,000 historical conversations onto the new platform — so the AI could learn from ZUS's real operation from day one. The Philippines went live in November 2025.
But ZUS wasn't just looking for a new tool. It was prepared to fully replace three systems that had run reliably for years — a decision few enterprises dare make at 1,000+ stores. To keep the transition from becoming a risk, the teams chose the Philippines as the starting point: a smaller market, but enough to learn how things really run, hit real problems, and solve them before stepping into Malaysia, where conversation volume is many times higher. And throughout, ZUS leadership wasn't watching from the sidelines — they were in it.
A context-aware AI Agent — resolving to the end, not just forwarding
The difference ZUS was after wasn't an AI that answers more — it was an AI that resolves what customers actually need: order status lookups, return-policy updates, store-specific information. The Filum AI Agent is integrated directly with ZUS systems via API — when a customer asks about an order, the AI queries in real time and answers immediately, without handing off. When a conversation gets complex (a specific complaint, a judgment call), the AI hands off to an agent with all the context it has gathered, so the person can pick up immediately without asking from scratch.
After the first two months in Malaysia and Singapore: more than half of incoming conversations were fully resolved by AI, with no human handoff. And nearly 87% of conversation goals were completed to the Customer Happiness team's own criteria — not just responding, but resolving.
One platform, three markets — local configs, shared operation
Each market ZUS runs has its own way for customers to reach Customer Happiness: Malaysia via WhatsApp, the ZUS App and Webchat; the Philippines via Facebook and Email; Singapore via Webchat and WhatsApp. Previously, that difference meant fragmentation — each local team on its own system, its own reports, its own market standards. With Filum.ai, each market keeps the configuration that fits its specifics — but the whole operation can be monitored and coordinated from one place.
It showed when Singapore prepared to go live: from Malaysia stabilizing to Singapore completing rollout took just 6 weeks. Not because Singapore was simpler — but because the platform was already there, already deeply integrated with ZUS systems from earlier markets. Adding a market means adding configuration, not rebuilding from scratch.
Automating the post-conversation process
Few people see the post-conversation process as a big problem — but for the Customer Happiness team of a hundreds-of-stores chain, this is where data leaks every day. Each conversation ends with a chain of manual steps: create ticket, pick category, write notes, sync. When the next conversation is already waiting, whatever step can be skipped gets skipped — and nearly half of tickets ended up with no category.
The consequence wasn't just missing data — it was untrustworthy data. Leadership had numbers, but couldn't tell whether they reflected reality or just the days agents had time to fill in every step. Now, after each conversation, the AI creates the ticket, tags the category, and fills in the summary — the agent confirms in seconds. For the first time, ZUS has a weekly report leadership can actually trust — to see which issues are emerging in which market, which channels need optimizing, which trends to address before they escalate.
Customer feedback doesn't only come through live chat. Google Reviews from hundreds of stores are gathered into one place — the Customer Happiness team responds to a single standard across the whole chain, with no store left out.
When each shift is left with only the work that needs a human
It used to be that the start of a shift meant opening a queue — order questions, opening hours, promotions stacked alongside complex complaints, with no distinction. Each one needed a person. Now, when an agent starts a shift, most routine requests are already handled: orders looked up, information sent, repeat questions answered — the AI finished them the moment the customer wrote in. What's left in the queue are the cases that need judgment: product complaints, transaction reviews, negative feedback to be handled to ZUS policy.
When taking over a conversation, the agent doesn't start from scratch. The AI has already assembled the context — chat history, issue type, related order information. Thirty people aren't stretched thinner as three markets run in parallel — because the increase from expansion is absorbed by AI before it reaches a person.
Results
- 55%+ of conversations fully resolved by AI, no human handoff — in Malaysia and Singapore over the first 2 months; the team focuses on cases that genuinely need human judgment.
- ~90% of handling turns involve AI — present in most conversations, helping agents respond faster and more consistently to SOP.
- ~87% goal achievement — completed to the team's own criteria; AI doesn't just respond, it resolves.
- 3 markets in 7 months — Philippines (Nov 2025), Malaysia (Apr 2026), Singapore (May 2026) — with the same 30-person team.
- 6 weeks from Malaysia going live to Singapore completing rollout — thanks to the platform's reusability.
Beyond those metrics, one result wasn't in the original plan: consistent classification data became the default in operations for the first time. ZUS leadership now has a weekly report reliable enough to spot trends — and to make operational decisions on consistent, complete data.
What's next
- Continuous optimization in Malaysia — ZUS aims to keep raising the AI self-resolution rate by expanding the knowledge base and refining handling logic.
- Expanding automation — deploying Filum Workflows to encode complex SOPs directly into the platform.
- Advancing AI capability — adopting Filum Agent Force to extend into more specialized scenarios for a multi-market F&B chain.
- Closing the feedback loop in the Philippines — measuring post-conversation CSAT in the first market, closing the loop from customer experience to operational improvement.
Closing
For ZUS Coffee, expanding into a new market no longer means hiring more people and retraining from scratch. The added volume — unavoidable with hundreds of new stores — is now absorbed by AI. Thirty people can focus on the cases that genuinely need human judgment and empathy, instead of being pulled into the same repeat questions day after day.
For a chain that positions itself as a "tech-driven coffee chain", that isn't just an operational result — it's the alignment of positioning and reality.
