“
At Carousell, delivering trusted support at scale is critical. Empowered by Level3AI's AgentOps team, we support millions of users across 6 markets and 5 languages, cutting resolution time by 50% while maintaining CSATs.
Gijs Verheijke
Director of Customer Experience
Overview
Carousell Group is the leading multi-category platform for secondhand in Greater Southeast Asia, on a mission to make secondhand the first choice. Founded in Singapore in 2012, the platform serves tens of millions of monthly active users across Singapore, Malaysia, Indonesia, the Philippines, Taiwan and Hong Kong, enabling people to buy and sell new and pre-loved items with the simplicity of snapping a photo and chatting.
With 70K monthly support tickets across all markets flowing through live chat and email in multiple languages, Carousell's customer support operation is always on. As the platform continued to grow, Carousell needed a customer experience AI agent that could deliver measurable cost savings without compromising service quality across its complex, multi-market operation.
The Problem
Back in 2024, the Carousell team was handling close to 15,000 weekly human tickets and relied on an existing rules-based AI stack that was limited to answering knowledge base questions, delivering only about a 10% reduction in ticket volume — far short of the transformation the business needed. Compounding this, the dominant pricing model across AI vendors, charging per resolved ticket, created a fundamental misalignment with business incentives: the better the AI performs, the more it costs.
A knowledge base built for humans, not AI. Carousell operates across multiple countries and multiple languages, with a knowledge base that has served human agents well for years. However, it was not structured for AI consumption — the tagging, naming conventions and market-level categorisation needed to be rearchitected to enable an AI agent to interpret queries accurately and resolve them autonomously at scale.
An operational challenge at scale. Moving from a previous-generation chatbot to a fully autonomous AI agent becomes an operational challenge at the scale Carousell operates: maintaining playbooks, monitoring performance and catching errors before they reach customers requires continuous engineering effort across six markets. With the rise of accessible LLM tooling, building this internally was tempting, but the scale of ongoing operational work made partnering with a specialist the smarter bet.
The Solution
Carousell chose Level3AI for its per-conversation pricing model, its guaranteed outcome approach and its forward deployment model, enabling a confident shift toward AI-powered customer experience at scale across all six markets.
Level3AI deployed Emily, a customer experience AI agent, across Carousell's live chat and email channels, serving as the first point of contact for all customer interactions across every market and language. Emily autonomously resolves queries spanning transaction disputes, payment issues, account security and listing policies, adhering to SOPs and policies. When escalation is required, she hands off to human agents with full context. This fundamentally changed the cost structure, allowing Carousell to absorb growing customer demand without scaling its BPO workforce.
None of this falls on Carousell to build or run alone. The Level3AI team trains and supports Carousell's AgentOps staff to manage Emily's performance efficiently, and continuously launches products and tools that make their work easier, faster and smarter.
Features such as Playbook Quality Inspector, Ticket Observability and MCP give Carousell's team control over different aspects of their operation. Playbook Quality Inspector catches mistakes and inconsistencies as playbooks are written, so new workflows launch faster and with fewer costly errors reaching production. Ticket Observability lets the team investigate any individual case themselves at ticket level, seeing how Emily reasoned through customers' responses and backend API data, which means issues get resolved and contained faster. And MCP gives Carousell's operations team direct access to measure performance, investigate conversations, look up SOPs and knowledge, check configuration health and test a change in UAT.
The Result
Carousell's AI-first transformation delivered results that redefined the economics of its customer support operation. The deployment achieved a 60% reduction in agent costs and a 50% improvement in time-to-resolve. Human ticket volume dropped by more than 50%, resulting in hundreds of thousands of dollars in annual cost savings.
This is also where the ongoing operational work pays off: continuous monitoring and product refinement through Level3AI's AgentOps team and tooling have pushed customer satisfaction to a record high since Emily went live — proof that the quality bar doesn't slip as the AI takes on more of the workload.
Emily now powers the entire global customer service operation across six countries and five languages from a single platform. Carousell's support operation has shifted from a reactive, headcount-driven model to a strategic, AI-augmented function where the team focuses on system improvement, process design and continuous AI optimisation rather than repetitive ticket handling.
Key metrics
•
-60%
in agent costs
•
-50%
in resolution time
Industry
eCommerce
Product
Chat & Email
