How we helped
Working with Artium by AlixPartners, a global online marketplace, moved from a logjam of agentic ambitions to live multi-agent customer service in just four months, with a phased rollout that proved value before any volume commitment.
The Challenge
Customer service is among the first functions most companies target for AI, and among the hardest to automate well. Conventional automation depends on repeatable patterns: a fixed product catalog, standard order types, and predictable questions. A two-sided marketplace has none of these. Every listing is unique, every transaction pairs two individuals, and no two service interactions look quite the same, which leaves rules-based systems with nothing stable to match against.
For this marketplace, the opportunity was clear but had been unmanageable in scale. A single division alone had identified 88 potential use cases for agentic AI. The workflows in question spanned account security, returns, and research. None of them could be completed without routing and handoffs between multiple people. The company had a long list of possibilities, no way to tell which would work, and no appetite for committing resources to an approach that remained unproven.
Our Approach
We started with three agentic use cases out of the 88 and took each one all the way into production, rather than running broad pilots that risked achieving little. The three agents sat in account security, returns, and research, each handling a workflow a customer service representative would otherwise have picked up by hand.
The harder problem was coordination, so we built a multi-agent orchestration layer that routes work between agents and manages the handoffs that previously required people, allowing tasks to move through security checks, returns processing, and research without dropping between systems.
Rollout was deliberately staged. A fourth agent, added to facilitate internal review with a human in the loop throughout, went live on 1% of traffic. Once performance held, it moved to 25% within a month, and continued upward toward full coverage. Each increment gave the business evidence before it gave the business exposure, which is what allowed the program to scale without a leap of faith at the start.
Key results
More than $400 million in cost-to-serve savings identified
Agents were built and deployed in four months, leading to major savings projections on track for realization
Approximately 20% improvement in customer retention on the new platform compared with its predecessor
Significant improvements on the new platform raised customer retention
An orchestration layer that can be used beyond the first three use cases
Each additional agent now plugs into coordination logic that already exists, which changes the economics of every subsequent build and turns a list of 88 possibilities into a roadmap the business can actually fund.
The client gained much more than four agents. The company now has a whole new set of capabilities: a proven method for taking agentic use cases from idea to production, infrastructure that makes the next one cheaper than the last, and the evidence to justify scaling at a pace the business is comfortable with.
