Kogan's deployment of an AI agent built with Salesforce's Agentforce represents a deliberate rejection of perfectionism in favour of iterative production deployment. The Australian e-tailer moved from a deterministic bot resolving just six percent of inbound inquiries to a generative AI agent handling 10 distinct use cases within six weeks, with WISMO (where is my order?) accounting for roughly 60 percent of customer requests. The implementation spans order status retrieval, returns processing, warranty handling, product recommendations across 16 million SKUs, membership queries, voucher redemption, charge investigations, troubleshooting, identity verification, and damage assessment via image recognition. Critically, Kogan's CTO Goran Stefkovski articulated a philosophy that directly challenges how many organisations approach AI pilots: the insistence on waiting for unrealistic resolution rates from day one is precisely what prevents deployment. This framing matters for teams evaluating whether to build incrementally or hold out for higher confidence thresholds before going live.
The technical architecture reveals how forward-deployed vendor engineers are becoming embedded in customer implementations to accelerate time-to-value. Kogan leveraged four Agentforce services—NextGen Authoring, Agent Script, Observability, and Testing Center—suggesting that platform maturity around agent orchestration, monitoring, and validation is now table stakes for enterprise CX tooling. The damage detection use case demonstrates a secondary benefit: generative AI's ability to identify fraudulent submissions (duplicate images across cases) without requiring classical machine learning infrastructure. For teams already running Agentforce or comparable platforms, this raises a practical question about whether your current observability and testing frameworks are equipped to surface these emergent fraud-detection capabilities, or whether you're still treating AI agents primarily as efficiency plays rather than risk-mitigation tools.
The phased rollout strategy—deliberately not applying the agent to all product types initially—indicates that Kogan is treating this as a controlled expansion rather than a big-bang replacement of human-led support. This approach contradicts the narrative that AI agents should immediately absorb all inbound volume. For support leaders managing team transitions, the implication is clear: selective deployment by use case complexity and risk profile allows you to maintain service quality whilst building internal confidence in agent performance. The six-week path from prototype to production, enabled by vendor partnership, also suggests that the bottleneck for most organisations isn't technical capability but organisational willingness to accept early-stage resolution rates and iterate in production.
Kogan taps AI agent to handle customer service iTnews