Resilient Supply Chain Podcast: The 2% Problem in AI-Enabled Operations
Resilient Supply Chain Podcast: The 2% Problem in AI-Enabled Operations
AI systems can deliver impressive performance while leaving organisations exposed at precisely the moments when failure matters most. In episode 142 of the Resilient Supply Chain Podcast, host Tom Raftery speaks with Josh Gould, Group CEO of The Big Word, about the operational boundary between automation and human judgement. The discussion examines why AI performance should be judged by what happens in exceptional, high-consequence cases rather than average success rates alone - a distinction with direct implications for supply chain and operations leaders deploying automation at scale. The full episode is available at www.resilientsupplychainpodcast.com
When 98% Is Not Enough
The central tension is deceptively simple: automation performs best when work is repeatable, but resilience is often determined by the cases that are neither predictable nor routine. Gould describes an interpreting operation capable of responding to 98% of calls within 60 seconds. That appears to be a strong service metric until the remaining 2% is examined: these can include police and ambulance calls where delay carries considerably greater consequences.
The implication extends well beyond language services. Aggregate performance indicators can obscure the distribution of operational risk. For supply chain leaders, the critical question is therefore not simply how frequently automation succeeds, but which transactions fail and what happens next.
Designing the Human Handoff
The conversation challenges the assumption that organisations must choose between AI and people. Gould argues that AI is most valuable when routine activity is automated while difficult cases can escalate quickly to human judgement. That makes the escalation path part of the system architecture rather than an emergency workaround. “AI assists human, it doesn’t replace human,” Gould says.
The distinction becomes especially important where speed and assurance compete. In some legal settings, Gould explains, an immediate machine-generated interpretation may meet the need for speed, while a human transcription follows for formal proceedings.
For operations leaders, the parallel is clear: automation without a defined exception path can improve throughput while simultaneously creating a new operational dependency.
Productivity Without Simplistic Headcount Logic
Gould also describes a more nuanced labour impact than the common assumption that AI adoption leads directly to job elimination. His organisation has used AI to increase calls handled per employee while slowing the rate at which headcount needed to expand.
He says combining AI and human workflows reduced the cost of producing a dollar of revenue from roughly 25% to 9%, with some savings passed to customers. The strategic point is not that automation automatically cuts cost, but that workflow design determines whether technology converts into measurable operating leverage.
Scale Creates Distance
A separate tension emerges around organisational growth. Gould compares senior management in a scaled organisation with operating at 10,000 feet: leaders gain broader information and longer-term perspective, but lose proximity to where execution actually occurs. That creates a governance problem familiar across global supply chains. Better system-level visibility does not guarantee better local decisions if leadership becomes detached from operational reality.
Gould is also unusually candid about technology investment, estimating that roughly two-thirds of what he invested across technology, departments and geographic expansion ultimately delivered little value. His broader prediction is similarly restrained: the “AI revolution” may become an “AI evolution”, taking decades rather than a few years.
For supply chain and operations leaders, the lesson is less about whether AI adoption continues than about how it is governed. Resilient automation requires organisations to measure the consequences hidden behind headline performance, define clear escalation paths, and preserve human judgement where exceptions carry disproportionate risk.
