AWS debuts Agentic Value Model to fix RPA-era ROI and unlock agent value

AWS's post arrives as agentic automation appears on AI center of excellence roadmaps, but the standard ROI model—hours saved times labor cost minus build cost—was built for rule-based robotic process automation (RPA). It assumes stable processes, rule-based tasks, and that saved hours equal banked value. McKinsey's "1:3:5 pattern" suggests successful AI transformations spend three dollars on process redesign and five on capability building for every dollar invested in agentic technology, yet most companies invert this formula.

The Agentic Value Model breaks value into four pools: time savings, exception handling, decision quality, and change resilience and maintenance economics. For exception handling, AWS guidance gives planning ranges—correcting an error can cost 1.5–4 times the original transaction, and human error can account for 2–15 percent of operational cost. Decision quality may matter most for low-volume, high-value decisions like credit, pricing, or risk, where a single better decision can dwarf a year of saved minutes. Maintenance economics goes both ways: agents absorb some variation without a rewrite but shift maintenance into evaluations, prompts, monitoring, and model operations.

An illustrative claims-triage example shows why redeployment matters: automating the routine 70 percent of 200,000 claims releases about 28,000 hours, or roughly $1.26 million of capacity, before oversight. If attrition and lower overtime capture half, the case should carry about $630,000, not the full figure. Of the 8 percent needing correction (16,000 claims), a 40 percent reduction adjusted by a 75 percent realization factor yields a modeled benefit near $151,000. Amazon Quick Automate deployments at Kitsa, dLocal, and Genpact show the dimensions at work, though none proves all four pools. The broader test, per McKinsey, is that nearly two-thirds of enterprises have experimented with agents but fewer than 10 percent have scaled them to tangible value.