Using hours-saved ROI will undercount agentic automation

If your team uses the classic hours-saved ROI to approve automation, you will underfund projects and face higher maintenance and exception-handling costs.

What actually changed

AWS argues that a new class of software — "agentic automation" — reasons and adapts to finish tasks, so the old RPA-era ROI model no longer captures most of its value. The traditional calculation (count transactions, measure minutes, multiply by a loaded labour rate, subtract build cost) was created for stable, high-volume, rule-driven processes. That model assumes stability, assumes tasks are purely rule-based and assumes the measured task equals the whole job; as a result it omits the costs and benefits agents introduce: maintenance when processes change, the work involved in handling exceptions, and the value of completing tasks end-to-end.

Who it affects

• AI centre of excellence (AI CoE) leaders who must build business cases for agentic automation. The source explicitly targets that audience.
• Operations and finance teams that currently use the RPA-era ROI worksheet to greenlight automation projects. If you apply the old formula to agents, your estimates will miss material items the announcement highlights.

What it costs or what it replaces

This replaces the single-line hours-saved ROI worksheet with a multi-line business case that accounts for items the RPA model ignores. The announcement does not state pricing for agentic tools or implementation. It does state that the RPA model omits:

• the cost of maintaining automations as processes change; and
• the labour and process cost of exceptions.

Because the new software can reason and adapt, the business case should also measure the value of completing tasks rather than only minutes saved, but the announcement does not provide a prescriptive pricing or cost template.

What we don't know

• The full set of metrics and line items included in AWS’s framework; the summary says a framework exists but does not detail it.
• Any ready-made templates or spreadsheets AWS provides for AI CoE use.
• How to quantify the downstream value of completing a job end-to-end (the announcement flags the gap but does not give formulas).
• Pricing, licences or implementation costs for agentic automation from AWS or partners; the announcement does not state pricing.

What to do next

1. Update your worksheet this week: reproduce your current RPA-era ROI row (transactions × minutes × loaded rate − build cost), then add two new line items — estimated annual maintenance effort for automation changes, and annual hours spent handling exceptions. Use recent change logs and ticket counts to set these numbers.

1. Rapidly shortlist candidate workflows (one afternoon): pick 5–10 processes where (a) processes change frequently, (b) exceptions are common, or (c) completing the whole task creates measurable downstream value. These criteria follow directly from the announcement’s critique of the RPA model.

1. Run a one-week pilot calculation for one shortlisted workflow: collect baseline minutes, exception count and post-completion outcome(s) you can observe this week (for example, tickets closed, approvals completed, or customer touches reduced). Recalculate ROI with the added maintenance and exception lines and compare to the classic hours-saved result. Use that comparison to decide whether to pursue a small agentic automation pilot with your AI CoE.

What to do next

(duplicate section removed)