Optimize process

Compare a recurring process with a proposed version of it, and estimate the difference honestly. A person and an agent are both possible performers of each step, including the agent running this analysis.

Map one cycle

Trace one real cycle from trigger to finished result, before and after the change: each step, who or what performs it, and the waiting between steps. When no change is proposed yet, propose one: remove the causes of waiting and rework first, and keep every step that carries a required decision, review, or control. Ask only for the facts that would change the estimate, and mark the rest as assumptions.

Estimate the impact

• Keep effort and elapsed time apart: an agent may finish a step in minutes while a human spends an hour reviewing it, and a queue may still make the cycle take days.
• Count agent work at its likely runtime plus the human setup and review it needs. Never convert it into human-only days or weeks.
• Keep one-time setup apart from recurring effort, and time saved per cycle apart from shorter end-to-end lead time.
• Use observed timings when they exist. Otherwise give a range in minutes or hours only when its assumptions are stated, and a direction of change when numbers would be guesswork.
• Do not invent error rates, cost savings, satisfaction scores, or percentage improvements.
• Name the assumption with the greatest effect on the estimate, and what to observe in a small trial to check it.

Output